Module astrapy.data.cursors.farr_cursor
Expand source code
# Copyright DataStax, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import annotations
from copy import deepcopy
from inspect import iscoroutinefunction
from typing import Any, Awaitable, Callable, Generic, cast
from astrapy import AsyncCollection, Collection
from astrapy.constants import (
FilterType,
HybridSortType,
ProjectionType,
)
from astrapy.data.cursors.cursor import (
TNEW,
TRAW,
AbstractCursor,
CursorState,
T,
_ensure_vector,
_revise_timeouts_for_cursor_copy,
)
from astrapy.data.cursors.query_engine import _CollectionFindAndRerankQueryEngine
from astrapy.data.cursors.reranked_result import RerankedResult
from astrapy.data_types import DataAPIVector
from astrapy.exceptions import (
CursorException,
MultiCallTimeoutManager,
)
from astrapy.utils.unset import _UNSET, UnsetType
class CollectionFindAndRerankCursor(
Generic[TRAW, T], AbstractCursor[RerankedResult[TRAW]]
):
"""
A synchronous cursor over documents, as returned by a `find_and_rerank` invocation
on a Collection. A cursor can be iterated over, materialized into a list,
and queried/manipulated in various ways.
Some cursor operations mutate it in-place (such as consuming its documents),
other return a new cursor without changing the original one. See the documentation
for the various methods and the Collection `find_and_rerank` method for more details
and usage patterns.
This cursor has two type parameters: TRAW and T. The first is the type
of the "raw" documents as they are found on the collection, the second
is the type of the items after the optional mapping function (see the `.map()`
method).
If no mapping is specified, `T = RerankedResult[TRAW]`: the items yielded by
the cursor are a `RerankedResult` wrapping the type (possibly after projection)
of the documents found on the collection: in other words, such a cursor returns
the documents, as they come back from the API, with their associated scores
from the find-and-rerank operation.
In general, consuming a cursor returns items of type T, except for the
`consume_buffer` primitive that draws directly from the buffer and always
returns items of type RerankedResult[TRAW].
Example:
>>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.)
>>> cursor = collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=5,
... include_scores=True,
... )
>>> for r_result in cursor:
... print(f"{r_result.document['wkd']}: {r_result.scores['$rerank']}")
...
Wed: -9.1015625
Mon: -10.2421875
Tue: -10.2421875
Sun: -11.375
Fri: -12.515625
"""
_query_engine: _CollectionFindAndRerankQueryEngine[TRAW]
_request_timeout_ms: int | None
_overall_timeout_ms: int | None
_request_timeout_label: str | None
_overall_timeout_label: str | None
_timeout_manager: MultiCallTimeoutManager
_filter: FilterType | None
_projection: ProjectionType | None
_sort: HybridSortType | None
_limit: int | None
_hybrid_limits: int | dict[str, int] | None
_include_scores: bool | None
_include_sort_vector: bool | None
_rerank_on: str | None
_rerank_query: str | None
_mapper: Callable[[RerankedResult[TRAW]], T] | None
def __init__(
self,
*,
collection: Collection[TRAW],
request_timeout_ms: int | None,
overall_timeout_ms: int | None,
request_timeout_label: str | None = None,
overall_timeout_label: str | None = None,
filter: FilterType | None = None,
projection: ProjectionType | None = None,
sort: HybridSortType | None = None,
limit: int | None = None,
hybrid_limits: int | dict[str, int] | None = None,
include_scores: bool | None = None,
include_sort_vector: bool | None = None,
rerank_on: str | None = None,
rerank_query: str | None = None,
mapper: Callable[[RerankedResult[TRAW]], T] | None = None,
) -> None:
self._filter = deepcopy(filter)
self._projection = projection
self._sort = deepcopy(sort)
self._limit = limit
self._hybrid_limits = deepcopy(hybrid_limits)
self._include_scores = include_scores
self._include_sort_vector = include_sort_vector
self._rerank_on = rerank_on
self._rerank_query = rerank_query
self._mapper = mapper
self._request_timeout_ms = request_timeout_ms
self._overall_timeout_ms = overall_timeout_ms
self._request_timeout_label = request_timeout_label
self._overall_timeout_label = overall_timeout_label
self._query_engine = _CollectionFindAndRerankQueryEngine(
collection=collection,
async_collection=None,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
)
AbstractCursor.__init__(self)
self._timeout_manager = MultiCallTimeoutManager(
overall_timeout_ms=self._overall_timeout_ms,
timeout_label=self._overall_timeout_label,
)
def _copy(
self: CollectionFindAndRerankCursor[TRAW, T],
*,
request_timeout_ms: int | None | UnsetType = _UNSET,
overall_timeout_ms: int | None | UnsetType = _UNSET,
request_timeout_label: str | None | UnsetType = _UNSET,
overall_timeout_label: str | None | UnsetType = _UNSET,
filter: FilterType | None | UnsetType = _UNSET,
projection: ProjectionType | None | UnsetType = _UNSET,
sort: dict[str, Any] | None | UnsetType = _UNSET,
limit: int | None | UnsetType = _UNSET,
hybrid_limits: int | dict[str, int] | None | UnsetType = _UNSET,
include_scores: bool | None | UnsetType = _UNSET,
include_sort_vector: bool | None | UnsetType = _UNSET,
rerank_on: str | None | UnsetType = _UNSET,
rerank_query: str | None | UnsetType = _UNSET,
) -> CollectionFindAndRerankCursor[TRAW, T]:
if self._query_engine.collection is None:
raise RuntimeError("Query engine has no collection.")
return CollectionFindAndRerankCursor(
collection=self._query_engine.collection,
request_timeout_ms=self._request_timeout_ms
if isinstance(request_timeout_ms, UnsetType)
else request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms
if isinstance(overall_timeout_ms, UnsetType)
else overall_timeout_ms,
request_timeout_label=self._request_timeout_label
if isinstance(request_timeout_label, UnsetType)
else request_timeout_label,
overall_timeout_label=self._overall_timeout_label
if isinstance(overall_timeout_label, UnsetType)
else overall_timeout_label,
filter=self._filter if isinstance(filter, UnsetType) else filter,
projection=self._projection
if isinstance(projection, UnsetType)
else projection,
sort=self._sort if isinstance(sort, UnsetType) else sort,
limit=self._limit if isinstance(limit, UnsetType) else limit,
hybrid_limits=self._hybrid_limits
if isinstance(hybrid_limits, UnsetType)
else hybrid_limits,
include_scores=self._include_scores
if isinstance(include_scores, UnsetType)
else include_scores,
include_sort_vector=self._include_sort_vector
if isinstance(include_sort_vector, UnsetType)
else include_sort_vector,
rerank_on=self._rerank_on
if isinstance(rerank_on, UnsetType)
else rerank_on,
rerank_query=self._rerank_query
if isinstance(rerank_query, UnsetType)
else rerank_query,
mapper=self._mapper,
)
def _try_ensure_fill_buffer(self) -> None:
"""
If buffer is empty, try to fill with next page, if applicable.
If not possible, silently do nothing.
This method never changes the cursor state.
"""
if self._state == CursorState.CLOSED:
return
if not self._buffer:
if self._next_page_state is not None or self._state == CursorState.IDLE:
new_buffer, next_page_state, resp_status = (
self._query_engine._fetch_page(
page_state=self._next_page_state,
timeout_context=self._timeout_manager.remaining_timeout(
cap_time_ms=self._request_timeout_ms,
cap_timeout_label=self._request_timeout_label,
),
)
)
self._next_page_state = next_page_state
self._last_response_status = resp_status
self._pages_retrieved += 1
self._buffer = new_buffer
def __repr__(self) -> str:
return (
f'{self.__class__.__name__}("{self.data_source.name}", '
f"{self._state.value}, "
f"consumed so far: {self.consumed})"
)
def __iter__(
self: CollectionFindAndRerankCursor[TRAW, T],
) -> CollectionFindAndRerankCursor[TRAW, T]:
self._ensure_alive()
return self
def __next__(self) -> T:
if self.state == CursorState.CLOSED:
raise StopIteration
self._try_ensure_fill_buffer()
if not self._buffer:
self._state = CursorState.CLOSED
raise StopIteration
self._state = CursorState.STARTED
# consume one item from buffer
traw0, rest_buffer = self._buffer[0], self._buffer[1:]
self._buffer = rest_buffer
self._consumed += 1
return cast(T, self._mapper(traw0) if self._mapper is not None else traw0)
@property
def data_source(self) -> Collection[TRAW]:
"""
The Collection object that originated this cursor through a `find_and_rerank`
operation.
Returns:
a Collection instance.
"""
if self._query_engine.collection is None:
raise RuntimeError("Query engine has no collection.")
return self._query_engine.collection
def clone(self) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Create a copy of this cursor with:
- the same parameters (timeouts, filter, projection, etc)
- and the cursor is rewound to its pristine IDLE state.
Returns:
a new CollectionFindAndRerankCursor, similar to this one but without mapping
and rewound to its initial state.
Example:
>>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.)
>>> cursor = collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=3,
... ).map(lambda r_result: r_result.document["wkd"].upper())
>>> for idx, value in zip([0, 1], cursor):
... print(f"{idx} ==> {value}")
...
0 ==> MON
1 ==> TUE
>>> cloned_cursor = cursor.clone()
>>> for value in cloned_cursor:
... print(f"(cloned) {value}")
...
(cloned) MON
(cloned) TUE
(cloned) SUN
>>>
>>> print(f"n ==> {next(cursor)}")
n ==> SUN
"""
if self._query_engine.collection is None:
raise RuntimeError("Query engine has no collection.")
return CollectionFindAndRerankCursor(
collection=self._query_engine.collection,
request_timeout_ms=self._request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms,
request_timeout_label=self._request_timeout_label,
overall_timeout_label=self._overall_timeout_label,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
mapper=self._mapper,
)
def filter(
self, filter: FilterType | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new filter setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
filter: a new filter setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `filter` which is the provided value.
"""
self._ensure_idle()
return self._copy(filter=filter)
def project(
self, projection: ProjectionType | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new projection setting.
This operation is allowed only if the cursor state is still IDLE and if
no mapping has been set on it.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
projection: a new projection setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `projection` which is the provided value.
"""
self._ensure_idle()
if self._mapper is not None:
raise CursorException(
"Cannot set projection after map.",
cursor_state=self._state.value,
)
return self._copy(projection=projection)
def sort(
self, sort: HybridSortType | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new sort setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
sort: a new sort setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `sort` which is the provided value.
"""
self._ensure_idle()
return self._copy(sort=sort)
def limit(self, limit: int | None) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new limit setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
limit: a new limit setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `limit` which is the provided value.
"""
self._ensure_idle()
return self._copy(limit=limit)
def hybrid_limits(
self, hybrid_limits: int | dict[str, int] | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new hybrid_limits setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
hybrid_limits: a new setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `hybrid_limits` which is the provided value.
"""
self._ensure_idle()
return self._copy(hybrid_limits=hybrid_limits)
def include_scores(
self, include_scores: bool | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new include_scores setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
include_scores: a new include_scores setting to apply
to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `include_scores` which is the provided value.
"""
self._ensure_idle()
return self._copy(include_scores=include_scores)
def include_sort_vector(
self, include_sort_vector: bool | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new include_sort_vector setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
include_sort_vector: a new include_sort_vector setting to apply
to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `include_sort_vector` which is the provided value.
"""
self._ensure_idle()
return self._copy(include_sort_vector=include_sort_vector)
def rerank_on(
self, rerank_on: str | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new rerank_on setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
rerank_on: a new setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `rerank_on` which is the provided value.
