Skip to content

Commit

Permalink
Documentation page for ASTRADB
Browse files Browse the repository at this point in the history
  • Loading branch information
@clun
clun committed Apr 25, 2024
1 parent ef3df82 commit c4a0b93
Show file tree
Hide file tree
Showing 3 changed files with 144 additions and 27 deletions.
127 changes: 122 additions & 5 deletions docs/docs/integrations/embedding-stores/astra-db.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -53,9 +53,9 @@ To get a token click the `[Generate Token]` button on the right. It will generat

> The full documentation regarding AstraDB can be found [here](https://docs.datastax.com/en/astra/astra-db-vector/api-reference/dataapiclient.html).
### 2.1. Setup your Collections
### 2.1. Connecting to Astra

- Client initialization
The following code show how to initialize the client and access the database.

```java
String token = System.getenv("ASTRA_DB_APPLICATION_TOKEN");
Expand All @@ -69,9 +69,126 @@ DataAPIClient client = new DataAPIClient(token);
Database db = client.getDatabase(astraApiEndpoint);
```

### 2.2. Ingestion
| Field | Description |
|-----------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------|
| **[DataAPIClient](https://datastaxdevs.github.io/astra-db-java/latest/com/datastax/astra/client/DataAPIClient.html)** | This class is the main entry point for the Astra Client. It allows to create databases and different admin operations |
| **[Database](https://datastaxdevs.github.io/astra-db-java/latest/com/datastax/astra/client/Database.html)** | The API endpoint you copied in the previous step. |

- From a Tenant (DataAPIClient) you can access one to many databases. The `Database` object is the entry point to interact with the database.
-
- From a Database (Database) you can access one the many namespaces (logical). The default namespace is `default_keyspace.

### 2.2. AstraDB Collection

A Database can have one to multiple collections. A collection is a logical grouping of data. A collection can store different type of data abd can contains a `$vector` field. Those collections can be used to store any informations and not only vectors. It can then be used for ChatMemory or any cache needed.

> AstraDB collections can use different types of identifiers for its documents. Default is the UUIDv4 (java UUID) but more can be
> use like ObjectId (MongoDB) , UUIDv7 (Snowflake) or other type of identifier. To get the complete list consult the [documentation](https://docs.datastax.com/en/astra/astra-db-vector/api-reference/collections.html#the-defaultid-option)
In Langchain4j, `AstraDBEmbeddingStore` is associated to one Collection with a `$vector` field. The `$vector` field is used to store the embeddings. The following code shows how to create a collection with a `$vector` field. By default there is no special field to store the text segment of the chunk. By CONVENTION, the store use field name `content`.
```java
// Create a vector collection
Collection<Document> col = db.createCollection("langchain4j_embedding_store",
CollectionOptions
.builder()
.vectorDimension(1536) // related to your embedding mode
.vectorSimilarity(SimilarityMetric.COSINE) //
.indexingDeny("content") // avoid to index text segment
.build());
```

If the collection exists (most of the time), you can access it with the following code:
```java
Collection<Document> col2 = db
.getCollection("langchain4j_embedding_store");
```

### 2.3. Init EmbeddingStore

To initialize the `AstraDBEmbeddingStore` simply give the collection object as argument.

```java
EmbeddingStore<TextSegment> embeddingStore = new AstraDBEmbeddingStore(col);
```

We could provide this utility method to help with creation of the store:

```java
/**
* Create an AstraDB Embedding Store.
*/
EmbeddingStore<TextSegment> createEmbeddingStore(
String astraToken, String apiEndpoint,
String collectionName, int dimension,
SimilarityMetric metric) {

return new AstraDBEmbeddingStore(
// AstraDB Client
new DataAPIClient(astraToken)
.getDatabase(apiEndpoint)
.createCollection(collectionName, dimension, metric));
}
```

### 2.4. Usage

Please find enclosed a simple example of how to use the `AstraDBEmbeddingStore`:

```java
// Given a embedding model
EmbeddingModel embeddingModel = ...;

// Given a text file on disk
Path textFile = ....;

// Get AstraDBEmbeddingStore
EmbeddingStore<TextSegment> embeddingStore = createEmbeddingStore(
System.getenv("ASTRA_DB_APPLICATION_TOKEN"),
System.getenv("ASTRA_DB_API_ENDPOINT"),
"langchain4j_embedding_store",
1536,
SimilarityMetric.COSINE);

// Ingestion
EmbeddingStoreIngestor.builder()
.documentSplitter(recursive(100, 10, new OpenAiTokenizer(GPT_3_5_TURBO)))
.embeddingModel(embeddingModel)
.embeddingStore(embeddingStore)
.build()
.ingest(loadDocument(textFile, new TextDocumentParser()));

// Sample Vector Search
ContentRetriever contentRetriever = EmbeddingStoreContentRetriever.builder()
.embeddingModel(embeddingModel)
.embeddingStore(embeddingStore)
.maxResults(2)
.minScore(0.5)
.build();

Assistant ai = AiServices.builder(Assistant.class)
.contentRetriever(contentRetriever)
.chatLanguageModel(initChatLanguageModelOpenAi())
.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
.build();
String response = ai.answer("What vegetable is Happy?");


// Meta-Data Filtering
RetrievalAugmentor retrievalAugmentor = DefaultRetrievalAugmentor
.builder()
.contentRetriever(contentRetriever)
.contentInjector(DefaultContentInjector.builder()
.metadataKeysToInclude(asList("document_format", "text"))
.build())
.build();

// configuring it to use the components we've created above.
Assistant aiWithMetaData = AiServices.builder(Assistant.class)
.retrievalAugmentor(retrievalAugmentor)
.chatLanguageModel(getChatLanguageModelChatBison())
.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
.build();
```

