elastic/elasticsearch-labs

Notebooks & Example Apps for Search & AI Applications with Elasticsearch

Elasticsearch Examples & Apps

Visit Search Labs for the latest articles and tutorials on using Elasticsearch for search and AI/ML-powered search experiences

This repo contains executable Python notebooks, sample apps, and resources for testing out the Elastic platform:

• Learn how to use Elasticsearch as a vector database to store embeddings, power hybrid and semantic search experiences.
• Build use cases such as retrieval augmented generation (RAG), summarization, and question answering (QA).
• Test Elastic's leading-edge, out-of-the-box capabilities like the Elastic Learned Sparse Encoder and reciprocal rank fusion (RRF), which produce best-in-class results without training or tuning.
• Integrate with projects like OpenAI, Hugging Face, and LangChain, and use Elasticsearch as the backbone of your LLM-powered applications.

Elastic enables all modern search experiences powered by AI/ML.

• Bookmark or subscribe to Elasticsearch Labs on Github
• Read our latest articles at elastic.co/search-labs

Apps

• Chatbot RAG App
• Internal Knowledge Search
• Relevance Workbench

Python notebooks 📒

The notebooks folder contains a range of executable Python notebooks, so you can test these features out for yourself. Colab provides an easy-to-use Python virtual environment in the browser.

Generative AI

• question-answering.ipynb
• chatbot.ipynb

Playground RAG Notebooks

Try out Playground in Kibana with the following notebooks:

• OpenAI Example
• Anthropic Claude 3 Example

LangChain

• question-answering.ipynb
• langchain-self-query-retriever.ipynb
• Question Answering with Self Query Retriever
• BM25 and Self-querying retriever with elasticsearch and LangChain
• langchain-vector-store.ipynb
• langchain-vector-store-using-elser.ipynb
• langchain-using-own-model.ipynb

Document Chunking

• Document Chunking with Ingest Pipelines
• Document Chunking with LangChain Splitters
• Calculating tokens for Semantic Search (ELSER and E5)
• Fetch surrounding chucks

Search

• 00-quick-start.ipynb
• 01-keyword-querying-filtering.ipynb
• 02-hybrid-search.ipynb
• 03-ELSER.ipynb
• 04-multilingual.ipynb
• 05-query-rules.ipynb
• 06-synonyms-api.ipynb
• 07-inference.ipynb
• 08-learning-to-rank.ipynb
• 09-semantic-text.ipynb

Integrations

• loading-model-from-hugging-face.ipynb
• openai-semantic-search-RAG.ipynb
• amazon-bedrock-langchain-qa-example.ipynb
• Semantic Search using the Inference API with the Cohere Service

Model Upgrades

• upgrading-index-to-use-elser.ipynb

Contributing 🎁

See contributing guidelines.

Support 🛟

The Search team at Elastic maintains this repository and is happy to help.

Official Support Services

If you have an Elastic subscription, you are entitled to Support services for your Elasticsearch deployment. See our welcome page for working with our support team. These services do not apply to the sample application code contained in this repository.

Discuss Forum

Try posting your question to the Elastic discuss forums and tag it with #esre-elasticsearch-relevance-engine

Elastic Slack

You can also find us in the #search-esre-relevance-engine channel of the Elastic Community Slack

License ⚖️

This software is licensed under the Apache License, version 2 ("ALv2").

elasticsearch-labs's Issues

CI notebook failures possibly due to new Langchain integration

There are two tests from our CI that are now failing, presumably after they started using the updated langchain integration.

The errors:

Running notebooks/document-chunking/with-langchain-splitters.ipynb... Failed
...
BadRequestError: BadRequestError(400, 'x_content_parse_exception', "[1:117] [text_embedding] model_id doesn't support values of type: VALUE_NULL")

Running notebooks/langchain/langchain-using-own-model.ipynb... Failed
...
AttributeError: 'ElasticsearchStore' object has no attribute 'client'

@maxjakob any ideas?

