Model interpretability with Azure OpenAI

From code to natural language

Model interpretability refers to the ease with which a human can understand the reasoning behind the predictions made by a machine learning model. It is the ability to understand how a model is making decisions and what factors are contributing to its predictions.

There are many reasons why model interpretability is relevant in the context of Machine Learning and Deep Learning. A few examples are:

• Making the code more efficient →this could reduce the training time
• Translating between different programming languages →this reduces the barriers among different coding skills
• Making the debugging easier →this speeds up error identification and fixing process
• Generating documentation →this could assist business users to understand the code behind models and how they actually work.

Since its General Availability, Azure OpenAI service already offers the API for Codex family models. Models belonging to this class are a fine-tuned version of the GPT-3 family, specialized for understanding and generating code.

My goal is that of being assisted by Azure OpenAI playground, using a Codex model, to have a clearer explanation of a Computer Vision project in Python…