## Description
Adds Supabase Vector as a self-querying retriever.
- Designed to be backwards compatible with existing `filter` logic on
`SupabaseVectorStore`.
- Adds new filter `postgrest_filter` to `SupabaseVectorStore`
`similarity_search()` methods
- Supports entire PostgREST [filter query
language](https://postgrest.org/en/stable/references/api/tables_views.html#read)
(used by self-querying retriever, but also works as an escape hatch for
more query control)
- `SupabaseVectorTranslator` converts Langchain filter into the above
PostgREST query
- Adds Jupyter Notebook for the self-querying retriever
- Adds tests
## Tag maintainer
@hwchase17
## Twitter handle
[@ggrdson](https://twitter.com/ggrdson)
- Description: Adding support for self-querying to Vectara integration
- Issue: per customer request
- Tag maintainer: @rlancemartin @baskaryan
- Twitter handle: @ofermend
Also updated some documentation, added self-query testing, and a demo
notebook with self-query example.
- Implemented the MilvusTranslator for self-querying using Milvus vector
store
- Made unit tests to test its functionality
- Documented the Milvus self-querying
Various miscellaneous fixes to most pages in the 'Retrievers' section of
the documentation:
- "VectorStore" and "vectorstore" changed to "vector store" for
consistency
- Various spelling, grammar, and formatting improvements for readability
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Now with ElasticsearchStore VectorStore merged, i've added support for
the self-query retriever.
I've added a notebook also to demonstrate capability. I've also added
unit tests.
**Credit**
@elastic and @phoey1 on twitter.
Given a user question, this will -
* Use LLM to generate a set of queries.
* Query for each.
* The URLs from search results are stored in self.urls.
* A check is performed for any new URLs that haven't been processed yet
(not in self.url_database).
* Only these new URLs are loaded, transformed, and added to the
vectorstore.
* The vectorstore is queried for relevant documents based on the
questions generated by the LLM.
* Only unique documents are returned as the final result.
This code will avoid reprocessing of URLs across multiple runs of
similar queries, which should improve the performance of the retriever.
It also keeps track of all URLs that have been processed, which could be
useful for debugging or understanding the retriever's behavior.
---------
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
- Until now, hybrid search was limited to modules requiring external
services, such as Weaviate/Pinecone Hybrid Search. However, I have
developed a hybrid retriever that can merge a list of retrievers using
the [Reciprocal Rank
Fusion](https://plg.uwaterloo.ca/~gvcormac/cormacksigir09-rrf.pdf)
algorithm. This new approach, similar to Weaviate hybrid search, does
not require the initialization of any external service.
- Dependencies: No - Twitter handle: dayuanjian21687
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
I've extended the support of async API to local Qdrant mode. It is faked
but allows prototyping without spinning a container. The tests are
improved to test the in-memory case as well.
@baskaryan @rlancemartin @eyurtsev @agola11
** This should land Monday the 17th **
Chroma is upgrading from `0.3.29` to `0.4.0`. `0.4.0` is easier to
build, more durable, faster, smaller, and more extensible. This comes
with a few changes:
1. A simplified and improved client setup. Instead of having to remember
weird settings, users can just do `EphemeralClient`, `PersistentClient`
or `HttpClient` (the underlying direct `Client` implementation is also
still accessible)
2. We migrated data stores away from `duckdb` and `clickhouse`. This
changes the api for the `PersistentClient` that used to reference
`chroma_db_impl="duckdb+parquet"`. Now we simply set
`is_persistent=true`. `is_persistent` is set for you to `true` if you
use `PersistentClient`.
3. Because we migrated away from `duckdb` and `clickhouse` - this also
means that users need to migrate their data into the new layout and
schema. Chroma is committed to providing extension notification and
tooling around any schema and data migrations (for example - this PR!).
After upgrading to `0.4.0` - if users try to access their data that was
stored in the previous regime, the system will throw an `Exception` and
instruct them how to use the migration assistant to migrate their data.