Complete run with these failures: https://github.com/elastic/elasticsearch-labs/actions/runs/9113170450

chatbot-rag-app does not work for self-signed (local development)

For self-signed environment, the code does not work. we need to set the certificate in order to establish a connection with Elasticsearch. There are two places to change:

1. index_data.py

ELASTCSEARCH_CERT_PATH = "/usr/share/certs/http_ca.crt"

if ELASTICSEARCH_URL:
    elasticsearch_client = Elasticsearch(
        hosts=[ELASTICSEARCH_URL], 
        ca_certs = ELASTCSEARCH_CERT_PATH, 
        verify_certs = True
    )

This adds the certificate info into the connection.

2. elasticsearch-client.py

ELASTCSEARCH_CERT_PATH = "/usr/share/certs/http_ca.crt"


if ELASTICSEARCH_URL:
    elasticsearch_client = Elasticsearch(
        hosts=[ELASTICSEARCH_URL], 
        ca_certs = ELASTCSEARCH_CERT_PATH, 
        verify_certs = True
    )

We can index the data using the command like:

docker run \
  --volume="$PWD/http_ca.crt:/usr/share/certs/http_ca.crt:ro" \
  --rm --env-file .env chatbot-rag-app flask create-index

and we can run the frondend like:

docker run \
  --volume="$PWD/http_ca.crt:/usr/share/certs/http_ca.crt:ro" \
  --rm -p 4000:4000 --env-file .env -d chatbot-rag-app

Confusion about the use of self-hosted Elasticsearch with ELSER in RAG App

Hello,

the documentation for the RAG app mentions that it is possible to use a self-hosted Elasticsearch instance. However, the use of ELSER seems to be necessary for the application to work.

Can you please clarify if and how a self-hosted Elasticsearch instance can be used without ELSER? In particular, it would be helpful to know whether a cloud instance or special ELSER configurations are required to fully utilise the RAG app.

Thank you very much for your support and clarification!

Blog Proposal - Ingest data from Snowflake to Elasticsearch

Ingest data from Snowflake to Elasticsearch

Blog Content

Wrong dims in the model configuration for embed-english-v3.0 in Cohere notebook

I tried to run the cohere notebook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/cohere/cohere-elasticsearch.ipynb. In the code, it configures:

client.indices.create(
    index="cohere-embeddings",
    settings={"index": {"default_pipeline": "cohere_embeddings"}},
    mappings={
        "properties": {
            "text_embedding": {
                "type": "dense_vector",
                "dims": 1024,
                "element_type": "byte",
            },
            "text": {"type": "text"},
            "id": {"type": "integer"},
            "title": {"type": "text"},
        }
    },
)

When I run the code for the model, and finally index data into Elasticsearch. It complains:

{
  "error": {
    "root_cause": [
      {
        "type": "document_parsing_exception",
        "reason": "[1:2078] failed to parse: The [dense_vector] field [text_embedding] in doc [document with id '4983'] has a different number of dimensions [384] than defined in the mapping [1024]"
      }
    ],
    "type": "document_parsing_exception",
    "reason": "[1:2078] failed to parse: The [dense_vector] field [text_embedding] in doc [document with id '4983'] has a different number of dimensions [384] than defined in the mapping [1024]",
    "caused_by": {
      "type": "illegal_argument_exception",
      "reason": "The [dense_vector] field [text_embedding] in doc [document with id '4983'] has a different number of dimensions [384] than defined in the mapping [1024]"
    }
  },
  "status": 400
}

From the error returned, the dims for the model is 384 instead of 1024. I think the documentation at https://www.elastic.co/guide/en/elasticsearch/reference/current/cohere-es.html is also wrong. After I reconfigure it to 384, and it works.

inference id is wrongly written when creating the model

I just ran the notebook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/cohere/cohere-elasticsearch.ipynb. It was not successful. I found that the inference id was not so correct.

client.inference.put_model(
    task_type="text_embedding",
    inference_id="cohere_embeddings1",
    body={
        "service": "cohere",
        "service_settings": {
            "api_key": COHERE_API_KEY,
            "model_id": "embed-english-v3.0",
            "embedding_type": "byte",
        },
    },
)

It defines the inference_id as "cohere_embeddings1", however, it is never referred in other part of the codes. I found our tutorial at https://www.elastic.co/guide/en/elasticsearch/reference/current/cohere-es.html. The correct code is:

COHERE_API_KEY = "cohere_api_key"

client.inference.put_model(
    task_type="text_embedding",
    inference_id="cohere_embeddings",
    body={
        "service": "cohere",
        "service_settings": {
            "api_key": COHERE_API_KEY,
            "model_id": "embed-english-v3.0",
            "embedding_type": "byte"
        }
    },
)

Create a pointer to code for dev genai workshops

in the spirit of opennes

What is the reason for "elasticsearch.BadRequestError: BadRequestError(400, 'x_content_parse_exception', '[1:137] unknown field [xlm_roberta]')"?

