Archives/langchain
2
0
Fork 1
mirror of https://github.com/hwchase17/langchain synced 2024-11-02 09:40:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Implemented MongoDB Atlas Self-Query Retriever (#13321)

Browse Source 
# Description 
This PR implements Self-Query Retriever for MongoDB Atlas vector store.

I've implemented the comparators and operators that are supported by
MongoDB Atlas vector store according to the section titled "Atlas Vector
Search Pre-Filter" from
https://www.mongodb.com/docs/atlas/atlas-vector-search/vector-search-stage/.

Namely:
```
allowed_comparators = [
      Comparator.EQ,
      Comparator.NE,
      Comparator.GT,
      Comparator.GTE,
      Comparator.LT,
      Comparator.LTE,
      Comparator.IN,
      Comparator.NIN,
  ]

"""Subset of allowed logical operators."""
allowed_operators = [
    Operator.AND,
    Operator.OR
]
```
Translations from comparators/operators to MongoDB Atlas filter
operators(you can find the syntax in the "Atlas Vector Search
Pre-Filter" section from the previous link) are done using the following
dictionary:
```
map_dict = {
            Operator.AND: "$and",
            Operator.OR: "$or",
            Comparator.EQ: "$eq",
            Comparator.NE: "$ne",
            Comparator.GTE: "$gte",
            Comparator.LTE: "$lte",
            Comparator.LT: "$lt",
            Comparator.GT: "$gt",
            Comparator.IN: "$in",
            Comparator.NIN: "$nin",
        }
```

In visit_structured_query() the filters are passed as "pre_filter" and
not "filter" as in the MongoDB link above since langchain's
implementation of MongoDB atlas vector
store(libs\langchain\langchain\vectorstores\mongodb_atlas.py) in
_similarity_search_with_score() sets the "filter" key to have the value
of the "pre_filter" argument.
```
params["filter"] = pre_filter
```
Test cases and documentation have also been added.

# Issue
#11616 

# Dependencies
No new dependencies have been added.

# Documentation
I have created the notebook mongodb_atlas_self_query.ipynb outlining the
steps to get the self-query mechanism working.

I worked closely with [@Farhan-Faisal](https://github.com/Farhan-Faisal)
on this PR.

---------

Co-authored-by: Bagatur <baskaryan@gmail.com>
This commit is contained in:
AthulVincent 2023-11-29 22:05:06 -05:00 committed by GitHub
parent c2e3963da4
commit 67c55cb5b0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 529 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

321
docs/docs/integrations/retrievers/self_query/mongodb_atlas.ipynb Normal file
View File

