introduction"> Introduction

In Pinecone, there are two APIs available for updating vector data: upsert, which performs full updates, and update, which performs partial updates. This article explains the distinction between these APIs.

Creating an Index

Execute the following commands or code to create an index in Pinecone.

bash

$ pip install -U pinecone-client

python

import pinecone

pinecone.init(
    api_key="API_KEY",
    environment="ENVIRONMENT"
)

if "sample" not in pinecone.list_indexes():
    pinecone.create_index("sample", dimension=1536)

index = pinecone.Index("sample")

Adding Data

python

index.upsert(
    vectors=[{
        "id": "1",
        "values': [0.0] * 1536,
        "metadata": {
            "content": "This is a sample vector"
        }
    }],
    namespace='my_namespace'
)

res = index.fetch(
    ids=["1"],
    namespace="my_namespace"
)

print(res["vectors"]["1"]["metadata"])

{'content': 'This is a sample vector'}

Updating Data

To update data, you can specify the id using the update or upsert API. When using upsert, it can be done as follows:

python

index.upsert(
    vectors=[{
        "id": "1",
        "values": [0.0] * 1536,
        "metadata": {
            "content": "Updated"
        }
    }],
    namespace="my_namespace"
)

res = index.fetch(
    ids=["1"],
    namespace="my_namespace"
)

print(res["vectors"]["1"]["metadata"])

{'content': 'Updated'}

Creating vectors consumes computing resources. If vector regeneration is unnecessary, it is advisable to use update to update only the metadata.

python

index.update(
    id="1",
    setMetadata={
        "content": "Only metatada updated"
    },
    namespace="my_namespace"
)

res = index.fetch(
    ids=["1"],
    namespace="my_namespace"
)

print(res["vectors"]["1"]["metadata"])

{'content': 'Only metatada updated'}

Differentiating Between Upsert and Update in Pinecone

Introduction

Creating an Index

Adding Data

Updating Data

Pinecone Sparse-Dense Vectors

Selective Metadata Indexing in Pinecone

Ryusei Kakujo