Embedding Visualization with LangChain + OpenAI

This repository demonstrates how to generate text embeddings using OpenAI models through LangChain, and then visualize the semantic similarity between words in 3D space using PCA and Matplotlib.

Features

• Generate embeddings for arbitrary text using OpenAI’s embedding models.
• Support for multiple models:
  • text-embedding-3-large
  • text-embedding-3-small
  • text-embedding-ada-002 (default)
• Dimensionality reduction using PCA.
• Interactive 3D scatter plot visualization of embeddings.

Requirements

Install dependencies with:

pip install -r requirements.txt

requirements.txt

langchain-openai
python-dotenv
numpy
matplotlib
scikit-learn

You will also need an OpenAI API key.

Setup

1. Clone this repository:

   git clone https://github.com/your-username/embedding-visualizer.git
   cd embedding-visualizer

2. Create a .env file in the root directory and add your OpenAI API key:

   echo "OPENAI_API_KEY=your_api_key_here" > .env

3. Choose your embedding model by editing the EMBEDDING_MODEL variable in the script:

   EMBEDDING_MODEL="text-embedding-ada-002"

Usage

Run the script to generate embeddings and plot them:

python embeddings_plot.py

This will:

1. Generate embeddings for the hardcoded list of words:

   texts = ["nfl", "football", "soccer", "basketball", "baseball"]

2. Reduce them to 3D space using PCA.
3. Save the visualization to 3d_plot_small.png.

Example output:

📊 A 3D scatter plot showing the relative similarity of sports terms.

Project Structure

.
├── embeddings_plot.py   # Main script
├── requirements.txt     # Dependencies
└── .env                 # API key (not committed)

Customization

• To change the words being compared, edit the texts list in embeddings_plot.py.
• To try a different embedding model, set EMBEDDING_MODEL accordingly.
• To adjust plot resolution, modify the dpi parameter in:

  plt.savefig("3d_plot_small.png", dpi=1000, bbox_inches='tight')

Example Plot