Cobalt: ML Model Understanding and Repair

Cobalt is a toolkit for illuminating and improving AI models and data.

---

Why Cobalt?

The time you have to understand and fix your model’s errors is limited, expensive, and hard to scale to the size of your dataset. Cobalt automates the otherwise painful step of looking for patterns in how your models are performing. We're here to make topological data analysis (TDA) easy to use.

---

Get Started

Create a new virtual environment and install Cobalt with pip:

pip install cobalt-ai

Then register your copy of Cobalt by opening a Python shell and running:

import cobalt
cobalt.register_license()

See the setup instructions for more details.

---

Cobalt Features

• Easily start analysis for a model or dataset with a few lines of code. Cobalt supports text, image, and tabular datasets.
• Automatically identify problematic groups of data in your model, saving days or weeks of troubleshooting effort.
• Quickly compare models and assess the deployment risk of each model for your use case.
  Example: Comparing embedding models for product search.
• Use the groups discovered by Cobalt to:
  • Curate your data.
  • Retrain and fine-tune models.
  • Develop intuitive test cases based on Cobalt's intelligent groups.
• Explore an interactive visualization of your dataset, model errors, or embedding model using our TDA-based dimensionality reduction.

---

Documentation

• Full Documentation
• Example: Choosing the Best Embedding Model for Your E-commerce Vector Database

---

Community

• Slack: Join the BlueLightAI Community
• Website: BlueLightAI
• LinkedIn: BlueLightAI on LinkedIn
• YouTube: BlueLightAI on YouTube