1. What do I want?

   I want to install an additional NLP model to Elasticsearch. The model is called multilingual-e5-base.

2. What did I do?

   Therefore, I followed the steps in Elasticsearch
   documentation.

3. What I expected?

   The multilingual-e5-base is successfully installed.

4. What I tried?

Here is the relevant section in my Python Notebook for the installation:

!source ../.env && \
eland_import_hub_model \
    --url ${REMOTE_HOST} \
    --es-username ${USER} \
    --es-password ${API_KEY} \
    --hub-model-id intfloat/multilingual-e5-base \
    --es-model-id multilingual-e5-base \
    --task-type text_embedding \
    --start

Error message:

raise HTTP_EXCEPTIONS.get(meta.status, ApiError)(
elasticsearch.BadRequestError: BadRequestError(400, 'x_content_parse_exception', '[1:137] unknown field [xlm_roberta]')

Note: I posted the same question on stackoverflow link

inference search notebook error with model_id vs inference_id after 8.13 upgrade

Issue from @derickson

My environment

• colab
• elasticsearch python library 8.13
• elasticsearch 8.13.0 in ESS
• Notebook: https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/search/07-inference.ipynb

The Semantic search step uses the follwing as it's knn query

{
        "field": "plot_embedding",
        "query_vector_builder": {
            "text_embedding": {
                "inference_id": "my_openai_embedding_model",
                "model_text": "Fighting movie",
            }
        },
        "k": 10,
        "num_candidates": 100,
    }

which does not work in 8.13

changing to "model_id" instead of "inference_id" solves the issue. This notebook was written for 8.12

the note at the end is innacurate

"NOTE: If you use Elasticsearch 8.12, you must change inference_id in the snippet above to model_id."

First error reason: Could not find trained model [.elser_model_1] when trying workplace-search

Can't execute index-data.py

step to reproduce

1. clone repo
2. cd workplace-search
3. install requirements.txt
4. Cange ES Client connection to http://localhost:9200
5. run index-data.py

Loaded 1 documents
Split 1 documents into 104 chunks
Creating Elasticsearch sparse vector store in Elastic Cloud: None
Error adding texts: 104 document(s) failed to index.
First error reason: Could not find trained model [.elser_model_1]
Traceback (most recent call last):
File "/home/aleksei/code/elasticsearch-labs/example-apps/workplace-search/./data/index-data.py", line 52, in
ElasticsearchStore.from_documents(
File "/home/aleksei/.local/lib/python3.10/site-packages/langchain/vectorstores/elasticsearch.py", line 1125, in from_documents
elasticsearchStore.add_documents(documents, bulk_kwargs=bulk_kwargs)
File "/home/aleksei/.local/lib/python3.10/site-packages/langchain/schema/vectorstore.py", line 122, in add_documents
return self.add_texts(texts, metadatas, **kwargs)
File "/home/aleksei/.local/lib/python3.10/site-packages/langchain/vectorstores/elasticsearch.py", line 971, in add_texts
raise e
File "/home/aleksei/.local/lib/python3.10/site-packages/langchain/vectorstores/elasticsearch.py", line 954, in add_texts
success, failed = bulk(
File "/home/aleksei/.local/lib/python3.10/site-packages/elasticsearch/helpers/actions.py", line 521, in bulk
for ok, item in streaming_bulk(
File "/home/aleksei/.local/lib/python3.10/site-packages/elasticsearch/helpers/actions.py", line 436, in streaming_bulk
for data, (ok, info) in zip(
File "/home/aleksei/.local/lib/python3.10/site-packages/elasticsearch/helpers/actions.py", line 355, in _process_bulk_chunk
yield from gen
File "/home/aleksei/.local/lib/python3.10/site-packages/elasticsearch/helpers/actions.py", line 274, in _process_bulk_chunk_success
raise BulkIndexError(f"{len(errors)} document(s) failed to index.", errors)
elasticsearch.helpers.BulkIndexError: 104 document(s) failed to index.