"""
self._ensure_idle()
return self._copy(rerank_on=rerank_on)
def rerank_query(
self, rerank_query: str | None
) -> CollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new rerank_query setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the Collection `find_and_rerank` method.
Args:
rerank_query: a new setting to apply to the returned new cursor.
Returns:
a new CollectionFindAndRerankCursor with the same settings as this one,
except for `rerank_query` which is the provided value.
"""
self._ensure_idle()
return self._copy(rerank_query=rerank_query)
def map(
self, mapper: Callable[[T], TNEW]
) -> CollectionFindAndRerankCursor[TRAW, TNEW]:
"""
Return a copy of this cursor with a mapping function to transform
the returned items. Calling this method on a cursor with a mapping
already set results in the mapping functions being composed.
This operation is allowed only if the cursor state is still IDLE.
Args:
mapper: a function transforming the objects returned by the cursor
into something else (i.e. a function T => TNEW).
If the map is imposed on a cursor without mapping yet, its input
argument must be a `RerankedResult[TRAW]`, where TRAW
stands for the type of the documents from the collection.
Returns:
a new CollectionFindAndRerankCursor with a new mapping function on the results,
possibly composed with any pre-existing mapping function.
Example:
>>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.)
>>> cursor = collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=3,
... )
>>> for r_result in cursor:
... print(r_result.document)
...
{'_id': 'A', 'wkd': 'Mon'}
{'_id': 'B', 'wkd': 'Tue'}
{'_id': 'G', 'wkd': 'Sun'}
>>> cursor_mapped = cursor.clone().map(
... lambda r_result: r_result.document["wkd"]
... )
>>> for value in cursor_mapped:
... print(value)
...
Mon
Tue
Sun
>>> cursor_mapped_twice = cursor_mapped.clone().map(
... lambda wkd: f"<{wkd[:2].lower()}>"
... )
>>> for value in cursor_mapped_twice:
... print(value)
...
<mo>
<tu>
<su>
"""
self._ensure_idle()
if self._query_engine.collection is None:
raise RuntimeError("Query engine has no collection.")
composite_mapper: Callable[[RerankedResult[TRAW]], TNEW]
if self._mapper is not None:
def _composite(document: RerankedResult[TRAW]) -> TNEW:
return mapper(self._mapper(document)) # type: ignore[misc]
composite_mapper = _composite
else:
composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper)
return CollectionFindAndRerankCursor(
collection=self._query_engine.collection,
request_timeout_ms=self._request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms,
request_timeout_label=self._request_timeout_label,
overall_timeout_label=self._overall_timeout_label,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
mapper=composite_mapper,
)
def for_each(
self,
function: Callable[[T], bool | None],
*,
general_method_timeout_ms: int | None = None,
timeout_ms: int | None = None,
) -> None:
"""
Consume the remaining documents in the cursor, invoking a provided callback
function on each of them.
Calling this method on a CLOSED cursor results in an error.
The callback function can return any value. The return value is generally
discarded, with the following exception: if the function returns the boolean
`False`, it is taken to signify that the method should quit early, leaving the
cursor half-consumed (ACTIVE state). If this does not occur, this method
results in the cursor entering CLOSED state once it is exhausted.
Args:
function: a callback function whose only parameter is of the type returned
by the cursor. This callback is invoked once per each document yielded
by the cursor. If the callback returns a `False`, the `for_each`
invocation stops early and returns without consuming further documents.
general_method_timeout_ms: a timeout, in milliseconds, for the whole
duration of this method. If not provided, there is no such timeout.
Note that the per-request timeout set on the cursor still applies.
timeout_ms: an alias for `general_method_timeout_ms`.
Example:
>>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.)
>>> from astrapy.cursors import CursorState, RerankedResult
>>>
>>> cursor = collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=3,
... )
>>> def printer(r_result: RerankedResult):
... print(f"-> {r_result.document['wkd']}")
...
>>> cursor.for_each(printer)
-> Mon
-> Tue
-> Sun
>>>
>>> if cursor.state != CursorState.CLOSED:
... print(f"alive: {cursor.to_list()}")
... else:
... print("(closed)")
...
(closed)
>>> cursor2 = cursor.clone()
>>> def checker(r_result: RerankedResult):
... print(f"-> {r_result.document['wkd']}")
... return r_result.document["wkd"] != "Tue"
...
>>> cursor2.for_each(checker)
-> Mon
-> Tue
>>>
>>> if cursor2.state != CursorState.CLOSED:
... print(f"alive: {list(cursor2)}")
... else:
... print("(closed)")
...
alive: [RerankedResult(document={'_id': 'G', 'wkd': 'Sun'}, scores={})]
"""
self._ensure_alive()
copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy(
new_general_method_timeout_ms=general_method_timeout_ms,
new_timeout_ms=timeout_ms,
old_request_timeout_ms=self._request_timeout_ms,
)
_cursor = self._copy(
request_timeout_ms=copy_req_ms,
overall_timeout_ms=copy_ovr_ms,
)
self._imprint_internal_state(_cursor)
for document in _cursor:
res = function(document)
if res is False:
break
_cursor._imprint_internal_state(self)
def to_list(
self,
*,
general_method_timeout_ms: int | None = None,
timeout_ms: int | None = None,
) -> list[T]:
"""
Materialize all documents that remain to be consumed from a cursor into a list.
Calling this method on a CLOSED cursor results in an error.
If the cursor is IDLE, the result will be the whole set of documents returned
by the `find_and_rerank` operation; otherwise, the documents already consumed
by the cursor will not be in the resulting list.
Calling this method is not recommended if a huge list of results is anticipated:
it would involve a large number of data exchanges with the Data API and possibly
a massive memory usage to construct the list. In such cases, a lazy pattern
of iterating and consuming the documents is to be preferred.
Args:
general_method_timeout_ms: a timeout, in milliseconds, for the whole
duration of this method. If not provided, there is no such timeout.
Note that the per-request timeout set on the cursor still applies.
timeout_ms: an alias for `general_method_timeout_ms`.
Returns:
a list of documents (or other values depending on the mapping
function, if one is set). These are all items that were left
to be consumed on the cursor when `to_list` is called.
Example:
>>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.)
>>> collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=4,
... ).map(
... lambda r_result: r_result.document["wkd"]
... ).to_list()
['Wed', 'Mon', 'Tue', 'Sun']
>>>
>>> cursor = collection.find_and_rerank(
... sort={"$hybrid": "Weekdays?"},
... projection={"wkd": True},
... limit=4,
... ).map(lambda r_result: r_result.document["wkd"])
>>> print(f"First item: {cursor.__next__()}.")
First item: Wed.
>>> cursor.to_list()
['Mon', 'Tue', 'Sun']
"""
self._ensure_alive()
copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy(
new_general_method_timeout_ms=general_method_timeout_ms,
new_timeout_ms=timeout_ms,
old_request_timeout_ms=self._request_timeout_ms,
)
_cursor = self._copy(
request_timeout_ms=copy_req_ms,
overall_timeout_ms=copy_ovr_ms,
)
self._imprint_internal_state(_cursor)
documents = [document for document in _cursor]
_cursor._imprint_internal_state(self)
return documents
def has_next(self) -> bool:
"""
Whether the cursor actually has more documents to return.
`has_next` can be called on any cursor, but on a CLOSED cursor
will always return False.
This method can trigger the fetch operation of a new page, if the current
buffer is empty.
Calling `has_next` on an IDLE cursor triggers the first page fetch, but the
cursor stays in the IDLE state until actual consumption starts.
Returns:
a boolean value of True if there is at least one further item
available to consume; False otherwise (including the case of CLOSED
cursor).
"""
if self._state == CursorState.CLOSED:
return False
self._try_ensure_fill_buffer()
return len(self._buffer) > 0
def get_sort_vector(self) -> list[float] | DataAPIVector | None:
"""
Return the query vector used in the vector (ANN) search that was run as
part of the search expressed by this cursor, if applicable.
Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch,
but the cursor stays in the IDLE state until actual consumption starts.
The method can be invoked on a CLOSED cursor and will return either None
or the sort vector used in the search.
Returns:
the query vector used in the search, if it was requested by passing
`include_sort_vector=True` to the `find_and_rerank` call that originated
the cursor.
If the sort vector is not available, None is returned.
Otherwise, the vector is returned as either a DataAPIVector
or a plain list of number depending on the setting for
`APIOptions.serdes_options`.
"""
self._try_ensure_fill_buffer()
if self._last_response_status:
return _ensure_vector(
self._last_response_status.get("sortVector"),
self.data_source.api_options.serdes_options,
)
else:
return None
class AsyncCollectionFindAndRerankCursor(
Generic[TRAW, T], AbstractCursor[RerankedResult[TRAW]]
):
"""
An asynchronous cursor over documents, as returned by a `find_and_rerank` invocation
on an AsyncCollection. A cursor can be iterated over, materialized into a list,
and queried/manipulated in various ways.
Some cursor operations mutate it in-place (such as consuming its documents),
other return a new cursor without changing the original one. See the documentation
for the various methods and the AsyncCollection `find_and_rerank` method for more
details and usage patterns.
This cursor has two type parameters: TRAW and T. The first is the type
of the "raw" documents as they are found on the collection, the second
is the type of the items after the optional mapping function (see the `.map()`
method).
If no mapping is specified, `T = RerankedResult[TRAW]`: the items yielded by
the cursor are a `RerankedResult` wrapping the type (possibly after projection)
of the documents found on the collection: in other words, such a cursor returns
the documents, as they come back from the API, with their associated scores
from the find-and-rerank operation.
In general, consuming a cursor returns items of type T, except for the
`consume_buffer` primitive that draws directly from the buffer and always
returns items of type RerankedResult[TRAW].
This class is the async counterpart of the CollectionFindAndRerankCursor, for use
with asyncio. Other than the async interface, its behavior is identical: please
refer to the documentation for `CollectionFindAndRerankCursor` for examples
and details.
"""
_query_engine: _CollectionFindAndRerankQueryEngine[TRAW]
_request_timeout_ms: int | None
_overall_timeout_ms: int | None
_request_timeout_label: str | None
_overall_timeout_label: str | None
_timeout_manager: MultiCallTimeoutManager
_filter: FilterType | None
_projection: ProjectionType | None
_sort: HybridSortType | None
_limit: int | None
_hybrid_limits: int | dict[str, int] | None
_include_scores: bool | None
_include_sort_vector: bool | None
_rerank_on: str | None
_rerank_query: str | None
_mapper: Callable[[RerankedResult[TRAW]], T] | None
def __init__(
self,
*,
collection: AsyncCollection[TRAW],
request_timeout_ms: int | None,
overall_timeout_ms: int | None,
request_timeout_label: str | None = None,
overall_timeout_label: str | None = None,
filter: FilterType | None = None,
projection: ProjectionType | None = None,
sort: HybridSortType | None = None,
limit: int | None = None,
hybrid_limits: int | dict[str, int] | None = None,
include_scores: bool | None = None,
include_sort_vector: bool | None = None,
rerank_on: str | None = None,
rerank_query: str | None = None,
mapper: Callable[[RerankedResult[TRAW]], T] | None = None,
) -> None:
self._filter = deepcopy(filter)
self._projection = projection
self._sort = deepcopy(sort)
self._limit = limit
self._hybrid_limits = deepcopy(hybrid_limits)
self._include_scores = include_scores
self._include_sort_vector = include_sort_vector
self._rerank_on = rerank_on
self._rerank_query = rerank_query
self._mapper = mapper
self._request_timeout_ms = request_timeout_ms
self._overall_timeout_ms = overall_timeout_ms
self._request_timeout_label = request_timeout_label
self._overall_timeout_label = overall_timeout_label
self._query_engine = _CollectionFindAndRerankQueryEngine(
collection=None,
async_collection=collection,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
)
AbstractCursor.__init__(self)
self._timeout_manager = MultiCallTimeoutManager(
overall_timeout_ms=self._overall_timeout_ms,
timeout_label=self._overall_timeout_label,
)
def _copy(
self: AsyncCollectionFindAndRerankCursor[TRAW, T],
*,
request_timeout_ms: int | None | UnsetType = _UNSET,
overall_timeout_ms: int | None | UnsetType = _UNSET,
request_timeout_label: str | None | UnsetType = _UNSET,
overall_timeout_label: str | None | UnsetType = _UNSET,
filter: FilterType | None | UnsetType = _UNSET,
projection: ProjectionType | None | UnsetType = _UNSET,
sort: dict[str, Any] | None | UnsetType = _UNSET,
limit: int | None | UnsetType = _UNSET,
hybrid_limits: int | dict[str, int] | None | UnsetType = _UNSET,
include_scores: bool | None | UnsetType = _UNSET,
include_sort_vector: bool | None | UnsetType = _UNSET,
rerank_on: str | None | UnsetType = _UNSET,
rerank_query: str | None | UnsetType = _UNSET,
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
if self._query_engine.async_collection is None:
raise RuntimeError("Query engine has no async collection.")