@ -0,0 +1,321 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# MongoDB Atlas\n",
"\n",
"[MongoDB Atlas](https://www.mongodb.com/) is a document database that can be \n",
"used as a vector databse.\n",
"\n",
"In the walkthrough, we'll demo the `SelfQueryRetriever` with a `MongoDB Atlas` vector store."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Creating a MongoDB Atlas vectorstore\n",
"First we'll want to create a MongoDB Atlas VectorStore and seed it with some data. We've created a small demo set of documents that contain summaries of movies.\n",
"\n",
"NOTE: The self-query retriever requires you to have `lark` installed (`pip install lark`). We also need the `pymongo` package."
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"#!pip install lark pymongo"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We want to use `OpenAIEmbeddings` so we have to get the OpenAI API Key."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"\n",
"OPENAI_API_KEY = \"Use your OpenAI key\"\n",
"\n",
"os.environ[\"OPENAI_API_KEY\"] = OPENAI_API_KEY"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from langchain.embeddings.openai import OpenAIEmbeddings\n",
"from langchain.schema import Document\n",
"from langchain.vectorstores import MongoDBAtlasVectorSearch\n",
"from pymongo import MongoClient\n",
"\n",
"CONNECTION_STRING = \"Use your MongoDB Atlas connection string\"\n",
"DB_NAME = \"Name of your MongoDB Atlas database\"\n",
"COLLECTION_NAME = \"Name of your collection in the database\"\n",
"INDEX_NAME = \"Name of a search index defined on the collection\"\n",
"\n",
"MongoClient = MongoClient(CONNECTION_STRING)\n",
"collection = MongoClient[DB_NAME][COLLECTION_NAME]\n",
"\n",
"embeddings = OpenAIEmbeddings()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"docs = [\n",
" Document(\n",
" page_content=\"A bunch of scientists bring back dinosaurs and mayhem breaks loose\",\n",
" metadata={\"year\": 1993, \"rating\": 7.7, \"genre\": \"action\"},\n",
" ),\n",
" Document(\n",
" page_content=\"Leo DiCaprio gets lost in a dream within a dream within a dream within a ...\",\n",
" metadata={\"year\": 2010, \"genre\": \"thriller\", \"rating\": 8.2},\n",
" ),\n",
" Document(\n",
" page_content=\"A bunch of normal-sized women are supremely wholesome and some men pine after them\",\n",
" metadata={\"year\": 2019, \"rating\": 8.3, \"genre\": \"drama\"},\n",
" ),\n",
" Document(\n",
" page_content=\"Three men walk into the Zone, three men walk out of the Zone\",\n",
" metadata={\"year\": 1979, \"rating\": 9.9, \"genre\": \"science fiction\"},\n",
" ),\n",
" Document(\n",
" page_content=\"A psychologist / detective gets lost in a series of dreams within dreams within dreams and Inception reused the idea\",\n",
" metadata={\"year\": 2006, \"genre\": \"thriller\", \"rating\": 9.0},\n",
" ),\n",
" Document(\n",
" page_content=\"Toys come alive and have a blast doing so\",\n",
" metadata={\"year\": 1995, \"genre\": \"animated\", \"rating\": 9.3},\n",
" ),\n",
"]\n",
"\n",
"vectorstore = MongoDBAtlasVectorSearch.from_documents(\n",
" docs,\n",
" embeddings,\n",
" collection=collection,\n",
" index_name=INDEX_NAME,\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Now, let's create a vector search index on your cluster. In the below example, `embedding` is the name of the field that contains the embedding vector. Please refer to the [documentation](https://www.mongodb.com/docs/atlas/atlas-search/field-types/knn-vector) to get more details on how to define an Atlas Vector Search index.\n",
"You can name the index `{COLLECTION_NAME}` and create the index on the namespace `{DB_NAME}.{COLLECTION_NAME}`. Finally, write the following definition in the JSON editor on MongoDB Atlas:\n",
"\n",
"```json\n",
"{\n",
" \"mappings\": {\n",
" \"dynamic\": true,\n",
" \"fields\": {\n",
" \"embedding\": {\n",
" \"dimensions\": 1536,\n",
" \"similarity\": \"cosine\",\n",
" \"type\": \"knnVector\"\n",
" },\n",
" \"genre\": {\n",
" \"type\": \"token\"\n",
" },\n",
" \"ratings\": {\n",
" \"type\": \"number\"\n",
" },\n",
" \"year\": {\n",
" \"type\": \"number\"\n",
" }\n",
" }\n",
" }\n",
"}\n",
"```"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Creating our self-querying retriever\n",
"Now we can instantiate our retriever. To do this we'll need to provide some information upfront about the metadata fields that our documents support and a short description of the document contents."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from langchain.chains.query_constructor.base import AttributeInfo\n",
"from langchain.llms import OpenAI\n",
"from langchain.retrievers.self_query.base import SelfQueryRetriever\n",
"\n",
"metadata_field_info = [\n",
" AttributeInfo(\n",
" name=\"genre\",\n",
" description=\"The genre of the movie\",\n",
" type=\"string\",\n",
" ),\n",
" AttributeInfo(\n",
" name=\"year\",\n",
" description=\"The year the movie was released\",\n",
" type=\"integer\",\n",
" ),\n",
" AttributeInfo(\n",
" name=\"rating\", description=\"A 1-10 rating for the movie\", type=\"float\"\n",
" ),\n",
"]\n",
"document_content_description = \"Brief summary of a movie\""
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"llm = OpenAI(temperature=0)\n",
"retriever = SelfQueryRetriever.from_llm(\n",
" llm, vectorstore, document_content_description, metadata_field_info, verbose=True\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Testing it out\n",
"And now we can try actually using our retriever!"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example only specifies a relevant query\n",
"retriever.get_relevant_documents(\"What are some movies about dinosaurs\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example specifies a filter\n",
"retriever.get_relevant_documents(\"What are some highly rated movies (above 9)?\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example only specifies a query and a filter\n",
"retriever.get_relevant_documents(\n",
" \"I want to watch a movie about toys rated higher than 9\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example specifies a composite filter\n",
"retriever.get_relevant_documents(\n",
" \"What's a highly rated (above or equal 9) thriller film?\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example specifies a query and composite filter\n",
"retriever.get_relevant_documents(\n",
" \"What's a movie after 1990 but before 2005 that's all about dinosaurs, \\\n",
" and preferably has a lot of action\"\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Filter k\n",
"\n",
"We can also use the self query retriever to specify `k`: the number of documents to fetch.\n",
"\n",
"We can do this by passing `enable_limit=True` to the constructor."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"retriever = SelfQueryRetriever.from_llm(\n",
" llm,\n",
" vectorstore,\n",
" document_content_description,\n",
" metadata_field_info,\n",
" verbose=True,\n",
" enable_limit=True,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# This example only specifies a relevant query\n",
"retriever.get_relevant_documents(\"What are two movies about dinosaurs?\")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": ".venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.5"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

