Chroma vector store setup in LangChain - Step-by-Step Execution

Concept Flow - Chroma vector store setup

Import Chroma and dependencies

↓

Create embeddings model

↓

Initialize Chroma vector store

↓

Add documents to vector store

↓

Query vector store for similar docs

↓

Use results

This flow shows how to set up Chroma vector store: import, create embeddings, initialize store, add docs, then query.

Execution Sample

LangChain

from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings

embeddings = OpenAIEmbeddings()
vectorstore = Chroma(embedding_function=embeddings)
vectorstore.add_texts(["Hello world", "Langchain rocks"])

This code imports Chroma and embeddings, creates embeddings, initializes Chroma vector store, and adds two texts.

Execution Table

Step	Action	Evaluation	Result
1	Import Chroma and OpenAIEmbeddings	Modules loaded	Ready to use classes
2	Create embeddings instance	embeddings = OpenAIEmbeddings()	embeddings object created
3	Initialize Chroma vector store	vectorstore = Chroma(embedding_function=embeddings)	Empty vector store ready
4	Add texts to vector store	vectorstore.add_texts(["Hello world", "Langchain rocks"])	Texts embedded and stored
5	Query vector store (not shown here)	vectorstore.similarity_search("Hello")	Returns similar documents

💡 Setup complete, vector store ready for queries

Variable Tracker

Variable	Start	After Step 2	After Step 3	After Step 4	Final
embeddings	None	OpenAIEmbeddings instance	OpenAIEmbeddings instance	OpenAIEmbeddings instance	OpenAIEmbeddings instance
vectorstore	None	None	Empty Chroma vector store	Chroma vector store with 2 texts	Chroma vector store with 2 texts

Key Moments - 2 Insights

Why do we need to create an embeddings instance before initializing Chroma?

Chroma needs an embedding function to convert texts into vectors. The embeddings instance provides this function, as shown in step 2 and used in step 3.

What happens when we call add_texts on the vector store?

The texts are converted into vectors using the embeddings function and stored inside Chroma. This is shown in step 4 where texts become embedded and stored.

Visual Quiz - 3 Questions

Look at the execution table, what is the state of 'vectorstore' after step 3?

A Chroma vector store with 2 texts stored

B Empty Chroma vector store ready to add texts

C None, vectorstore not created yet

D An embeddings instance

💡 Hint

Check the 'Result' column for step 3 in the execution table

At which step are the texts converted into vectors and stored?

A Step 2

B Step 3

C Step 4

D Step 5

💡 Hint

Look for 'Texts embedded and stored' in the execution table

If we skip creating the embeddings instance, what will happen when initializing Chroma?

A Error because embedding_function is missing

B Chroma initializes with default embeddings

C Vector store initializes empty anyway

D Texts are automatically embedded

💡 Hint

Refer to step 3 where embeddings instance is passed to Chroma

Concept Snapshot

Chroma vector store setup:
1. Import Chroma and embeddings.
2. Create embeddings instance.
3. Initialize Chroma with embeddings.
4. Add texts to store (embedding happens here).
5. Query vector store for similar texts.

Full Transcript

To set up a Chroma vector store in Langchain, first import the Chroma class and an embeddings model like OpenAIEmbeddings. Then create an embeddings instance. Next, initialize the Chroma vector store by passing the embeddings instance as the embedding_function parameter. After that, add your texts to the vector store using add_texts, which converts texts into vectors and stores them. Finally, you can query the vector store to find similar documents. This process ensures your texts are searchable by their meaning, not just words.

Practice

1. What is the main purpose of setting up a Chroma vector store in Langchain?

easy

A. To generate text summaries automatically

B. To store text as vectors for fast similarity search

C. To create a relational database for text storage

D. To visualize text data in charts

Solution

1. Step 1: Understand Chroma vector store role

   Chroma stores text as vectors to quickly find similar meanings, not for relational data or visualization.

2. Step 2: Identify main use case

   Its main use is fast similarity search using vector embeddings.

3. Final Answer:

   To store text as vectors for fast similarity search -> Option B

4. Quick Check:

   Chroma = vector storage for similarity [OK]

Hint: Chroma = vector storage for quick similarity search [OK]

Common Mistakes:

• Thinking Chroma is a database for tables
• Confusing it with text summarization tools
• Assuming it creates visual charts

2. Which of the following is the correct way to initialize a Chroma vector store with an embedding function named embed and collection name my_collection?

easy

A. store = Chroma(embedding=embed, collection='my_collection')

B. store = Chroma(collection='my_collection', embed=embedding_function)

C. store = Chroma('my_collection', embed_function=embed)

D. store = Chroma(embedding_function=embed, collection_name='my_collection')

Solution

1. Step 1: Check parameter names for Chroma initialization

   The correct parameters are embedding_function and collection_name.