return AsyncCollectionFindAndRerankCursor(
collection=self._query_engine.async_collection,
request_timeout_ms=self._request_timeout_ms
if isinstance(request_timeout_ms, UnsetType)
else request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms
if isinstance(overall_timeout_ms, UnsetType)
else overall_timeout_ms,
request_timeout_label=self._request_timeout_label
if isinstance(request_timeout_label, UnsetType)
else request_timeout_label,
overall_timeout_label=self._overall_timeout_label
if isinstance(overall_timeout_label, UnsetType)
else overall_timeout_label,
filter=self._filter if isinstance(filter, UnsetType) else filter,
projection=self._projection
if isinstance(projection, UnsetType)
else projection,
sort=self._sort if isinstance(sort, UnsetType) else sort,
limit=self._limit if isinstance(limit, UnsetType) else limit,
hybrid_limits=self._hybrid_limits
if isinstance(hybrid_limits, UnsetType)
else hybrid_limits,
include_scores=self._include_scores
if isinstance(include_scores, UnsetType)
else include_scores,
include_sort_vector=self._include_sort_vector
if isinstance(include_sort_vector, UnsetType)
else include_sort_vector,
rerank_on=self._rerank_on
if isinstance(rerank_on, UnsetType)
else rerank_on,
rerank_query=self._rerank_query
if isinstance(rerank_query, UnsetType)
else rerank_query,
mapper=self._mapper,
)
async def _try_ensure_fill_buffer(self) -> None:
"""
If buffer is empty, try to fill with next page, if applicable.
If not possible, silently do nothing.
This method never changes the cursor state.
"""
if self._state == CursorState.CLOSED:
return
if not self._buffer:
if self._next_page_state is not None or self._state == CursorState.IDLE:
(
new_buffer,
next_page_state,
resp_status,
) = await self._query_engine._async_fetch_page(
page_state=self._next_page_state,
timeout_context=self._timeout_manager.remaining_timeout(
cap_time_ms=self._request_timeout_ms,
cap_timeout_label=self._request_timeout_label,
),
)
self._next_page_state = next_page_state
self._last_response_status = resp_status
self._pages_retrieved += 1
self._buffer = new_buffer
def __repr__(self) -> str:
return (
f'{self.__class__.__name__}("{self.data_source.name}", '
f"{self._state.value}, "
f"consumed so far: {self.consumed})"
)
def __aiter__(
self: AsyncCollectionFindAndRerankCursor[TRAW, T],
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
self._ensure_alive()
return self
async def __anext__(self) -> T:
if self.state == CursorState.CLOSED:
raise StopAsyncIteration
await self._try_ensure_fill_buffer()
if not self._buffer:
self._state = CursorState.CLOSED
raise StopAsyncIteration
self._state = CursorState.STARTED
# consume one item from buffer
traw0, rest_buffer = self._buffer[0], self._buffer[1:]
self._buffer = rest_buffer
self._consumed += 1
return cast(T, self._mapper(traw0) if self._mapper is not None else traw0)
@property
def data_source(self) -> AsyncCollection[TRAW]:
"""
The AsyncCollection object that originated this cursor through a
`find_and_rerank` operation.
Returns:
an AsyncCollection instance.
"""
if self._query_engine.async_collection is None:
raise RuntimeError("Query engine has no async collection.")
return self._query_engine.async_collection
def clone(self) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Create a copy of this cursor with:
- the same parameters (timeouts, filter, projection, etc)
- and the cursor is rewound to its pristine IDLE state.
For usage examples, please refer to the same method of the
equivalent synchronous CollectionFindCursor class, and apply the necessary
adaptations to the async interface.
Returns:
a new AsyncCollectionFindAndRerankCursor, similar to this one but
rewound to its initial state.
"""
if self._query_engine.async_collection is None:
raise RuntimeError("Query engine has no async collection.")
return AsyncCollectionFindAndRerankCursor(
collection=self._query_engine.async_collection,
request_timeout_ms=self._request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms,
request_timeout_label=self._request_timeout_label,
overall_timeout_label=self._overall_timeout_label,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
mapper=self._mapper,
)
def filter(
self, filter: FilterType | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new filter setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
filter: a new filter setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `filter` which is the provided value.
"""
self._ensure_idle()
return self._copy(filter=filter)
def project(
self, projection: ProjectionType | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new projection setting.
This operation is allowed only if the cursor state is still IDLE and if
no mapping has been set on it.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
projection: a new projection setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `projection` which is the provided value.
"""
self._ensure_idle()
if self._mapper is not None:
raise CursorException(
"Cannot set projection after map.",
cursor_state=self._state.value,
)
return self._copy(projection=projection)
def sort(
self, sort: HybridSortType | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new sort setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
sort: a new sort setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `sort` which is the provided value.
"""
self._ensure_idle()
return self._copy(sort=sort)
def limit(self, limit: int | None) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new limit setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
limit: a new limit setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `limit` which is the provided value.
"""
self._ensure_idle()
return self._copy(limit=limit)
def hybrid_limits(
self, hybrid_limits: int | dict[str, int] | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new hybrid_limits setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
hybrid_limits: a new setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `hybrid_limits` which is the provided value.
"""
self._ensure_idle()
return self._copy(hybrid_limits=hybrid_limits)
def include_scores(
self, include_scores: bool | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new include_scores setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
include_scores: a new include_scores setting to apply
to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `include_scores` which is the provided value.
"""
self._ensure_idle()
return self._copy(include_scores=include_scores)
def include_sort_vector(
self, include_sort_vector: bool | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new include_sort_vector setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
include_sort_vector: a new include_sort_vector setting to apply
to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `include_sort_vector` which is the provided value.
"""
self._ensure_idle()
return self._copy(include_sort_vector=include_sort_vector)
def rerank_on(
self, rerank_on: str | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new rerank_on setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
rerank_on: a new setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `rerank_on` which is the provided value.
"""
self._ensure_idle()
return self._copy(rerank_on=rerank_on)
def rerank_query(
self, rerank_query: str | None
) -> AsyncCollectionFindAndRerankCursor[TRAW, T]:
"""
Return a copy of this cursor with a new rerank_query setting.
This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists
in passing arguments to the AsyncCollection `find_and_rerank` method.
Args:
rerank_query: a new setting to apply to the returned new cursor.
Returns:
a new AsyncCollectionFindAndRerankCursor with the same settings as this one,
except for `rerank_query` which is the provided value.
"""
self._ensure_idle()
return self._copy(rerank_query=rerank_query)
def map(
self, mapper: Callable[[T], TNEW]
) -> AsyncCollectionFindAndRerankCursor[TRAW, TNEW]:
"""
Return a copy of this cursor with a mapping function to transform
the returned items. Calling this method on a cursor with a mapping
already set results in the mapping functions being composed.
This operation is allowed only if the cursor state is still IDLE.
For usage examples, please refer to the same method of the
equivalent synchronous CollectionFindCursor class, and apply the necessary
adaptations to the async interface.
Args:
mapper: a function transforming the objects returned by the cursor
into something else (i.e. a function T => TNEW).
If the map is imposed on a cursor without mapping yet, its input
argument must be a `RerankedResult[TRAW]`, where TRAW
stands for the type of the documents from the collection.
Returns:
a new AsyncCollectionFindAndRerankCursor with a new mapping function on the
results, possibly composed with any pre-existing mapping function.
"""
self._ensure_idle()
if self._query_engine.async_collection is None:
raise RuntimeError("Query engine has no async collection.")
composite_mapper: Callable[[RerankedResult[TRAW]], TNEW]
if self._mapper is not None:
def _composite(document: RerankedResult[TRAW]) -> TNEW:
return mapper(self._mapper(document)) # type: ignore[misc]
composite_mapper = _composite
else:
composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper)
return AsyncCollectionFindAndRerankCursor(
collection=self._query_engine.async_collection,
request_timeout_ms=self._request_timeout_ms,
overall_timeout_ms=self._overall_timeout_ms,
request_timeout_label=self._request_timeout_label,
overall_timeout_label=self._overall_timeout_label,
filter=self._filter,
projection=self._projection,
sort=self._sort,
limit=self._limit,
hybrid_limits=self._hybrid_limits,
include_scores=self._include_scores,
include_sort_vector=self._include_sort_vector,
rerank_on=self._rerank_on,
rerank_query=self._rerank_query,
mapper=composite_mapper,
)
async def for_each(
self,
function: Callable[[T], bool | None] | Callable[[T], Awaitable[bool | None]],
*,
general_method_timeout_ms: int | None = None,
timeout_ms: int | None = None,
) -> None:
"""
Consume the remaining documents in the cursor, invoking a provided callback
function -- or coroutine -- on each of them.
Calling this method on a CLOSED cursor results in an error.
The callback function can return any value. The return value is generally
discarded, with the following exception: if the function returns the boolean
`False`, it is taken to signify that the method should quit early, leaving the
cursor half-consumed (ACTIVE state). If this does not occur, this method
results in the cursor entering CLOSED state once it is exhausted.
For usage examples, please refer to the same method of the
equivalent synchronous CollectionFindCursor class, and apply the necessary
adaptations to the async interface.
Args:
function: a callback function, or a coroutine, whose only parameter is of
the type returned by the cursor.
This callback is invoked once per each document yielded
by the cursor. If the callback returns a `False`, the `for_each`
invocation stops early and returns without consuming further documents.
general_method_timeout_ms: a timeout, in milliseconds, for the whole
duration of this method. If not provided, there is no such timeout.
Note that the per-request timeout set on the cursor still applies.
timeout_ms: an alias for `general_method_timeout_ms`.
"""
self._ensure_alive()
copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy(
new_general_method_timeout_ms=general_method_timeout_ms,
new_timeout_ms=timeout_ms,
old_request_timeout_ms=self._request_timeout_ms,
)
_cursor = self._copy(
request_timeout_ms=copy_req_ms,
overall_timeout_ms=copy_ovr_ms,
)
self._imprint_internal_state(_cursor)
is_coro = iscoroutinefunction(function)
async for document in _cursor:
if is_coro:
res = await function(document) # type: ignore[misc]
else:
res = function(document)
if res is False:
break
_cursor._imprint_internal_state(self)
async def to_list(
self,
*,
general_method_timeout_ms: int | None = None,
timeout_ms: int | None = None,
) -> list[T]:
"""
Materialize all documents that remain to be consumed from a cursor into a list.
Calling this method on a CLOSED cursor results in an error.
If the cursor is IDLE, the result will be the whole set of documents returned
by the `find_and_rerank` operation; otherwise, the documents already consumed
by the cursor will not be in the resulting list.