3
libs/langchain/langchain/retrievers/self_query/base.py
View File

@ -21,6 +21,7 @@ from langchain.retrievers.self_query.dashvector import DashvectorTranslator
from langchain.retrievers.self_query.deeplake import DeepLakeTranslator
from langchain.retrievers.self_query.elasticsearch import ElasticsearchTranslator
from langchain.retrievers.self_query.milvus import MilvusTranslator
from langchain.retrievers.self_query.mongodb_atlas import MongoDBAtlasTranslator
from langchain.retrievers.self_query.myscale import MyScaleTranslator
from langchain.retrievers.self_query.opensearch import OpenSearchTranslator
from langchain.retrievers.self_query.pinecone import PineconeTranslator
@ -36,6 +37,7 @@ from langchain.vectorstores import (
DeepLake,
ElasticsearchStore,
Milvus,
MongoDBAtlasVectorSearch,
MyScale,
OpenSearchVectorSearch,
Pinecone,
@ -66,6 +68,7 @@ def _get_builtin_translator(vectorstore: VectorStore) -> Visitor:
SupabaseVectorStore: SupabaseVectorTranslator,
TimescaleVector: TimescaleVectorTranslator,
OpenSearchVectorSearch: OpenSearchTranslator,
MongoDBAtlasVectorSearch: MongoDBAtlasTranslator,
}
if isinstance(vectorstore, Qdrant):
return QdrantTranslator(metadata_key=vectorstore.metadata_payload_key)

74
libs/langchain/langchain/retrievers/self_query/mongodb_atlas.py Normal file
View File

@ -0,0 +1,74 @@
"""Logic for converting internal query language to a valid MongoDB Atlas query."""
from typing import Dict, Tuple, Union
from langchain.chains.query_constructor.ir import (
Comparator,
Comparison,
Operation,
Operator,
StructuredQuery,
Visitor,
)
MULTIPLE_ARITY_COMPARATORS = [Comparator.IN, Comparator.NIN]
class MongoDBAtlasTranslator(Visitor):
"""Translate Mongo internal query language elements to valid filters."""
"""Subset of allowed logical comparators."""
allowed_comparators = [
Comparator.EQ,
Comparator.NE,
Comparator.GT,
Comparator.GTE,
Comparator.LT,
Comparator.LTE,
Comparator.IN,
Comparator.NIN,
]
"""Subset of allowed logical operators."""
allowed_operators = [Operator.AND, Operator.OR]
## Convert a operator or a comparator to Mongo Query Format
def _format_func(self, func: Union[Operator, Comparator]) -> str:
self._validate_func(func)
map_dict = {
Operator.AND: "$and",
Operator.OR: "$or",
Comparator.EQ: "$eq",
Comparator.NE: "$ne",
Comparator.GTE: "$gte",
Comparator.LTE: "$lte",
Comparator.LT: "$lt",
Comparator.GT: "$gt",
Comparator.IN: "$in",
Comparator.NIN: "$nin",
}
return map_dict[func]
def visit_operation(self, operation: Operation) -> Dict:
args = [arg.accept(self) for arg in operation.arguments]
return {self._format_func(operation.operator): args}
def visit_comparison(self, comparison: Comparison) -> Dict:
if comparison.comparator in MULTIPLE_ARITY_COMPARATORS and not isinstance(
comparison.value, list
):
comparison.value = [comparison.value]
comparator = self._format_func(comparison.comparator)
attribute = comparison.attribute
return {attribute: {comparator: comparison.value}}
def visit_structured_query(
self, structured_query: StructuredQuery
) -> Tuple[str, dict]:
if structured_query.filter is None:
kwargs = {}
else:
kwargs = {"pre_filter": structured_query.filter.accept(self)}
return structured_query.query, kwargs