Wrong fields used in the filter and aggregation

For the notebook at https://github.com/elastic/elasticsearch-labs/blob/main/supporting-blog-content/vector-search-implementation-guide-api/vector_search_implementation_guide_api.ipynb, "my_metadata" is a text field, and it should not be used in the term filter, and it is the same for the aggregation. The correct one should be "my_metadata.keyword".

Model id is needed to get Gemma LLM locally working on private data

This is the code https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/gemma/rag-gemma-huggingface-elastic.ipynb.

There are two improvements here:

1. For package installation, pytorch is needed. We need to add "torch" module into the list
2. Since .elser_model_2 has been released, and it is the GA release for use. The following code:

es = ElasticsearchStore.from_documents(
    docs,
    es_cloud_id=ELASTIC_CLOUD_ID,
    es_api_key=ELASTIC_API_KEY,
    index_name=elastic_index_name,
    strategy=ElasticsearchStore.SparseVectorRetrievalStrategy(),
)

works only for .elser_model_1. In order to get it working for .elser_model_2, we have to pass the model id into the strategy:

es = ElasticsearchStore.from_documents(
    docs,
    es_cloud_id=ELASTIC_CLOUD_ID,
    es_api_key=ELASTIC_API_KEY,
    index_name=elastic_index_name,
    strategy=ElasticsearchStore.SparseVectorRetrievalStrategy(model_id=".elser_model_2"),
)

If we run the code without the model_id set and we have the .elser_model_2 deployed in the cluster, we might run into a problem reported here at langchain-ai/langchain#18917 even we later on correctly set the model_id in the function call.

AUTO_BUCKET is not working in the ES|QL notebook

For the notebook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/esql/esql-getting-started.ipynb, the following code is not working:

esql_query = """
FROM sample_data
| KEEP @timestamp
| EVAL bucket = AUTO_BUCKET(@timestamp, 24, "2023-10-23T00:00:00Z", "2023-10-23T23:59:59Z")
"""
response = client.esql.query(query=esql_query)
format_response(response)

The error message is shown as:

BadRequestError: BadRequestError(400, 'verification_exception', 'Found 1 problem\nline 4:17: Unknown function [AUTO_BUCKET], did you mean [bucket]?')

![image](https://github.com/elastic/elasticsearch-labs/assets/8455957/be2018cb-a9dd-46fd-a281-30b0c9de00c9)

mappings link does not exist for the internal-knowledge-search example

As described in the readme file, the following link does not exist at all: https://github.com/elastic/elasticsearch-labs/blob/main/example-apps/internal-knowledge-search/src/config/documentsToSearchResultMappings.json

In addition, I read the blog at https://www.elastic.co/search-labs/blog/articles/dls-internal-knowledge-search. I I am not wrong, the DIS should be DLS, right?

Ragas and Elasticsearch

I've built a Streamlit based RAG pipeline workbench that lets developers experiment using different prompt techniques, LLMs, model temperatures, and document chunking approaches. The results are objectively measured and compared using Ragas and exported to Elastic for visualisation using a dashboard that is part of the repo and can be imported by the user. I've also included LLM interaction logging as part of the app in order to help developers keep track of costs and compare the differences between the models they're using. Ultimately I would also like to write a blog around this and I have taken assignment of the search labs ticket of the same title.

Changing examples to use hosts instead of Cloud ID

Most of the examples in this repo use Could ID instead of hosts, while chatting with @miguelgrinberg and @joemcelroy we noticed that we should probably update them.

What is in most examples is something like this:

ELASTIC_CLOUD_ID = getpass("Elastic Cloud ID: ")

# https://www.elastic.co/search-labs/tutorials/install-elasticsearch/elastic-cloud#creating-an-api-key
ELASTIC_API_KEY = getpass("Elastic Api Key: ")

A version of this could be:

# Create an Elasticsearch client instance
client = Elasticsearch(
    getpass("Host: "),  # Securely get the host for the Elasticsearch cluster
    api_key=getpass("API Key: "),  # Securely get the API key for authentication
)

[BLOG proposal]

Blog content

Reviewed by Aditya Tripathi, Carly Richmond, and Philipp Krenn.