Calling this method is not recommended if a huge list of results is anticipated:
it would involve a large number of data exchanges with the Data API and possibly
a massive memory usage to construct the list. In such cases, a lazy pattern
of iterating and consuming the documents is to be preferred.
For usage examples, please refer to the same method of the
equivalent synchronous CollectionFindCursor class, and apply the necessary
adaptations to the async interface.
Args:
general_method_timeout_ms: a timeout, in milliseconds, for the whole
duration of this method. If not provided, there is no such timeout.
Note that the per-request timeout set on the cursor still applies.
timeout_ms: an alias for `general_method_timeout_ms`.
Returns:
a list of documents (or other values depending on the mapping
function, if one is set). These are all items that were left
to be consumed on the cursor when `to_list` is called.
"""
self._ensure_alive()
copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy(
new_general_method_timeout_ms=general_method_timeout_ms,
new_timeout_ms=timeout_ms,
old_request_timeout_ms=self._request_timeout_ms,
)
_cursor = self._copy(
request_timeout_ms=copy_req_ms,
overall_timeout_ms=copy_ovr_ms,
)
self._imprint_internal_state(_cursor)
documents = [document async for document in _cursor]
_cursor._imprint_internal_state(self)
return documents
async def has_next(self) -> bool:
"""
Whether the cursor actually has more documents to return.
`has_next` can be called on any cursor, but on a CLOSED cursor
will always return False.
This method can trigger the fetch operation of a new page, if the current
buffer is empty.
Calling `has_next` on an IDLE cursor triggers the first page fetch, but the
cursor stays in the IDLE state until actual consumption starts.
Returns:
a boolean value of True if there is at least one further item
available to consume; False otherwise (including the case of CLOSED
cursor).
"""
if self._state == CursorState.CLOSED:
return False
await self._try_ensure_fill_buffer()
return len(self._buffer) > 0
async def get_sort_vector(self) -> list[float] | DataAPIVector | None:
"""
Return the query vector used in the vector (ANN) search that was run as
part of the search expressed by this cursor, if applicable.
Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch,
but the cursor stays in the IDLE state until actual consumption starts.
The method can be invoked on a CLOSED cursor and will return either None
or the sort vector used in the search.
Returns:
the query vector used in the search, if it was requested by passing
`include_sort_vector=True` to the `find_and_rerank` call that originated
the cursor.
If the sort vector is not available, None is returned.
Otherwise, the vector is returned as either a DataAPIVector
or a plain list of number depending on the setting for
`APIOptions.serdes_options`.
"""
await self._try_ensure_fill_buffer()
if self._last_response_status:
return _ensure_vector(
self._last_response_status.get("sortVector"),
self.data_source.api_options.serdes_options,
)
else:
return NoneClasses
class AsyncCollectionFindAndRerankCursor (*, collection: AsyncCollection[TRAW], request_timeout_ms: int | None, overall_timeout_ms: int | None, request_timeout_label: str | None = None, overall_timeout_label: str | None = None, filter: FilterType | None = None, projection: ProjectionType | None = None, sort: HybridSortType | None = None, limit: int | None = None, hybrid_limits: int | dict[str, int] | None = None, include_scores: bool | None = None, include_sort_vector: bool | None = None, rerank_on: str | None = None, rerank_query: str | None = None, mapper: Callable[[RerankedResult[TRAW]], T] | None = None)-
An asynchronous cursor over documents, as returned by a
find_and_rerankinvocation on an AsyncCollection. A cursor can be iterated over, materialized into a list, and queried/manipulated in various ways.Some cursor operations mutate it in-place (such as consuming its documents), other return a new cursor without changing the original one. See the documentation for the various methods and the AsyncCollection
find_and_rerankmethod for more details and usage patterns.This cursor has two type parameters: TRAW and T. The first is the type of the "raw" documents as they are found on the collection, the second is the type of the items after the optional mapping function (see the
.map()method). If no mapping is specified,T = RerankedResult[TRAW]: the items yielded by the cursor are aRerankedResultwrapping the type (possibly after projection) of the documents found on the collection: in other words, such a cursor returns the documents, as they come back from the API, with their associated scores from the find-and-rerank operation. In general, consuming a cursor returns items of type T, except for theconsume_bufferprimitive that draws directly from the buffer and always returns items of type RerankedResult[TRAW].This class is the async counterpart of the CollectionFindAndRerankCursor, for use with asyncio. Other than the async interface, its behavior is identical: please refer to the documentation for
CollectionFindAndRerankCursorfor examples and details.Expand source code
class AsyncCollectionFindAndRerankCursor( Generic[TRAW, T], AbstractCursor[RerankedResult[TRAW]] ): """ An asynchronous cursor over documents, as returned by a `find_and_rerank` invocation on an AsyncCollection. A cursor can be iterated over, materialized into a list, and queried/manipulated in various ways. Some cursor operations mutate it in-place (such as consuming its documents), other return a new cursor without changing the original one. See the documentation for the various methods and the AsyncCollection `find_and_rerank` method for more details and usage patterns. This cursor has two type parameters: TRAW and T. The first is the type of the "raw" documents as they are found on the collection, the second is the type of the items after the optional mapping function (see the `.map()` method). If no mapping is specified, `T = RerankedResult[TRAW]`: the items yielded by the cursor are a `RerankedResult` wrapping the type (possibly after projection) of the documents found on the collection: in other words, such a cursor returns the documents, as they come back from the API, with their associated scores from the find-and-rerank operation. In general, consuming a cursor returns items of type T, except for the `consume_buffer` primitive that draws directly from the buffer and always returns items of type RerankedResult[TRAW]. This class is the async counterpart of the CollectionFindAndRerankCursor, for use with asyncio. Other than the async interface, its behavior is identical: please refer to the documentation for `CollectionFindAndRerankCursor` for examples and details. """ _query_engine: _CollectionFindAndRerankQueryEngine[TRAW] _request_timeout_ms: int | None _overall_timeout_ms: int | None _request_timeout_label: str | None _overall_timeout_label: str | None _timeout_manager: MultiCallTimeoutManager _filter: FilterType | None _projection: ProjectionType | None _sort: HybridSortType | None _limit: int | None _hybrid_limits: int | dict[str, int] | None _include_scores: bool | None _include_sort_vector: bool | None _rerank_on: str | None _rerank_query: str | None _mapper: Callable[[RerankedResult[TRAW]], T] | None def __init__( self, *, collection: AsyncCollection[TRAW], request_timeout_ms: int | None, overall_timeout_ms: int | None, request_timeout_label: str | None = None, overall_timeout_label: str | None = None, filter: FilterType | None = None, projection: ProjectionType | None = None, sort: HybridSortType | None = None, limit: int | None = None, hybrid_limits: int | dict[str, int] | None = None, include_scores: bool | None = None, include_sort_vector: bool | None = None, rerank_on: str | None = None, rerank_query: str | None = None, mapper: Callable[[RerankedResult[TRAW]], T] | None = None, ) -> None: self._filter = deepcopy(filter) self._projection = projection self._sort = deepcopy(sort) self._limit = limit self._hybrid_limits = deepcopy(hybrid_limits) self._include_scores = include_scores self._include_sort_vector = include_sort_vector self._rerank_on = rerank_on self._rerank_query = rerank_query self._mapper = mapper self._request_timeout_ms = request_timeout_ms self._overall_timeout_ms = overall_timeout_ms self._request_timeout_label = request_timeout_label self._overall_timeout_label = overall_timeout_label self._query_engine = _CollectionFindAndRerankQueryEngine( collection=None, async_collection=collection, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, ) AbstractCursor.__init__(self) self._timeout_manager = MultiCallTimeoutManager( overall_timeout_ms=self._overall_timeout_ms, timeout_label=self._overall_timeout_label, ) def _copy( self: AsyncCollectionFindAndRerankCursor[TRAW, T], *, request_timeout_ms: int | None | UnsetType = _UNSET, overall_timeout_ms: int | None | UnsetType = _UNSET, request_timeout_label: str | None | UnsetType = _UNSET, overall_timeout_label: str | None | UnsetType = _UNSET, filter: FilterType | None | UnsetType = _UNSET, projection: ProjectionType | None | UnsetType = _UNSET, sort: dict[str, Any] | None | UnsetType = _UNSET, limit: int | None | UnsetType = _UNSET, hybrid_limits: int | dict[str, int] | None | UnsetType = _UNSET, include_scores: bool | None | UnsetType = _UNSET, include_sort_vector: bool | None | UnsetType = _UNSET, rerank_on: str | None | UnsetType = _UNSET, rerank_query: str | None | UnsetType = _UNSET, ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") return AsyncCollectionFindAndRerankCursor( collection=self._query_engine.async_collection, request_timeout_ms=self._request_timeout_ms if isinstance(request_timeout_ms, UnsetType) else request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms if isinstance(overall_timeout_ms, UnsetType) else overall_timeout_ms, request_timeout_label=self._request_timeout_label if isinstance(request_timeout_label, UnsetType) else request_timeout_label, overall_timeout_label=self._overall_timeout_label if isinstance(overall_timeout_label, UnsetType) else overall_timeout_label, filter=self._filter if isinstance(filter, UnsetType) else filter, projection=self._projection if isinstance(projection, UnsetType) else projection, sort=self._sort if isinstance(sort, UnsetType) else sort, limit=self._limit if isinstance(limit, UnsetType) else limit, hybrid_limits=self._hybrid_limits if isinstance(hybrid_limits, UnsetType) else hybrid_limits, include_scores=self._include_scores if isinstance(include_scores, UnsetType) else include_scores, include_sort_vector=self._include_sort_vector if isinstance(include_sort_vector, UnsetType) else include_sort_vector, rerank_on=self._rerank_on if isinstance(rerank_on, UnsetType) else rerank_on, rerank_query=self._rerank_query if isinstance(rerank_query, UnsetType) else rerank_query, mapper=self._mapper, ) async def _try_ensure_fill_buffer(self) -> None: """ If buffer is empty, try to fill with next page, if applicable. If not possible, silently do nothing. This method never changes the cursor state. """ if self._state == CursorState.CLOSED: return if not self._buffer: if self._next_page_state is not None or self._state == CursorState.IDLE: ( new_buffer, next_page_state, resp_status, ) = await self._query_engine._async_fetch_page( page_state=self._next_page_state, timeout_context=self._timeout_manager.remaining_timeout( cap_time_ms=self._request_timeout_ms, cap_timeout_label=self._request_timeout_label, ), ) self._next_page_state = next_page_state self._last_response_status = resp_status self._pages_retrieved += 1 self._buffer = new_buffer def __repr__(self) -> str: return ( f'{self.__class__.__name__}("{self.data_source.name}", ' f"{self._state.value}, " f"consumed so far: {self.consumed})" ) def __aiter__( self: AsyncCollectionFindAndRerankCursor[TRAW, T], ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: self._ensure_alive() return self async def __anext__(self) -> T: if self.state == CursorState.CLOSED: raise StopAsyncIteration await self._try_ensure_fill_buffer() if not self._buffer: self._state = CursorState.CLOSED raise StopAsyncIteration self._state = CursorState.STARTED # consume one item from buffer traw0, rest_buffer = self._buffer[0], self._buffer[1:] self._buffer = rest_buffer self._consumed += 1 return cast(T, self._mapper(traw0) if self._mapper is not None else traw0) @property def data_source(self) -> AsyncCollection[TRAW]: """ The AsyncCollection object that originated this cursor through a `find_and_rerank` operation. Returns: an AsyncCollection instance. """ if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") return self._query_engine.async_collection def clone(self) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Returns: a new AsyncCollectionFindAndRerankCursor, similar to this one but rewound to its initial state. """ if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") return AsyncCollectionFindAndRerankCursor( collection=self._query_engine.async_collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=self._mapper, ) def filter( self, filter: FilterType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: filter: a new filter setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `filter` which is the provided value. """ self._ensure_idle() return self._copy(filter=filter) def project( self, projection: ProjectionType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: projection: a new projection setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `projection` which is the provided value. """ self._ensure_idle() if self._mapper is not None: raise CursorException( "Cannot set projection after map.", cursor_state=self._state.value, ) return self._copy(projection=projection) def sort( self, sort: HybridSortType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: sort: a new sort setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `sort` which is the provided value. """ self._ensure_idle() return self._copy(sort=sort) def limit(self, limit: int | None) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: limit: a new limit setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `limit` which is the provided value. """ self._ensure_idle() return self._copy(limit=limit) def hybrid_limits( self, hybrid_limits: int | dict[str, int] | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: hybrid_limits: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `hybrid_limits` which is the provided value. """ self._ensure_idle() return self._copy(hybrid_limits=hybrid_limits) def include_scores( self, include_scores: bool | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: include_scores: a new include_scores setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `include_scores` which is the provided value. """ self._ensure_idle() return self._copy(include_scores=include_scores) def include_sort_vector( self, include_sort_vector: bool | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: include_sort_vector: a new include_sort_vector setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `include_sort_vector` which is the provided value. """ self._ensure_idle() return self._copy(include_sort_vector=include_sort_vector) def rerank_on( self, rerank_on: str | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: rerank_on: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `rerank_on` which is the provided value. """ self._ensure_idle() return self._copy(rerank_on=rerank_on) def rerank_query( self, rerank_query: str | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: rerank_query: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `rerank_query` which is the provided value. """ self._ensure_idle() return self._copy(rerank_query=rerank_query) def map( self, mapper: Callable[[T], TNEW] ) -> AsyncCollectionFindAndRerankCursor[TRAW, TNEW]: """ Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed. This operation is allowed only if the cursor state is still IDLE. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: mapper: a function transforming the objects returned by the cursor into something else (i.e. a function T => TNEW). If the map is imposed on a cursor without mapping yet, its input argument must be a `RerankedResult[TRAW]`, where TRAW stands for the type of the documents from the collection. Returns: a new AsyncCollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function. """ self._ensure_idle() if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") composite_mapper: Callable[[RerankedResult[TRAW]], TNEW] if self._mapper is not None: def _composite(document: RerankedResult[TRAW]) -> TNEW: return mapper(self._mapper(document)) # type: ignore[misc] composite_mapper = _composite else: composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper) return AsyncCollectionFindAndRerankCursor( collection=self._query_engine.async_collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=composite_mapper, ) async def for_each( self, function: Callable[[T], bool | None] | Callable[[T], Awaitable[bool | None]], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> None: """ Consume the remaining documents in the cursor, invoking a provided callback function -- or coroutine -- on each of them. Calling this method on a CLOSED cursor results in an error. The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean `False`, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: function: a callback function, or a coroutine, whose only parameter is of the type returned by the cursor. This callback is invoked once per each document yielded by the cursor. If the callback returns a `False`, the `for_each` invocation stops early and returns without consuming further documents. general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) is_coro = iscoroutinefunction(function) async for document in _cursor: if is_coro: res = await function(document) # type: ignore[misc] else: res = function(document) if res is False: break _cursor._imprint_internal_state(self) async def to_list( self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> list[T]: """ Materialize all documents that remain to be consumed from a cursor into a list. Calling this method on a CLOSED cursor results in an error. If the cursor is IDLE, the result will be the whole set of documents returned by the `find_and_rerank` operation; otherwise, the documents already consumed by the cursor will not be in the resulting list. Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Returns: a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when `to_list` is called. """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) documents = [document async for document in _cursor] _cursor._imprint_internal_state(self) return documents async def has_next(self) -> bool: """ Whether the cursor actually has more documents to return. `has_next` can be called on any cursor, but on a CLOSED cursor will always return False. This method can trigger the fetch operation of a new page, if the current buffer is empty. Calling `has_next` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. Returns: a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor). """ if self._state == CursorState.CLOSED: return False await self._try_ensure_fill_buffer() return len(self._buffer) > 0 async def get_sort_vector(self) -> list[float] | DataAPIVector | None: """ Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable. Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search. Returns: the query vector used in the search, if it was requested by passing `include_sort_vector=True` to the `find_and_rerank` call that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting for `APIOptions.serdes_options`. """ await self._try_ensure_fill_buffer() if self._last_response_status: return _ensure_vector( self._last_response_status.get("sortVector"), self.data_source.api_options.serdes_options, ) else: return NoneAncestors
- AbstractCursor
- abc.ABC
- typing.Generic
Instance variables
var data_source : AsyncCollection[~TRAW]-
The AsyncCollection object that originated this cursor through a
find_and_rerankoperation.Returns
an AsyncCollection instance.
Expand source code
@property def data_source(self) -> AsyncCollection[TRAW]: """ The AsyncCollection object that originated this cursor through a `find_and_rerank` operation. Returns: an AsyncCollection instance. """ if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") return self._query_engine.async_collection
Methods
def clone(self) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state.
For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface.
Returns
a new AsyncCollectionFindAndRerankCursor, similar to this one but rewound to its initial state.
Expand source code
def clone(self) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Returns: a new AsyncCollectionFindAndRerankCursor, similar to this one but rewound to its initial state. """ if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") return AsyncCollectionFindAndRerankCursor( collection=self._query_engine.async_collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=self._mapper, ) def filter(self, filter: FilterType | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
filter- a new filter setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
filterwhich is the provided value.Expand source code
def filter( self, filter: FilterType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: filter: a new filter setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `filter` which is the provided value. """ self._ensure_idle() return self._copy(filter=filter) async def for_each(self, function: Callable[[T], bool | None] | Callable[[T], Awaitable[bool | None]], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None) ‑> None-
Consume the remaining documents in the cursor, invoking a provided callback function – or coroutine – on each of them.
Calling this method on a CLOSED cursor results in an error.
The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean
False, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted.For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface.
Args
function- a callback function, or a coroutine, whose only parameter is of
the type returned by the cursor.
This callback is invoked once per each document yielded
by the cursor. If the callback returns a
False, thefor_eachinvocation stops early and returns without consuming further documents. general_method_timeout_ms- a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies.
timeout_ms- an alias for
general_method_timeout_ms.
Expand source code
async def for_each( self, function: Callable[[T], bool | None] | Callable[[T], Awaitable[bool | None]], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> None: """ Consume the remaining documents in the cursor, invoking a provided callback function -- or coroutine -- on each of them. Calling this method on a CLOSED cursor results in an error. The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean `False`, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: function: a callback function, or a coroutine, whose only parameter is of the type returned by the cursor. This callback is invoked once per each document yielded by the cursor. If the callback returns a `False`, the `for_each` invocation stops early and returns without consuming further documents. general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) is_coro = iscoroutinefunction(function) async for document in _cursor: if is_coro: res = await function(document) # type: ignore[misc] else: res = function(document) if res is False: break _cursor._imprint_internal_state(self) async def get_sort_vector(self) ‑> list[float] | DataAPIVector | None-
Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable.
Calling
get_sort_vectoron an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts.The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search.
Returns
the query vector used in the search, if it was requested by passing
include_sort_vector=Trueto thefind_and_rerankcall that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting forAPIOptions.serdes_options.Expand source code
async def get_sort_vector(self) -> list[float] | DataAPIVector | None: """ Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable. Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search. Returns: the query vector used in the search, if it was requested by passing `include_sort_vector=True` to the `find_and_rerank` call that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting for `APIOptions.serdes_options`. """ await self._try_ensure_fill_buffer() if self._last_response_status: return _ensure_vector( self._last_response_status.get("sortVector"), self.data_source.api_options.serdes_options, ) else: return None async def has_next(self) ‑> bool-
Whether the cursor actually has more documents to return.
has_nextcan be called on any cursor, but on a CLOSED cursor will always return False.This method can trigger the fetch operation of a new page, if the current buffer is empty.
Calling
has_nexton an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts.Returns
a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor).
Expand source code
async def has_next(self) -> bool: """ Whether the cursor actually has more documents to return. `has_next` can be called on any cursor, but on a CLOSED cursor will always return False. This method can trigger the fetch operation of a new page, if the current buffer is empty. Calling `has_next` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. Returns: a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor). """ if self._state == CursorState.CLOSED: return False await self._try_ensure_fill_buffer() return len(self._buffer) > 0 def hybrid_limits(self, hybrid_limits: int | dict[str, int] | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
hybrid_limits- a new setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
hybrid_limitswhich is the provided value.Expand source code
def hybrid_limits( self, hybrid_limits: int | dict[str, int] | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: hybrid_limits: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `hybrid_limits` which is the provided value. """ self._ensure_idle() return self._copy(hybrid_limits=hybrid_limits) def include_scores(self, include_scores: bool | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
include_scores- a new include_scores setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
include_scoreswhich is the provided value.Expand source code
def include_scores( self, include_scores: bool | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: include_scores: a new include_scores setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `include_scores` which is the provided value. """ self._ensure_idle() return self._copy(include_scores=include_scores) def include_sort_vector(self, include_sort_vector: bool | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
include_sort_vector- a new include_sort_vector setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
include_sort_vectorwhich is the provided value.Expand source code
def include_sort_vector( self, include_sort_vector: bool | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: include_sort_vector: a new include_sort_vector setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `include_sort_vector` which is the provided value. """ self._ensure_idle() return self._copy(include_sort_vector=include_sort_vector) def limit(self, limit: int | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
limit- a new limit setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
limitwhich is the provided value.Expand source code
def limit(self, limit: int | None) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: limit: a new limit setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `limit` which is the provided value. """ self._ensure_idle() return self._copy(limit=limit) def map(self, mapper: Callable[[T], TNEW]) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~TNEW]-
Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed.
This operation is allowed only if the cursor state is still IDLE.
For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface.
Args
mapper- a function transforming the objects returned by the cursor
into something else (i.e. a function T => TNEW).
If the map is imposed on a cursor without mapping yet, its input
argument must be a
RerankedResult[TRAW], where TRAW stands for the type of the documents from the collection.
Returns
a new AsyncCollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function.
Expand source code
def map( self, mapper: Callable[[T], TNEW] ) -> AsyncCollectionFindAndRerankCursor[TRAW, TNEW]: """ Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed. This operation is allowed only if the cursor state is still IDLE. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: mapper: a function transforming the objects returned by the cursor into something else (i.e. a function T => TNEW). If the map is imposed on a cursor without mapping yet, its input argument must be a `RerankedResult[TRAW]`, where TRAW stands for the type of the documents from the collection. Returns: a new AsyncCollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function. """ self._ensure_idle() if self._query_engine.async_collection is None: raise RuntimeError("Query engine has no async collection.") composite_mapper: Callable[[RerankedResult[TRAW]], TNEW] if self._mapper is not None: def _composite(document: RerankedResult[TRAW]) -> TNEW: return mapper(self._mapper(document)) # type: ignore[misc] composite_mapper = _composite else: composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper) return AsyncCollectionFindAndRerankCursor( collection=self._query_engine.async_collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=composite_mapper, ) def project(self, projection: ProjectionType | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
projection- a new projection setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
projectionwhich is the provided value.Expand source code
def project( self, projection: ProjectionType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: projection: a new projection setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `projection` which is the provided value. """ self._ensure_idle() if self._mapper is not None: raise CursorException( "Cannot set projection after map.", cursor_state=self._state.value, ) return self._copy(projection=projection) def rerank_on(self, rerank_on: str | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
rerank_on- a new setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
rerank_onwhich is the provided value.Expand source code
def rerank_on( self, rerank_on: str | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: rerank_on: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `rerank_on` which is the provided value. """ self._ensure_idle() return self._copy(rerank_on=rerank_on) def rerank_query(self, rerank_query: str | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
rerank_query- a new setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
rerank_querywhich is the provided value.Expand source code
def rerank_query( self, rerank_query: str | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: rerank_query: a new setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `rerank_query` which is the provided value. """ self._ensure_idle() return self._copy(rerank_query=rerank_query) def sort(self, sort: HybridSortType | None) ‑> AsyncCollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection
find_and_rerankmethod.Args
sort- a new sort setting to apply to the returned new cursor.