### 2.3. Vector Search

### 2.4. Meta-Data Filtering

42 changes: 21 additions & 21 deletions docs/docs/integrations/embedding-stores/index.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,24 +4,24 @@ hide_title: false
sidebar_position: 0
---

| Provider | Storing Metadata | Filtering by Metadata | Local | Cloud |
|---------------------------------------------------------------------------------------|------------------|-----------------------|-------|-------|
| [In-memory](/integrations/embedding-stores/in-memory) | ✅ | ✅ | | |
| [Astra DB](/integrations/embedding-stores/astra-db) | ✅ | | | |
| [Azure AI Search](/integrations/embedding-stores/azure-ai-search) | ✅ | | | |
| [Azure CosmosDB Mongo vCore](/integrations/embedding-stores/azure-cosmos-mongo-vcore) | ✅ | | | |
| [Cassandra](/integrations/embedding-stores/cassandra) | ✅ | | | |
| [Chroma](/integrations/embedding-stores/chroma) | ✅ | | | |
| [Elasticsearch](/integrations/embedding-stores/elasticsearch) | ✅ | ✅ | | |
| [Infinispan](/integrations/embedding-stores/infinispan) | ✅ | | | |
| [Milvus](/integrations/embedding-stores/milvus) | ✅ | ✅ | | |
| [MongoDB Atlas](/integrations/embedding-stores/mongodb-atlas) | ✅ | | | |
| [Neo4j](/integrations/embedding-stores/neo4j) | | | | |
| [OpenSearch](/integrations/embedding-stores/opensearch) | ✅ | | | |
| [PGVector](/integrations/embedding-stores/pgvector) | ✅ | | | |
| [Pinecone](/integrations/embedding-stores/pinecone) | | | | |
| [Qdrant](/integrations/embedding-stores/qdrant) | ✅ | | | |
| [Redis](/integrations/embedding-stores/redis) | ✅ | | | |
| [Vearch](/integrations/embedding-stores/vearch) | ✅ | | | |
| [Vespa](/integrations/embedding-stores/vespa) | | | | |
| [Weaviate](/integrations/embedding-stores/weaviate) | | | | |
| Provider | Storing Metadata | Filtering by Metadata | Local | Cloud |
|---------------------------------------------------------------------------------------|:------------------:|:---------------------:|:-------:|:-------:|
| [In-memory](/integrations/embedding-stores/in-memory) | ✅ | ✅ | | |
| [AstraDB](/integrations/embedding-stores/astra-db) | ✅ | ✅ | | ✅ |
| [Azure AI Search](/integrations/embedding-stores/azure-ai-search) | ✅ | | | |
| [Azure CosmosDB Mongo vCore](/integrations/embedding-stores/azure-cosmos-mongo-vcore) | ✅ | | | |
| [Apache Cassandra™](/integrations/embedding-stores/cassandra) | ✅ | ✅ (partial) | ✅ | ✅ |
| [Chroma](/integrations/embedding-stores/chroma) | ✅ | | | |
| [Elasticsearch](/integrations/embedding-stores/elasticsearch) | ✅ | ✅ | | |
| [Infinispan](/integrations/embedding-stores/infinispan) | ✅ | | | |
| [Milvus](/integrations/embedding-stores/milvus) | ✅ | ✅ | | |
| [MongoDB Atlas](/integrations/embedding-stores/mongodb-atlas) | ✅ | | | |
| [Neo4j](/integrations/embedding-stores/neo4j) | | | | |
| [OpenSearch](/integrations/embedding-stores/opensearch) | ✅ | | | |
| [PGVector](/integrations/embedding-stores/pgvector) | ✅ | | | |
| [Pinecone](/integrations/embedding-stores/pinecone) | | | | |
| [Qdrant](/integrations/embedding-stores/qdrant) | ✅ | | | |
| [Redis](/integrations/embedding-stores/redis) | ✅ | | | |
| [Vearch](/integrations/embedding-stores/vearch) | ✅ | | | |
| [Vespa](/integrations/embedding-stores/vespa) | | | | |
| [Weaviate](/integrations/embedding-stores/weaviate) | | | | |
2 changes: 1 addition & 1 deletion ...-astradb/src/main/java/dev/langchain4j/store/embedding/astradb/AstraDBEmbeddingStore.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ public AstraDBEmbeddingStore(@NonNull Collection<Document> client) {
* @param concurrentThreads
* concurrent threads
*/
public AstraDBEmbeddingStore(@NonNull Collection<Document> client, int itemsPerChunk, int concurrentThreads) {
public AstraDBEmbeddingStore(@NonNull Collection<Document> client, int itemsPerChunk, int concurrentThreads) {
if (itemsPerChunk>20 || itemsPerChunk<1) {
throw new IllegalArgumentException("'itemsPerChunk' should be in between 1 and 20");
}
Expand Down

0 comments on commit c4a0b93

Please sign in to comment.