wrong field is used in the filtering

Let me repost my previous report here. https://github.com/elastic/search-labs-elastic-co/issues/81

Unnecessary if statement in search tutorial

An unnecessary if statement exists in the following page.

https://www.elastic.co/search-labs/tutorials/search-tutorial/full-text-search/filters

elastic_transport.ConnectionTimeout: Connection timed out

when i run :
python manage.py search_index --rebuild
have always this error :
anaconda3/envs/DJ_ElasticSearch/lib/python3.6/site-packages/elastic_transport/_node/_http_urllib3.py", line 202, in perform_request
raise err from None
elastic_transport.ConnectionTimeout: Connection timed out

How can i solve that?

k-Nearest Neighbor (kNN) Search, The result will be duplicate data, is that correct？

Excuse me, I did all the tests according to the documents, but I found a lot of duplicate data appeared in the search results. Is this reasonable to expect

[chatbot-rag-app] jq failed in Windows

I cannot install packages

1. clone repo to windows 11
2. cd example-apps/chatbot-rag-app
3. conda create-n test_env python=3.11
4. install requirements.txt
5. Failed to build jq

Building wheels for collected packages: jq
  Building wheel for jq (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Building wheel for jq (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [5 lines of output]
      running bdist_wheel
      running build
      running build_ext
       Executing: ./configure CFLAGS=-fPIC --prefix=C:\Users\wonseop\AppData\Local\Temp\pip-install-tzsfm24e\jq_79650ef0c8804b599c4ca739dc810450\_deps\build\onig-install-6.9.4
      error: [WinError 2] The system cannot find the file specified
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for jq
Failed to build jq
ERROR: Could not build wheels for jq, which is required to install pyproject.toml-based projects

Someone please help me~

Guides and examples for 2nd phase retrieval

Providing examples on how to re-rank using:

• bi-encoders and cross-encoders
• Learn to rank via feature extractors
• boosting on attributes like dates, popularity

Cohere service is not available in the 8.12 release

For the notebook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/cohere/inference-cohere.ipynb. In the following code:

API_KEY = getpass("Enter Cohere API key:  ")

client.inference.put_model(
    task_type="text_embedding",
    model_id="cohere_embeddings",
    body={
        "service": "cohere",
        "service_settings": {
            "api_key": API_KEY,
            "model_id": "embed-english-v3.0",
            "embedding_type": "int8",
        },
        "task_settings": {},
    },
)

Cohere does not work for Elastic Stack normal release. It returns the following error:

BadRequestError: BadRequestError(400, 'status_exception', 'Unknown service [cohere]')

For Elastic Stack 8.12 installation.

According to our documentation at https://www.elastic.co/guide/en/elasticsearch/reference/current/put-inference-api.html. Currently, only the following services are supported:

ELSER
OpenAI
Hugging Face

According to the reply from our Slack discussion, "Cohere" is only currently supported on Serverless.

It would be good to point out this to the developers.

nbtest fails on document chunking notebook

I'm getting the following error when running the document chunking notebook under nbtest:

ConflictError: ConflictError(409, 'status_exception', 'Trained model deployment [sentence-transformers__all-minilm-l6-v2] is not allocated to any nodes')

This is on a deployment that has never been used with this particular model.

Fetch surrounding chunks notebook

I have built a notebook to demonstrate how to fetch surrounding chunks within Elasticsearch. This is not the only way to do it, but it is definitely a valid approach. I am interested in contributing this notebook to our examples

Notebook: Google Colab

The example uses text from a Harry Potter book, splitting it by chapter and then into chunks. Each chunk is a nested passage containing the text, dense, and sparse representations of the chunk.

When searching, the demo will fetch the matching passage chunk along with surrounding chunks. If the chunk is the first chunk in the chapter, it will fetch n, n+1, and n+2. If it is the last chunk, it will fetch n, n-1, and n-2. Otherwise it will fetch n-1, n, n+1.