Returns
a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for
sortwhich is the provided value.Expand source code
def sort( self, sort: HybridSortType | None ) -> AsyncCollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the AsyncCollection `find_and_rerank` method. Args: sort: a new sort setting to apply to the returned new cursor. Returns: a new AsyncCollectionFindAndRerankCursor with the same settings as this one, except for `sort` which is the provided value. """ self._ensure_idle() return self._copy(sort=sort) async def to_list(self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None) ‑> list[~T]-
Materialize all documents that remain to be consumed from a cursor into a list.
Calling this method on a CLOSED cursor results in an error.
If the cursor is IDLE, the result will be the whole set of documents returned by the
find_and_rerankoperation; otherwise, the documents already consumed by the cursor will not be in the resulting list.Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred.
For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface.
Args
general_method_timeout_ms- a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies.
timeout_ms- an alias for
general_method_timeout_ms.
Returns
a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when
to_listis called.Expand source code
async def to_list( self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> list[T]: """ Materialize all documents that remain to be consumed from a cursor into a list. Calling this method on a CLOSED cursor results in an error. If the cursor is IDLE, the result will be the whole set of documents returned by the `find_and_rerank` operation; otherwise, the documents already consumed by the cursor will not be in the resulting list. Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred. For usage examples, please refer to the same method of the equivalent synchronous CollectionFindCursor class, and apply the necessary adaptations to the async interface. Args: general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Returns: a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when `to_list` is called. """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) documents = [document async for document in _cursor] _cursor._imprint_internal_state(self) return documents
Inherited members
class CollectionFindAndRerankCursor (*, collection: Collection[TRAW], request_timeout_ms: int | None, overall_timeout_ms: int | None, request_timeout_label: str | None = None, overall_timeout_label: str | None = None, filter: FilterType | None = None, projection: ProjectionType | None = None, sort: HybridSortType | None = None, limit: int | None = None, hybrid_limits: int | dict[str, int] | None = None, include_scores: bool | None = None, include_sort_vector: bool | None = None, rerank_on: str | None = None, rerank_query: str | None = None, mapper: Callable[[RerankedResult[TRAW]], T] | None = None)-
A synchronous cursor over documents, as returned by a
find_and_rerankinvocation on a Collection. A cursor can be iterated over, materialized into a list, and queried/manipulated in various ways.Some cursor operations mutate it in-place (such as consuming its documents), other return a new cursor without changing the original one. See the documentation for the various methods and the Collection
find_and_rerankmethod for more details and usage patterns.This cursor has two type parameters: TRAW and T. The first is the type of the "raw" documents as they are found on the collection, the second is the type of the items after the optional mapping function (see the
.map()method). If no mapping is specified,T = RerankedResult[TRAW]: the items yielded by the cursor are aRerankedResultwrapping the type (possibly after projection) of the documents found on the collection: in other words, such a cursor returns the documents, as they come back from the API, with their associated scores from the find-and-rerank operation. In general, consuming a cursor returns items of type T, except for theconsume_bufferprimitive that draws directly from the buffer and always returns items of type RerankedResult[TRAW].Example
>>> # (this assumes 'vectorize'. See <code>Collection.find\_and\_rerank</code> for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=5, ... include_scores=True, ... ) >>> for r_result in cursor: ... print(f"{r_result.document['wkd']}: {r_result.scores['$rerank']}") ... Wed: -9.1015625 Mon: -10.2421875 Tue: -10.2421875 Sun: -11.375 Fri: -12.515625Expand source code
class CollectionFindAndRerankCursor( Generic[TRAW, T], AbstractCursor[RerankedResult[TRAW]] ): """ A synchronous cursor over documents, as returned by a `find_and_rerank` invocation on a Collection. A cursor can be iterated over, materialized into a list, and queried/manipulated in various ways. Some cursor operations mutate it in-place (such as consuming its documents), other return a new cursor without changing the original one. See the documentation for the various methods and the Collection `find_and_rerank` method for more details and usage patterns. This cursor has two type parameters: TRAW and T. The first is the type of the "raw" documents as they are found on the collection, the second is the type of the items after the optional mapping function (see the `.map()` method). If no mapping is specified, `T = RerankedResult[TRAW]`: the items yielded by the cursor are a `RerankedResult` wrapping the type (possibly after projection) of the documents found on the collection: in other words, such a cursor returns the documents, as they come back from the API, with their associated scores from the find-and-rerank operation. In general, consuming a cursor returns items of type T, except for the `consume_buffer` primitive that draws directly from the buffer and always returns items of type RerankedResult[TRAW]. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=5, ... include_scores=True, ... ) >>> for r_result in cursor: ... print(f"{r_result.document['wkd']}: {r_result.scores['$rerank']}") ... Wed: -9.1015625 Mon: -10.2421875 Tue: -10.2421875 Sun: -11.375 Fri: -12.515625 """ _query_engine: _CollectionFindAndRerankQueryEngine[TRAW] _request_timeout_ms: int | None _overall_timeout_ms: int | None _request_timeout_label: str | None _overall_timeout_label: str | None _timeout_manager: MultiCallTimeoutManager _filter: FilterType | None _projection: ProjectionType | None _sort: HybridSortType | None _limit: int | None _hybrid_limits: int | dict[str, int] | None _include_scores: bool | None _include_sort_vector: bool | None _rerank_on: str | None _rerank_query: str | None _mapper: Callable[[RerankedResult[TRAW]], T] | None def __init__( self, *, collection: Collection[TRAW], request_timeout_ms: int | None, overall_timeout_ms: int | None, request_timeout_label: str | None = None, overall_timeout_label: str | None = None, filter: FilterType | None = None, projection: ProjectionType | None = None, sort: HybridSortType | None = None, limit: int | None = None, hybrid_limits: int | dict[str, int] | None = None, include_scores: bool | None = None, include_sort_vector: bool | None = None, rerank_on: str | None = None, rerank_query: str | None = None, mapper: Callable[[RerankedResult[TRAW]], T] | None = None, ) -> None: self._filter = deepcopy(filter) self._projection = projection self._sort = deepcopy(sort) self._limit = limit self._hybrid_limits = deepcopy(hybrid_limits) self._include_scores = include_scores self._include_sort_vector = include_sort_vector self._rerank_on = rerank_on self._rerank_query = rerank_query self._mapper = mapper self._request_timeout_ms = request_timeout_ms self._overall_timeout_ms = overall_timeout_ms self._request_timeout_label = request_timeout_label self._overall_timeout_label = overall_timeout_label self._query_engine = _CollectionFindAndRerankQueryEngine( collection=collection, async_collection=None, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, ) AbstractCursor.__init__(self) self._timeout_manager = MultiCallTimeoutManager( overall_timeout_ms=self._overall_timeout_ms, timeout_label=self._overall_timeout_label, ) def _copy( self: CollectionFindAndRerankCursor[TRAW, T], *, request_timeout_ms: int | None | UnsetType = _UNSET, overall_timeout_ms: int | None | UnsetType = _UNSET, request_timeout_label: str | None | UnsetType = _UNSET, overall_timeout_label: str | None | UnsetType = _UNSET, filter: FilterType | None | UnsetType = _UNSET, projection: ProjectionType | None | UnsetType = _UNSET, sort: dict[str, Any] | None | UnsetType = _UNSET, limit: int | None | UnsetType = _UNSET, hybrid_limits: int | dict[str, int] | None | UnsetType = _UNSET, include_scores: bool | None | UnsetType = _UNSET, include_sort_vector: bool | None | UnsetType = _UNSET, rerank_on: str | None | UnsetType = _UNSET, rerank_query: str | None | UnsetType = _UNSET, ) -> CollectionFindAndRerankCursor[TRAW, T]: if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") return CollectionFindAndRerankCursor( collection=self._query_engine.collection, request_timeout_ms=self._request_timeout_ms if isinstance(request_timeout_ms, UnsetType) else request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms if isinstance(overall_timeout_ms, UnsetType) else overall_timeout_ms, request_timeout_label=self._request_timeout_label if isinstance(request_timeout_label, UnsetType) else request_timeout_label, overall_timeout_label=self._overall_timeout_label if isinstance(overall_timeout_label, UnsetType) else overall_timeout_label, filter=self._filter if isinstance(filter, UnsetType) else filter, projection=self._projection if isinstance(projection, UnsetType) else projection, sort=self._sort if isinstance(sort, UnsetType) else sort, limit=self._limit if isinstance(limit, UnsetType) else limit, hybrid_limits=self._hybrid_limits if isinstance(hybrid_limits, UnsetType) else hybrid_limits, include_scores=self._include_scores if isinstance(include_scores, UnsetType) else include_scores, include_sort_vector=self._include_sort_vector if isinstance(include_sort_vector, UnsetType) else include_sort_vector, rerank_on=self._rerank_on if isinstance(rerank_on, UnsetType) else rerank_on, rerank_query=self._rerank_query if isinstance(rerank_query, UnsetType) else rerank_query, mapper=self._mapper, ) def _try_ensure_fill_buffer(self) -> None: """ If buffer is empty, try to fill with next page, if applicable. If not possible, silently do nothing. This method never changes the cursor state. """ if self._state == CursorState.CLOSED: return if not self._buffer: if self._next_page_state is not None or self._state == CursorState.IDLE: new_buffer, next_page_state, resp_status = ( self._query_engine._fetch_page( page_state=self._next_page_state, timeout_context=self._timeout_manager.remaining_timeout( cap_time_ms=self._request_timeout_ms, cap_timeout_label=self._request_timeout_label, ), ) ) self._next_page_state = next_page_state self._last_response_status = resp_status self._pages_retrieved += 1 self._buffer = new_buffer def __repr__(self) -> str: return ( f'{self.__class__.__name__}("{self.data_source.name}", ' f"{self._state.value}, " f"consumed so far: {self.consumed})" ) def __iter__( self: CollectionFindAndRerankCursor[TRAW, T], ) -> CollectionFindAndRerankCursor[TRAW, T]: self._ensure_alive() return self def __next__(self) -> T: if self.state == CursorState.CLOSED: raise StopIteration self._try_ensure_fill_buffer() if not self._buffer: self._state = CursorState.CLOSED raise StopIteration self._state = CursorState.STARTED # consume one item from buffer traw0, rest_buffer = self._buffer[0], self._buffer[1:] self._buffer = rest_buffer self._consumed += 1 return cast(T, self._mapper(traw0) if self._mapper is not None else traw0) @property def data_source(self) -> Collection[TRAW]: """ The Collection object that originated this cursor through a `find_and_rerank` operation. Returns: a Collection instance. """ if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") return self._query_engine.collection def clone(self) -> CollectionFindAndRerankCursor[TRAW, T]: """ Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state. Returns: a new CollectionFindAndRerankCursor, similar to this one but without mapping and rewound to its initial state. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ).map(lambda r_result: r_result.document["wkd"].upper()) >>> for idx, value in zip([0, 1], cursor): ... print(f"{idx} ==> {value}") ... 