131
libs/langchain/tests/unit_tests/retrievers/self_query/test_mongodb_atlas.py Normal file
View File

@ -0,0 +1,131 @@
from typing import Dict, Tuple
from langchain.chains.query_constructor.ir import (
Comparator,
Comparison,
Operation,
Operator,
StructuredQuery,
)
from langchain.retrievers.self_query.mongodb_atlas import MongoDBAtlasTranslator
DEFAULT_TRANSLATOR = MongoDBAtlasTranslator()
def test_visit_comparison_lt() -> None:
comp = Comparison(comparator=Comparator.LT, attribute="qty", value=20)
expected = {"qty": {"$lt": 20}}
actual = DEFAULT_TRANSLATOR.visit_comparison(comp)
assert expected == actual
def test_visit_comparison_eq() -> None:
comp = Comparison(comparator=Comparator.EQ, attribute="qty", value=10)
expected = {"qty": {"$eq": 10}}
actual = DEFAULT_TRANSLATOR.visit_comparison(comp)
assert expected == actual
def test_visit_comparison_ne() -> None:
comp = Comparison(comparator=Comparator.NE, attribute="name", value="foo")
expected = {"name": {"$ne": "foo"}}
actual = DEFAULT_TRANSLATOR.visit_comparison(comp)
assert expected == actual
def test_visit_comparison_in() -> None:
comp = Comparison(comparator=Comparator.IN, attribute="name", value="foo")
expected = {"name": {"$in": ["foo"]}}
actual = DEFAULT_TRANSLATOR.visit_comparison(comp)
assert expected == actual
def test_visit_comparison_nin() -> None:
comp = Comparison(comparator=Comparator.NIN, attribute="name", value="foo")
expected = {"name": {"$nin": ["foo"]}}
actual = DEFAULT_TRANSLATOR.visit_comparison(comp)
assert expected == actual
def test_visit_operation() -> None:
op = Operation(
operator=Operator.AND,
arguments=[
Comparison(comparator=Comparator.GTE, attribute="qty", value=10),
Comparison(comparator=Comparator.LTE, attribute="qty", value=20),
Comparison(comparator=Comparator.EQ, attribute="name", value="foo"),
],
)
expected = {
"$and": [
{"qty": {"$gte": 10}},
{"qty": {"$lte": 20}},
{"name": {"$eq": "foo"}},
]
}
actual = DEFAULT_TRANSLATOR.visit_operation(op)
assert expected == actual
def test_visit_structured_query_no_filter() -> None:
query = "What is the capital of France?"
structured_query = StructuredQuery(
query=query,
filter=None,
)
expected: Tuple[str, Dict] = (query, {})
actual = DEFAULT_TRANSLATOR.visit_structured_query(structured_query)
assert expected == actual
def test_visit_structured_query_one_attr() -> None:
query = "What is the capital of France?"
comp = Comparison(comparator=Comparator.IN, attribute="qty", value=[5, 15, 20])
structured_query = StructuredQuery(
query=query,
filter=comp,
)
expected = (
query,
{"pre_filter": {"qty": {"$in": [5, 15, 20]}}},
)
actual = DEFAULT_TRANSLATOR.visit_structured_query(structured_query)
assert expected == actual
def test_visit_structured_query_deep_nesting() -> None:
query = "What is the capital of France?"
op = Operation(
operator=Operator.AND,
arguments=[
Comparison(comparator=Comparator.EQ, attribute="name", value="foo"),
Operation(
operator=Operator.OR,
arguments=[
Comparison(comparator=Comparator.GT, attribute="qty", value=6),
Comparison(
comparator=Comparator.NIN,
attribute="tags",
value=["bar", "foo"],
),
],
),
],
)
structured_query = StructuredQuery(
query=query,
filter=op,
)
expected = (
query,
{
"pre_filter": {
"$and": [
{"name": {"$eq": "foo"}},
{"$or": [{"qty": {"$gt": 6}}, {"tags": {"$nin": ["bar", "foo"]}}]},
]
}
},
)
actual = DEFAULT_TRANSLATOR.visit_structured_query(structured_query)
assert expected == actual