Jupyter link in “Geospatial search" is broken

The "Jupyter notebook" links in the blog "Journey through the city: Geospatial search made simple with LLM and Elasticsearch" are all broken. I found that their values are https://colab.research.google.com/github/elastic/elasticsearch-labs/blob/main/geospatial-llm/09-geospatial-search.ipynb. However, the actual location for the notebook is https://github.com/elastic/elasticsearch-labs/blob/main/supporting-blog-content/geospatial-llm/09-geospatial-search.ipynb

Wrong file name is spelled in the search-ui tutorial

https://www.elastic.co/search-labs/tutorials/search-tutorial/full-text-search/filters. On the page,

Enrich error for "Getting started with ES|QL" notebook

For EQSL getting started notboook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/esql/esql-getting-started.ipynb. The problem lies here in the following piece of code:

# Define the enrich policy
policy = {
    "match": {
        "indices": "clientips",
        "match_field": "client_ip",
        "enrich_fields": ["env"],
    }
}

# Put the enrich policy
client.enrich.put_policy(name="clientip_policy", body=policy)

# Execute the enrich policy without waiting for completion
client.enrich.execute_policy(name="clientip_policy", wait_for_completion=False)

The policy is executed in asynchronous way, and run the following code immediately, it results error:

esql_query = """
FROM sample_data
| KEEP @timestamp, client_ip, event_duration
| EVAL client_ip = TO_STRING(client_ip)
| ENRICH clientip_policy ON client_ip WITH env
"""
response = client.esql.query(query=esql_query)
format_response(response)

I think the enrich index is not ready. To correct the problem, we need to make the call to synchronous:

client.enrich.execute_policy(name="clientip_policy", wait_for_completion=True)

BadRequestError(400, 'parsing_exception', 'Unknown key for a START_OBJECT in [knn].')

Something is wrong with the Hybrid Search query:

results = es.search(
        query={
            "bool": {
                **search_query,
                **filters
            }
        },
        knn={
            "field": "embedding",
            "query_vector": es.get_embedding(parsed_query),
            "k": 10,
           "num_candidates": 50,
        },
        rank={"rrf": {}},
        aggs={
            "category-agg": {
                "terms": {
                    "field": "category.keyword",
                }
            },
            "year-agg": {
                "date_histogram": {
                    "field": "updated_at",
                    "calendar_interval": "year",
                    "format": "yyyy",
                },
            },
        },
        size=5,
        from_=from_,
    )

I get this error:

BadRequestError(400, 'parsing_exception', 'Unknown key for a START_OBJECT in [knn].')

If I delete the knn Param I get this error:

BadRequestError(400, 'parsing_exception', 'Unknown key for a START_OBJECT in [rank].')

If I also delete also rank Param it works.

Can someone help me please?

index alias instead of index

Can I use index aliases instead of indexes?

Search text is not used in the embedding generation

at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/gemini/vector-search-gemini-elastic.ipynb

q = "How's weather in India?"

embedding = genai.embed_content(model=model,
                                content=sample_text,
                                task_type="retrieval_query")

resp = es.search(
  index = elastic_index_name,
  knn={
      "field": "text_embedding",
      "query_vector":  embedding['embedding'],
      "k": 10,
      "num_candidates": 100
    }
)

The variable "q" is never used in the embedding generation in the search process. The correct one should be:

embedding = genai.embed_content(model=model,
                                content=q,
                                task_type="retrieval_query")

Notebook for RAG with Llama 3 open-source and Elastic

Notebook for RAG with Llama 3 open-source and Elastic using Llamaindex and Langchain.

Include a note on setting up virtual environments in readme or notebooks

This is a relatively minority "issue" but thinking back to when I started learning python, virtual environments were still "new", it was not covered in most material. Over time I'd built up various quirks that venv would have avoided

Users who are interested in using the notebooks and aren't familiar deeply familiar with python may be able to follow along with the instructions but be unaware that activating a python virtual environment is a good approach to handling dependency issues that can arise when different versions are needed.

Including this somewhere could save a lot of headache and troubleshooting to fix whatever issues arise, while yielding a better user experience.

Potential Addition: Vector and Hybrid search examples in Go

Is there an interest for adding non-Python examples to the labs? Alongside @Anaethelion we're written Go examples to find Gophers in Elasticsearch.

Currently, it's in Markdown format rather than a Python notebook with markdown elements, so it would be good to understand if the search labs blog requires posts in a notebook or can support markdown.

Aware there's been discussion by @xeraa and @joemcelroy so let us know and we can raise a PR to add this contribution. Otherwise, it can live on the Elastic blog instead.