0 ==> MON 1 ==> TUE >>> cloned_cursor = cursor.clone() >>> for value in cloned_cursor: ... print(f"(cloned) {value}") ... (cloned) MON (cloned) TUE (cloned) SUN >>> >>> print(f"n ==> {next(cursor)}") n ==> SUN """ if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") return CollectionFindAndRerankCursor( collection=self._query_engine.collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=self._mapper, ) def filter( self, filter: FilterType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: filter: a new filter setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `filter` which is the provided value. """ self._ensure_idle() return self._copy(filter=filter) def project( self, projection: ProjectionType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: projection: a new projection setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `projection` which is the provided value. """ self._ensure_idle() if self._mapper is not None: raise CursorException( "Cannot set projection after map.", cursor_state=self._state.value, ) return self._copy(projection=projection) def sort( self, sort: HybridSortType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: sort: a new sort setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `sort` which is the provided value. """ self._ensure_idle() return self._copy(sort=sort) def limit(self, limit: int | None) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: limit: a new limit setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `limit` which is the provided value. """ self._ensure_idle() return self._copy(limit=limit) def hybrid_limits( self, hybrid_limits: int | dict[str, int] | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: hybrid_limits: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `hybrid_limits` which is the provided value. """ self._ensure_idle() return self._copy(hybrid_limits=hybrid_limits) def include_scores( self, include_scores: bool | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: include_scores: a new include_scores setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `include_scores` which is the provided value. """ self._ensure_idle() return self._copy(include_scores=include_scores) def include_sort_vector( self, include_sort_vector: bool | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: include_sort_vector: a new include_sort_vector setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `include_sort_vector` which is the provided value. """ self._ensure_idle() return self._copy(include_sort_vector=include_sort_vector) def rerank_on( self, rerank_on: str | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: rerank_on: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `rerank_on` which is the provided value. """ self._ensure_idle() return self._copy(rerank_on=rerank_on) def rerank_query( self, rerank_query: str | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: rerank_query: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `rerank_query` which is the provided value. """ self._ensure_idle() return self._copy(rerank_query=rerank_query) def map( self, mapper: Callable[[T], TNEW] ) -> CollectionFindAndRerankCursor[TRAW, TNEW]: """ Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed. This operation is allowed only if the cursor state is still IDLE. Args: mapper: a function transforming the objects returned by the cursor into something else (i.e. a function T => TNEW). If the map is imposed on a cursor without mapping yet, its input argument must be a `RerankedResult[TRAW]`, where TRAW stands for the type of the documents from the collection. Returns: a new CollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> for r_result in cursor: ... print(r_result.document) ... {'_id': 'A', 'wkd': 'Mon'} {'_id': 'B', 'wkd': 'Tue'} {'_id': 'G', 'wkd': 'Sun'} >>> cursor_mapped = cursor.clone().map( ... lambda r_result: r_result.document["wkd"] ... ) >>> for value in cursor_mapped: ... print(value) ... Mon Tue Sun >>> cursor_mapped_twice = cursor_mapped.clone().map( ... lambda wkd: f"<{wkd[:2].lower()}>" ... ) >>> for value in cursor_mapped_twice: ... print(value) ... <mo> <tu> <su> """ self._ensure_idle() if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") composite_mapper: Callable[[RerankedResult[TRAW]], TNEW] if self._mapper is not None: def _composite(document: RerankedResult[TRAW]) -> TNEW: return mapper(self._mapper(document)) # type: ignore[misc] composite_mapper = _composite else: composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper) return CollectionFindAndRerankCursor( collection=self._query_engine.collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=composite_mapper, ) def for_each( self, function: Callable[[T], bool | None], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> None: """ Consume the remaining documents in the cursor, invoking a provided callback function on each of them. Calling this method on a CLOSED cursor results in an error. The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean `False`, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted. Args: function: a callback function whose only parameter is of the type returned by the cursor. This callback is invoked once per each document yielded by the cursor. If the callback returns a `False`, the `for_each` invocation stops early and returns without consuming further documents. general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> from astrapy.cursors import CursorState, RerankedResult >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> def printer(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... >>> cursor.for_each(printer) -> Mon -> Tue -> Sun >>> >>> if cursor.state != CursorState.CLOSED: ... print(f"alive: {cursor.to_list()}") ... else: ... print("(closed)") ... (closed) >>> cursor2 = cursor.clone() >>> def checker(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... return r_result.document["wkd"] != "Tue" ... >>> cursor2.for_each(checker) -> Mon -> Tue >>> >>> if cursor2.state != CursorState.CLOSED: ... print(f"alive: {list(cursor2)}") ... else: ... print("(closed)") ... alive: [RerankedResult(document={'_id': 'G', 'wkd': 'Sun'}, scores={})] """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) for document in _cursor: res = function(document) if res is False: break _cursor._imprint_internal_state(self) def to_list( self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> list[T]: """ Materialize all documents that remain to be consumed from a cursor into a list. Calling this method on a CLOSED cursor results in an error. If the cursor is IDLE, the result will be the whole set of documents returned by the `find_and_rerank` operation; otherwise, the documents already consumed by the cursor will not be in the resulting list. Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred. Args: general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Returns: a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when `to_list` is called. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map( ... lambda r_result: r_result.document["wkd"] ... ).to_list() ['Wed', 'Mon', 'Tue', 'Sun'] >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map(lambda r_result: r_result.document["wkd"]) >>> print(f"First item: {cursor.__next__()}.") First item: Wed. >>> cursor.to_list() ['Mon', 'Tue', 'Sun'] """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) documents = [document for document in _cursor] _cursor._imprint_internal_state(self) return documents def has_next(self) -> bool: """ Whether the cursor actually has more documents to return. `has_next` can be called on any cursor, but on a CLOSED cursor will always return False. This method can trigger the fetch operation of a new page, if the current buffer is empty. Calling `has_next` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. Returns: a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor). """ if self._state == CursorState.CLOSED: return False self._try_ensure_fill_buffer() return len(self._buffer) > 0 def get_sort_vector(self) -> list[float] | DataAPIVector | None: """ Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable. Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search. Returns: the query vector used in the search, if it was requested by passing `include_sort_vector=True` to the `find_and_rerank` call that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting for `APIOptions.serdes_options`. """ self._try_ensure_fill_buffer() if self._last_response_status: return _ensure_vector( self._last_response_status.get("sortVector"), self.data_source.api_options.serdes_options, ) else: return NoneAncestors
- AbstractCursor
- abc.ABC
- typing.Generic
Instance variables
var data_source : Collection[~TRAW]-
The Collection object that originated this cursor through a
find_and_rerankoperation.Returns
a Collection instance.
Expand source code
@property def data_source(self) -> Collection[TRAW]: """ The Collection object that originated this cursor through a `find_and_rerank` operation. Returns: a Collection instance. """ if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") return self._query_engine.collection
Methods
def clone(self) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state.
Returns
a new CollectionFindAndRerankCursor, similar to this one but without mapping and rewound to its initial state.
Example
>>> # (this assumes 'vectorize'. See <code>Collection.find\_and\_rerank</code> for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ).map(lambda r_result: r_result.document["wkd"].upper()) >>> for idx, value in zip([0, 1], cursor): ... print(f"{idx} ==> {value}") ... 0 ==> MON 1 ==> TUE >>> cloned_cursor = cursor.clone() >>> for value in cloned_cursor: ... print(f"(cloned) {value}") ... (cloned) MON (cloned) TUE (cloned) SUN >>> >>> print(f"n ==> {next(cursor)}") n ==> SUNExpand source code
def clone(self) -> CollectionFindAndRerankCursor[TRAW, T]: """ Create a copy of this cursor with: - the same parameters (timeouts, filter, projection, etc) - and the cursor is rewound to its pristine IDLE state. Returns: a new CollectionFindAndRerankCursor, similar to this one but without mapping and rewound to its initial state. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ).map(lambda r_result: r_result.document["wkd"].upper()) >>> for idx, value in zip([0, 1], cursor): ... print(f"{idx} ==> {value}") ... 0 ==> MON 1 ==> TUE >>> cloned_cursor = cursor.clone() >>> for value in cloned_cursor: ... print(f"(cloned) {value}") ... (cloned) MON (cloned) TUE (cloned) SUN >>> >>> print(f"n ==> {next(cursor)}") n ==> SUN """ if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") return CollectionFindAndRerankCursor( collection=self._query_engine.collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=self._mapper, ) def filter(self, filter: FilterType | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
filter- a new filter setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
filterwhich is the provided value.Expand source code
def filter( self, filter: FilterType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new filter setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: filter: a new filter setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `filter` which is the provided value. """ self._ensure_idle() return self._copy(filter=filter) def for_each(self, function: Callable[[T], bool | None], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None) ‑> None-
Consume the remaining documents in the cursor, invoking a provided callback function on each of them.
Calling this method on a CLOSED cursor results in an error.
The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean
False, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted.Args
function- a callback function whose only parameter is of the type returned
by the cursor. This callback is invoked once per each document yielded
by the cursor. If the callback returns a
False, thefor_eachinvocation stops early and returns without consuming further documents. general_method_timeout_ms- a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies.
timeout_ms- an alias for
general_method_timeout_ms.
Example
>>> # (this assumes 'vectorize'. See <code>Collection.find\_and\_rerank</code> for more.) >>> from astrapy.cursors import CursorState, RerankedResult >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> def printer(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... >>> cursor.for_each(printer) -> Mon -> Tue -> Sun >>> >>> if cursor.state != CursorState.CLOSED: ... print(f"alive: {cursor.to_list()}") ... else: ... print("(closed)") ... (closed) >>> cursor2 = cursor.clone() >>> def checker(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... return r_result.document["wkd"] != "Tue" ... >>> cursor2.for_each(checker) -> Mon -> Tue >>> >>> if cursor2.state != CursorState.CLOSED: ... print(f"alive: {list(cursor2)}") ... else: ... print("(closed)") ... alive: [RerankedResult(document={'_id': 'G', 'wkd': 'Sun'}, scores={})]Expand source code
def for_each( self, function: Callable[[T], bool | None], *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> None: """ Consume the remaining documents in the cursor, invoking a provided callback function on each of them. Calling this method on a CLOSED cursor results in an error. The callback function can return any value. The return value is generally discarded, with the following exception: if the function returns the boolean `False`, it is taken to signify that the method should quit early, leaving the cursor half-consumed (ACTIVE state). If this does not occur, this method results in the cursor entering CLOSED state once it is exhausted. Args: function: a callback function whose only parameter is of the type returned by the cursor. This callback is invoked once per each document yielded by the cursor. If the callback returns a `False`, the `for_each` invocation stops early and returns without consuming further documents. general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> from astrapy.cursors import CursorState, RerankedResult >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> def printer(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... >>> cursor.for_each(printer) -> Mon -> Tue -> Sun >>> >>> if cursor.state != CursorState.CLOSED: ... print(f"alive: {cursor.to_list()}") ... else: ... print("(closed)") ... (closed) >>> cursor2 = cursor.clone() >>> def checker(r_result: RerankedResult): ... print(f"-> {r_result.document['wkd']}") ... return r_result.document["wkd"] != "Tue" ... >>> cursor2.for_each(checker) -> Mon -> Tue >>> >>> if cursor2.state != CursorState.CLOSED: ... print(f"alive: {list(cursor2)}") ... else: ... print("(closed)") ... alive: [RerankedResult(document={'_id': 'G', 'wkd': 'Sun'}, scores={})] """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) for document in _cursor: res = function(document) if res is False: break _cursor._imprint_internal_state(self) def get_sort_vector(self) ‑> list[float] | DataAPIVector | None-
Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable.
Calling
get_sort_vectoron an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts.The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search.