The migration assitant is a pip installable CLI: `pip install
chroma_migrate`. And is runnable by calling `chroma_migrate`
-- TODO ADD here is a short video demonstrating how it works.
Please reference the readme at
[chroma-core/chroma-migrate](https://github.com/chroma-core/chroma-migrate)
to see a full write-up of our philosophy on migrations as well as more
details about this particular migration.
Please direct any users facing issues upgrading to our Discord channel
called
[#get-help](https://discord.com/channels/1073293645303795742/1129200523111841883).
We have also created a [email
listserv](https://airtable.com/shrHaErIs1j9F97BE) to notify developers
directly in the future about breaking changes.
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Description: This PR adds the option to retrieve scores and explanations
in the WeaviateHybridSearchRetriever. This feature improves the
usability of the retriever by allowing users to understand the scoring
logic behind the search results and further refine their search queries.
Issue: This PR is a solution to the issue #7855
Dependencies: This PR does not introduce any new dependencies.
Tag maintainer: @rlancemartin, @eyurtsev
I have included a unit test for the added feature, ensuring that it
retrieves scores and explanations correctly. I have also included an
example notebook demonstrating its use.
Motivation, it seems that when dealing with a long context and "big"
number of relevant documents we must avoid using out of the box score
ordering from vector stores.
See: https://arxiv.org/pdf/2306.01150.pdf
So, I added an additional parameter that allows you to reorder the
retrieved documents so we can work around this performance degradation.
The relevance respect the original search score but accommodates the
lest relevant document in the middle of the context.
Extract from the paper (one image speaks 1000 tokens):
This seems to be common to all diff arquitectures. SO I think we need a
good generic way to implement this reordering and run some test in our
already running retrievers.
It could be that my approach is not the best one from the architecture
point of view, happy to have a discussion about that.
For me this was the best place to introduce the change and start
retesting diff implementations.
@rlancemartin, @eyurtsev
---------
Co-authored-by: Lance Martin <lance@langchain.dev>
- Description: Add a BM25 Retriever that do not need Elastic search
- Dependencies: rank_bm25(if it is not installed it will be install by
using pip, just like TFIDFRetriever do)
- Tag maintainer: @rlancemartin, @eyurtsev
- Twitter handle: DayuanJian21687
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Multiple people have asked in #5081 for a way to limit the documents
returned from an AzureCognitiveSearchRetriever. This PR adds the `top_n`
parameter to allow that.
Twitter handle:
[@UmerHAdil](twitter.com/umerHAdil)
**Description: a description of the change**
Fixed `make docs_build` and related scripts which caused errors. There
are several changes.
First, I made the build of the documentation and the API Reference into
two separate commands. This is because it takes less time to build. The
commands for documents are `make docs_build`, `make docs_clean`, and
`make docs_linkcheck`. The commands for API Reference are `make
api_docs_build`, `api_docs_clean`, and `api_docs_linkcheck`.
It looked like `docs/.local_build.sh` could be used to build the
documentation, so I used that. Since `.local_build.sh` was also building
API Rerefence internally, I removed that process. `.local_build.sh` also
added some Bash options to stop in error or so. Futher more added `cd
"${SCRIPT_DIR}"` at the beginning so that the script will work no matter
which directory it is executed in.
`docs/api_reference/api_reference.rst` is removed, because which is
generated by `docs/api_reference/create_api_rst.py`, and added it to
.gitignore.
Finally, the description of CONTRIBUTING.md was modified.
**Issue: the issue # it fixes (if applicable)**
https://github.com/hwchase17/langchain/issues/6413
**Dependencies: any dependencies required for this change**
`nbdoc` was missing in group docs so it was added. I installed it with
the `poetry add --group docs nbdoc` command. I am concerned if any
modifications are needed to poetry.lock. I would greatly appreciate it
if you could pay close attention to this file during the review.
**Tag maintainer**
- General / Misc / if you don't know who to tag: @baskaryan
If this PR needs any additional changes, I'll be happy to make them!