Merge and update corresponding blog post

We'd like to migrate https://www.elastic.co/blog/searching-by-music-leveraging-vector-search-audio-information-retrieval to Search Labs

Which is on the migration list : https://github.com/elastic/search-labs-elastic-co/issues/93 but it already has supporting code in salgado/music-search#1 which I've asked them to open a PR to merge into Search labs. When we migrate the blog over, please update the section of the blog to reference the merged code in Search Labs vs the original GH project.

index name is not consistent with the code

Let me repost my previous issue here. https://github.com/elastic/search-labs-elastic-co/issues/80

Get index mapping and Get reference document are inconsistent with noteboook in "OpenAI function calling with Elasticsearch"

I read through the blog at https://www.elastic.co/search-labs/blog/function-calling-with-elastic, and I found two functions defined in the notebook and blog are insconsistent:

• def get_index_mapping():
• get_ref_document()

The ones defined in the blogs are not correct:

def get_index_mapping():

    url = f"""{ES_ENDPOINT}/{ES_INDEX}/_mappings"""

    headers = {
        "Content-type": "application/json",
        "Authorization": f"""ApiKey {ES_API_KEY}""",
    }

def get_index_mapping():

    url = f"""{ES_ENDPOINT}/{ES_INDEX}/_mappings"""

    headers = {
        "Content-type": "application/json",
        "Authorization": f"""ApiKey {ES_API_KEY}""",
    }

    resp = requests.request("GET", url, headers=headers)
    resp = json.loads(resp.text)
    mapping = json.dumps(resp, indent=4)

    return mapping

The correct ones are shown in the attached pictures,

The pipeline use in the demo of 'plagiarism-detection-with-elasticsearch'

I believe the pipeline used for reindex should be plagiarism-checker-pipeline rather than plagiarism-pipeline

Cusstom ranking method

Hello everyone. I wanted to ask if there's any way that I can implement a custom algorithm for the search when combining KNN and Text Search.

https://raw.githubusercontent.com/elastic/elasticsearch-labs/main/notebooks/hugging-face/blogs.json does not exist

I reported the issue, but it might be in the wrong place. Please take a look at https://github.com/elastic/search-labs-elastic-co/issues/79

Thanks

Cloud ID is not properly formatted

  File "/.../elastic/chatbot-rag-app/.venv/lib/python3.11/site-packages/elastic_transport/client_utils.py", line 125, in parse_cloud_id
    raise ValueError("Cloud ID is not properly formatted") from None
ValueError: Cloud ID is not properly formatted

Using the Cloud ID found on my https://cloud.elastic.co/deployments/ page.

It seems that the client_utils.py may be expecting to find a '$' separator in part of the decoded id string.

This prevents ingesting data while attempting to connect to a cloud Elasticsearch instance.

Positional arguments do not work for TransformerModel and wrong pipeline name used in the code for plagiarism demo

For the example app at https://github.com/elastic/elasticsearch-labs/blob/main/supporting-blog-content/plagiarism-detection-with-elasticsearch/plagiarism_detection_es.ipynb. I find there are following issues:

1. According to code at https://github.com/elastic/eland/blob/main/eland/ml/pytorch/transformers.py,
   class TransformerModel: def __init__( self, *, model_id: str, task_type: str, es_version: Optional[Tuple[int, int, int]] = None, quantize: bool = False, access_token: Optional[str] = None, ):
   It needs to have keyword arguments for the constructor. I am not sure the demo uses which eland version. I use the latest version:
   $ pip3 list | grep eland eland 8.11.1

The call in the code:
tm = TransformerModel(hf_model_id, "text_embedding")

does not work. The working one is:

tm = TransformerModel(model_id=hf_model_id, task_type="text_embedding")

This applies to the code as well:
tm = TransformerModel(hf_model_id, "text_classification")

2. pipeline name is spelled wrongly in the code:

client.reindex(wait_for_completion=True, source={ "index": "plagiarism-docs" }, dest= { "index": "plagiarism-checker", "pipeline": "plagiarism-pipeline" } )

In the code, "plagiarism-pipeline" pipeline is never created. It should be "plagiarism-checker-pipeline".