Returns
the query vector used in the search, if it was requested by passing
include_sort_vector=Trueto thefind_and_rerankcall that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting forAPIOptions.serdes_options.Expand source code
def get_sort_vector(self) -> list[float] | DataAPIVector | None: """ Return the query vector used in the vector (ANN) search that was run as part of the search expressed by this cursor, if applicable. Calling `get_sort_vector` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. The method can be invoked on a CLOSED cursor and will return either None or the sort vector used in the search. Returns: the query vector used in the search, if it was requested by passing `include_sort_vector=True` to the `find_and_rerank` call that originated the cursor. If the sort vector is not available, None is returned. Otherwise, the vector is returned as either a DataAPIVector or a plain list of number depending on the setting for `APIOptions.serdes_options`. """ self._try_ensure_fill_buffer() if self._last_response_status: return _ensure_vector( self._last_response_status.get("sortVector"), self.data_source.api_options.serdes_options, ) else: return None def has_next(self) ‑> bool-
Whether the cursor actually has more documents to return.
has_nextcan be called on any cursor, but on a CLOSED cursor will always return False.This method can trigger the fetch operation of a new page, if the current buffer is empty.
Calling
has_nexton an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts.Returns
a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor).
Expand source code
def has_next(self) -> bool: """ Whether the cursor actually has more documents to return. `has_next` can be called on any cursor, but on a CLOSED cursor will always return False. This method can trigger the fetch operation of a new page, if the current buffer is empty. Calling `has_next` on an IDLE cursor triggers the first page fetch, but the cursor stays in the IDLE state until actual consumption starts. Returns: a boolean value of True if there is at least one further item available to consume; False otherwise (including the case of CLOSED cursor). """ if self._state == CursorState.CLOSED: return False self._try_ensure_fill_buffer() return len(self._buffer) > 0 def hybrid_limits(self, hybrid_limits: int | dict[str, int] | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
hybrid_limits- a new setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
hybrid_limitswhich is the provided value.Expand source code
def hybrid_limits( self, hybrid_limits: int | dict[str, int] | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new hybrid_limits setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: hybrid_limits: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `hybrid_limits` which is the provided value. """ self._ensure_idle() return self._copy(hybrid_limits=hybrid_limits) def include_scores(self, include_scores: bool | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
include_scores- a new include_scores setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
include_scoreswhich is the provided value.Expand source code
def include_scores( self, include_scores: bool | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_scores setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: include_scores: a new include_scores setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `include_scores` which is the provided value. """ self._ensure_idle() return self._copy(include_scores=include_scores) def include_sort_vector(self, include_sort_vector: bool | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
include_sort_vector- a new include_sort_vector setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
include_sort_vectorwhich is the provided value.Expand source code
def include_sort_vector( self, include_sort_vector: bool | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new include_sort_vector setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: include_sort_vector: a new include_sort_vector setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `include_sort_vector` which is the provided value. """ self._ensure_idle() return self._copy(include_sort_vector=include_sort_vector) def limit(self, limit: int | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
limit- a new limit setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
limitwhich is the provided value.Expand source code
def limit(self, limit: int | None) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new limit setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: limit: a new limit setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `limit` which is the provided value. """ self._ensure_idle() return self._copy(limit=limit) def map(self, mapper: Callable[[T], TNEW]) ‑> CollectionFindAndRerankCursor[~TRAW, ~TNEW]-
Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed.
This operation is allowed only if the cursor state is still IDLE.
Args
mapper- a function transforming the objects returned by the cursor
into something else (i.e. a function T => TNEW).
If the map is imposed on a cursor without mapping yet, its input
argument must be a
RerankedResult[TRAW], where TRAW stands for the type of the documents from the collection.
Returns
a new CollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function.
Example
>>> # (this assumes 'vectorize'. See <code>Collection.find\_and\_rerank</code> for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> for r_result in cursor: ... print(r_result.document) ... {'_id': 'A', 'wkd': 'Mon'} {'_id': 'B', 'wkd': 'Tue'} {'_id': 'G', 'wkd': 'Sun'} >>> cursor_mapped = cursor.clone().map( ... lambda r_result: r_result.document["wkd"] ... ) >>> for value in cursor_mapped: ... print(value) ... Mon Tue Sun >>> cursor_mapped_twice = cursor_mapped.clone().map( ... lambda wkd: f"<{wkd[:2].lower()}>" ... ) >>> for value in cursor_mapped_twice: ... print(value) ... <mo> <tu> <su>Expand source code
def map( self, mapper: Callable[[T], TNEW] ) -> CollectionFindAndRerankCursor[TRAW, TNEW]: """ Return a copy of this cursor with a mapping function to transform the returned items. Calling this method on a cursor with a mapping already set results in the mapping functions being composed. This operation is allowed only if the cursor state is still IDLE. Args: mapper: a function transforming the objects returned by the cursor into something else (i.e. a function T => TNEW). If the map is imposed on a cursor without mapping yet, its input argument must be a `RerankedResult[TRAW]`, where TRAW stands for the type of the documents from the collection. Returns: a new CollectionFindAndRerankCursor with a new mapping function on the results, possibly composed with any pre-existing mapping function. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=3, ... ) >>> for r_result in cursor: ... print(r_result.document) ... {'_id': 'A', 'wkd': 'Mon'} {'_id': 'B', 'wkd': 'Tue'} {'_id': 'G', 'wkd': 'Sun'} >>> cursor_mapped = cursor.clone().map( ... lambda r_result: r_result.document["wkd"] ... ) >>> for value in cursor_mapped: ... print(value) ... Mon Tue Sun >>> cursor_mapped_twice = cursor_mapped.clone().map( ... lambda wkd: f"<{wkd[:2].lower()}>" ... ) >>> for value in cursor_mapped_twice: ... print(value) ... <mo> <tu> <su> """ self._ensure_idle() if self._query_engine.collection is None: raise RuntimeError("Query engine has no collection.") composite_mapper: Callable[[RerankedResult[TRAW]], TNEW] if self._mapper is not None: def _composite(document: RerankedResult[TRAW]) -> TNEW: return mapper(self._mapper(document)) # type: ignore[misc] composite_mapper = _composite else: composite_mapper = cast(Callable[[RerankedResult[TRAW]], TNEW], mapper) return CollectionFindAndRerankCursor( collection=self._query_engine.collection, request_timeout_ms=self._request_timeout_ms, overall_timeout_ms=self._overall_timeout_ms, request_timeout_label=self._request_timeout_label, overall_timeout_label=self._overall_timeout_label, filter=self._filter, projection=self._projection, sort=self._sort, limit=self._limit, hybrid_limits=self._hybrid_limits, include_scores=self._include_scores, include_sort_vector=self._include_sort_vector, rerank_on=self._rerank_on, rerank_query=self._rerank_query, mapper=composite_mapper, ) def project(self, projection: ProjectionType | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
projection- a new projection setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
projectionwhich is the provided value.Expand source code
def project( self, projection: ProjectionType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new projection setting. This operation is allowed only if the cursor state is still IDLE and if no mapping has been set on it. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: projection: a new projection setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `projection` which is the provided value. """ self._ensure_idle() if self._mapper is not None: raise CursorException( "Cannot set projection after map.", cursor_state=self._state.value, ) return self._copy(projection=projection) def rerank_on(self, rerank_on: str | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
rerank_on- a new setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
rerank_onwhich is the provided value.Expand source code
def rerank_on( self, rerank_on: str | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_on setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: rerank_on: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `rerank_on` which is the provided value. """ self._ensure_idle() return self._copy(rerank_on=rerank_on) def rerank_query(self, rerank_query: str | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
rerank_query- a new setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
rerank_querywhich is the provided value.Expand source code
def rerank_query( self, rerank_query: str | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new rerank_query setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: rerank_query: a new setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `rerank_query` which is the provided value. """ self._ensure_idle() return self._copy(rerank_query=rerank_query) def sort(self, sort: HybridSortType | None) ‑> CollectionFindAndRerankCursor[~TRAW, ~T]-
Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE.
Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection
find_and_rerankmethod.Args
sort- a new sort setting to apply to the returned new cursor.
Returns
a new CollectionFindAndRerankCursor with the same settings as this one, except for
sortwhich is the provided value.Expand source code
def sort( self, sort: HybridSortType | None ) -> CollectionFindAndRerankCursor[TRAW, T]: """ Return a copy of this cursor with a new sort setting. This operation is allowed only if the cursor state is still IDLE. Instead of explicitly invoking this method, the typical usage consists in passing arguments to the Collection `find_and_rerank` method. Args: sort: a new sort setting to apply to the returned new cursor. Returns: a new CollectionFindAndRerankCursor with the same settings as this one, except for `sort` which is the provided value. """ self._ensure_idle() return self._copy(sort=sort) def to_list(self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None) ‑> list[~T]-
Materialize all documents that remain to be consumed from a cursor into a list.
Calling this method on a CLOSED cursor results in an error.
If the cursor is IDLE, the result will be the whole set of documents returned by the
find_and_rerankoperation; otherwise, the documents already consumed by the cursor will not be in the resulting list.Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred.
Args
general_method_timeout_ms- a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies.
timeout_ms- an alias for
general_method_timeout_ms.
Returns
a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when
to_listis called.Example
>>> # (this assumes 'vectorize'. See <code>Collection.find\_and\_rerank</code> for more.) >>> collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map( ... lambda r_result: r_result.document["wkd"] ... ).to_list() ['Wed', 'Mon', 'Tue', 'Sun'] >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map(lambda r_result: r_result.document["wkd"]) >>> print(f"First item: {cursor.__next__()}.") First item: Wed. >>> cursor.to_list() ['Mon', 'Tue', 'Sun']Expand source code
def to_list( self, *, general_method_timeout_ms: int | None = None, timeout_ms: int | None = None, ) -> list[T]: """ Materialize all documents that remain to be consumed from a cursor into a list. Calling this method on a CLOSED cursor results in an error. If the cursor is IDLE, the result will be the whole set of documents returned by the `find_and_rerank` operation; otherwise, the documents already consumed by the cursor will not be in the resulting list. Calling this method is not recommended if a huge list of results is anticipated: it would involve a large number of data exchanges with the Data API and possibly a massive memory usage to construct the list. In such cases, a lazy pattern of iterating and consuming the documents is to be preferred. Args: general_method_timeout_ms: a timeout, in milliseconds, for the whole duration of this method. If not provided, there is no such timeout. Note that the per-request timeout set on the cursor still applies. timeout_ms: an alias for `general_method_timeout_ms`. Returns: a list of documents (or other values depending on the mapping function, if one is set). These are all items that were left to be consumed on the cursor when `to_list` is called. Example: >>> # (this assumes 'vectorize'. See `Collection.find_and_rerank` for more.) >>> collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map( ... lambda r_result: r_result.document["wkd"] ... ).to_list() ['Wed', 'Mon', 'Tue', 'Sun'] >>> >>> cursor = collection.find_and_rerank( ... sort={"$hybrid": "Weekdays?"}, ... projection={"wkd": True}, ... limit=4, ... ).map(lambda r_result: r_result.document["wkd"]) >>> print(f"First item: {cursor.__next__()}.") First item: Wed. >>> cursor.to_list() ['Mon', 'Tue', 'Sun'] """ self._ensure_alive() copy_req_ms, copy_ovr_ms = _revise_timeouts_for_cursor_copy( new_general_method_timeout_ms=general_method_timeout_ms, new_timeout_ms=timeout_ms, old_request_timeout_ms=self._request_timeout_ms, ) _cursor = self._copy( request_timeout_ms=copy_req_ms, overall_timeout_ms=copy_ovr_ms, ) self._imprint_internal_state(_cursor) documents = [document for document in _cursor] _cursor._imprint_internal_state(self) return documents
Inherited members