2. Step 2: Match correct syntax

   store = Chroma(embedding_function=embed, collection_name='my_collection') uses embedding_function=embed and collection_name='my_collection', which is correct.

3. Final Answer:

   store = Chroma(embedding_function=embed, collection_name='my_collection') -> Option D

4. Quick Check:

   embedding_function + collection_name = store = Chroma(embedding_function=embed, collection_name='my_collection') [OK]

Hint: Use embedding_function and collection_name keys exactly [OK]

Common Mistakes:

• Using wrong parameter names like embed or collection
• Passing parameters in wrong order without keywords
• Confusing embedding_function with embed_function

3. Given the following code snippet, what will results contain after running store.similarity_search('Hello world')?

embed = SomeEmbeddingFunction()
store = Chroma(embedding_function=embed, collection_name='texts')
store.add_texts(['Hello world', 'Goodbye world'])
results = store.similarity_search('Hello world')

medium

A. A list containing 'Hello world' as the most similar text

B. An empty list because no texts were added

C. A list containing 'Goodbye world' only

D. A syntax error due to missing parameters

Solution

1. Step 1: Analyze texts added to the store

   The store has two texts: 'Hello world' and 'Goodbye world'.

2. Step 2: Understand similarity_search behavior

   Searching 'Hello world' will return texts most similar, so 'Hello world' itself is the closest match.

3. Final Answer:

   A list containing 'Hello world' as the most similar text -> Option A

4. Quick Check:

   Added texts include 'Hello world' = A list containing 'Hello world' as the most similar text [OK]

Hint: Similarity search returns closest added text matches [OK]

Common Mistakes:

• Assuming no texts were added so result is empty
• Thinking it returns unrelated texts
• Confusing syntax error with runtime behavior

4. Identify the error in this Chroma vector store setup code:

embed = SomeEmbeddingFunction()
store = Chroma(embedding_function=embed)
store.add_texts(['Sample text'])

medium

A. Missing collection_name parameter in Chroma initialization

B. Embedding function is not assigned correctly

C. add_texts method does not exist

D. Chroma requires a database URL parameter

Solution

1. Step 1: Check required parameters for Chroma

   Chroma requires both embedding_function and collection_name to be set.

2. Step 2: Identify missing parameter

   The code misses collection_name, which is mandatory.

3. Final Answer:

   Missing collection_name parameter in Chroma initialization -> Option A

4. Quick Check:

   collection_name missing = Missing collection_name parameter in Chroma initialization [OK]

Hint: Always include collection_name when creating Chroma store [OK]

Common Mistakes:

• Assuming embedding_function alone is enough
• Thinking add_texts method is invalid
• Believing database URL is mandatory

5. You want to store multiple documents in a Chroma vector store and then find the top 2 most similar documents to a query. Which code snippet correctly sets this up and performs the search?

hard

A. embed = Embedding()
store = Chroma(collection_name='docs')
docs = ['Doc1', 'Doc2', 'Doc3']
store.add_texts(docs)
results = store.similarity_search('query text', top=2)

B. embed = Embedding()
store = Chroma(embedding_function=embed, collection='docs')
docs = ['Doc1', 'Doc2', 'Doc3']
store.add_texts(docs)
results = store.similarity_search('query text', k=2)

C. embed = Embedding()
store = Chroma(embedding_function=embed, collection_name='docs')
docs = ['Doc1', 'Doc2', 'Doc3']
store.add_texts(docs)
results = store.similarity_search('query text', k=2)

D. embed = Embedding()
store = Chroma(embedding_function=embed, collection_name='docs')
docs = ['Doc1', 'Doc2', 'Doc3']
store.add_texts(docs)
results = store.similarity_search('query text')

Solution

1. Step 1: Verify Chroma initialization parameters

   embedding_function and collection_name must be set correctly; embed = Embedding() store = Chroma(embedding_function=embed, collection_name='docs') docs = ['Doc1', 'Doc2', 'Doc3'] store.add_texts(docs) results = store.similarity_search('query text', k=2) does this properly.

2. Step 2: Check similarity_search usage for top results

   embed = Embedding() store = Chroma(embedding_function=embed, collection_name='docs') docs = ['Doc1', 'Doc2', 'Doc3'] store.add_texts(docs) results = store.similarity_search('query text', k=2) uses k=2 to get top 2 results, which is the correct parameter.

3. Final Answer:

   Correct setup with embedding_function, collection_name, and k=2 in similarity_search -> Option C

4. Quick Check:

   embedding_function + collection_name + k=2 = embed = Embedding() store = Chroma(embedding_function=embed, collection_name='docs') docs = ['Doc1', 'Doc2', 'Doc3'] store.add_texts(docs) results = store.similarity_search('query text', k=2) [OK]

Hint: Use k=2 in similarity_search to get top 2 results [OK]

Common Mistakes:

• Using wrong parameter name like top instead of k
• Missing embedding_function or collection_name
• Not specifying number of results to return