3. The synchronous call for the reindex seems not working at all.

client.reindex(wait_for_completion=True, source={ "index": "plagiarism-docs" }, dest= { "index": "plagiarism-checker", "pipeline": "plagiarism-pipeline" } )

It always returns "Connection time out" error. I have to change it to False to make it work.

GOOGLE_APPLICATION_CREDENTIALS json-service-account-file cannot be found

The application cannot find the json-service-account-file per the given instructions and thus fails on a permission error.
The credential file needs to be copied into one of the directories that is copied into the container by the Dockerfile so it can be found by the app.
My suggestion is to copy the file into the data directory and then set the path accordingly

My suggestion is the following for the Vertex AI Section

Vertex AI

To use Vertex AI you need to set the following environment variables. More info here.

copy your json-service-account-file into the data directory and set the path accordingly

export LLM_TYPE=vertex
export VERTEX_PROJECT_ID=<gcp-project-id>
export VERTEX_REGION=<gcp-region> # Default is us-central1
export GOOGLE_APPLICATION_CREDENTIALS=/data/<json-service-account-file>

"'NoneType' object is not iterable" in the Cohere notebook

In the notebook at https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/integrations/cohere/cohere-elasticsearch.ipynb, in the last part of the code:

response = co.chat(message=query, documents=ranked_documents, model="command-r-plus")

source_documents = []
for citation in response.citations:
   for document_id in citation.document_ids:
       if document_id not in source_documents:
           source_documents.append(document_id)

print(f"Query: {query}")
print(f"Response: {response.text}")

It returns the following error:

`---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[15], line 4
      1 response = co.chat(message=query, documents=ranked_documents, model="command-r-plus")
      3 source_documents = []
----> 4 for citation in response.citations:
      5    for document_id in citation.document_ids:
      6        if document_id not in source_documents:

TypeError: 'NoneType' object is not iterable`

According to the documentation at https://docs.cohere.com/docs/chat-api. There is no such a field "citations" in the Chat API.

ModuleNotFoundError: No module named 'dotenv' when trying workplace-search

Can't execute index-data.py

step to reproduce

1. clone repo
2. cd workplace-search
3. create virtual env
4. install requirements.txt
5. run python3 data/index-data.py

(.venv) aleksei@EPRSBELL001B:~/code/elasticsearch-labs/example-apps/workplace-search$ python3 data/index-data.py
Traceback (most recent call last):
File "/home/aleksei/code/elasticsearch-labs/example-apps/workplace-search/data/index-data.py", line 3, in
from dotenv import load_dotenv
ModuleNotFoundError: No module named 'dotenv'

chatbot-example does not get the right result as shown in the notebook.

https://github.com/elastic/elasticsearch-labs/blob/main/notebooks/langchain/self-query-retriever-examples/chatbot-example.ipynb

I followed the exact steps as described in the notebook. In the last step, I used the same question, but I got the result "I do not know".

However, if I changed the question to "What movies are about dreams and it was released after the year 2009 but before 2011?". Then I will get the exact same result as in the notebook.

Customizing the UI?

Is there a way to customize the UI?

redundant texts in the article

https://www.elastic.co/search-labs/tutorials/search-tutorial/vector-search/nearest-neighbor-search. Here "It is" seems to be redundant.

Unable to run chatbot-rag-app using Azure OpenAI model

I attempted to run the chatbot-rag-app using an Azure OpenAI model, and found that when I entered a question in the prompt box in the GUI, the service would never produce a response.

I was able to successfully use the docker run --rm --env-file .env chatbot-rag-app flask create-index to index the sample documents to my Elasticsearch cluster, so all ES configuration elements are good.

I assumed I was passing a bad var value in the .env file, so I:

• Double-checked my OPENAI_API_KEY
• Attempted multiple permutations of OPENAI_VERSION - the model resource overview page in the Azure portal shows a model version of 0613, but the annotation in the project docs indicates that the version should appear like 2023-05-15, so I tried using both 0613 and 2023-06-13, but got the same results with both
• Double checked the OPENAI_BASE_URL var, using the Endpoint value shown on the Keys & Endpoints page in the azure portal, both with and without the tailing /

Attached are a redacted copy of my .env file (see above for the additional variations I tried), as well as the docker logs from a container instance where this behavior occurred

redacted_dot_env_file.txt
azureopenai_error_logs.txt