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Continuing with Tolkien inspired series of langchain tools. I bring to
you:
**The Fellowship of the Vectors**, AKA EmbeddingsClusteringFilter.
This document filter uses embeddings to group vectors together into
clusters, then allows you to pick an arbitrary number of documents
vector based on proximity to the cluster centers. That's a
representative sample of the cluster.
The original idea is from [Greg Kamradt](https://github.com/gkamradt)
from this video (Level4):
https://www.youtube.com/watch?v=qaPMdcCqtWk&t=365s
I added few tricks to make it a bit more versatile, so you can
parametrize what to do with duplicate documents in case of cluster
overlap: replace the duplicates with the next closest document or remove
it. This allow you to use it as an special kind of redundant filter too.
Additionally you can choose 2 diff orders: grouped by cluster or
respecting the original retriever scores.
In my use case I was using the docs grouped by cluster to run refine
chains per cluster to generate summarization over a large corpus of
documents.
Let me know if you want to change anything!
@rlancemartin, @eyurtsev, @hwchase17,
---------
Co-authored-by: rlm <pexpresss31@gmail.com>
Distance-based vector database retrieval embeds (represents) queries in
high-dimensional space and finds similar embedded documents based on
"distance". But, retrieval may produce difference results with subtle
changes in query wording or if the embeddings do not capture the
semantics of the data well. Prompt engineering / tuning is sometimes
done to manually address these problems, but can be tedious.
The `MultiQueryRetriever` automates the process of prompt tuning by
using an LLM to generate multiple queries from different perspectives
for a given user input query. For each query, it retrieves a set of
relevant documents and takes the unique union across all queries to get
a larger set of potentially relevant documents. By generating multiple
perspectives on the same question, the `MultiQueryRetriever` might be
able to overcome some of the limitations of the distance-based retrieval
and get a richer set of results.
---------
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Fixed several inconsistencies:
- file names and notebook titles should be similar otherwise ToC on the
[retrievers
page](https://python.langchain.com/en/latest/modules/indexes/retrievers.html)
and on the left ToC tab are different. For example, now, `Self-querying
with Chroma` is not correctly alphabetically sorted because its file
named `chroma_self_query.ipynb`
- `Stringing compressors and document transformers...` demoted from `#`
to `##`. Otherwise, it appears in Toc.
- several formatting problems
#### Who can review?
@hwchase17
@dev2049
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
## DocArray as a Retriever
[DocArray](https://github.com/docarray/docarray) is an open-source tool
for managing your multi-modal data. It offers flexibility to store and
search through your data using various document index backends. This PR
introduces `DocArrayRetriever` - which works with any available backend
and serves as a retriever for Langchain apps.
Also, I added 2 notebooks:
DocArray Backends - intro to all 5 currently supported backends, how to
initialize, index, and use them as a retriever
DocArray Usage - showcasing what additional search parameters you can
pass to create versatile retrievers
Example:
```python
from docarray.index import InMemoryExactNNIndex
from docarray import BaseDoc, DocList
from docarray.typing import NdArray
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.retrievers import DocArrayRetriever
# define document schema
class MyDoc(BaseDoc):
description: str
description_embedding: NdArray[1536]
embeddings = OpenAIEmbeddings()
# create documents
descriptions = ["description 1", "description 2"]
desc_embeddings = embeddings.embed_documents(texts=descriptions)
docs = DocList[MyDoc](
[
MyDoc(description=desc, description_embedding=embedding)
for desc, embedding in zip(descriptions, desc_embeddings)
]
)
# initialize document index with data
db = InMemoryExactNNIndex[MyDoc](docs)
# create a retriever
retriever = DocArrayRetriever(
index=db,
embeddings=embeddings,
search_field="description_embedding",
content_field="description",
)
# find the relevant document
doc = retriever.get_relevant_documents("action movies")
print(doc)
```
#### Who can review?
@dev2049
---------
Signed-off-by: jupyterjazz <saba.sturua@jina.ai>
