PyTorch Model Interpretability and Interface to Sklearn
Pradeepta Mishra · Apress eBooks · 2022
Model interpretability is an area that needs special attention because it is connected with model adoption in particular and AI adoption in general. Users will adopt a model and framework if they can explain the decisions or predictions generated by the deep learning model. In this chapter, you will explore a new framework called Captum, which consists of a set of algorithms that can explain or help us interpret the predictions, model results, and layers of a neural network model. In this chapter, you are also going to use another framework called skorch, which is a library compatible for Scikit-learn users. Machine learning users prefer the sklearn library to train models, perform grid searches, and identify the best hyper parameters of the models—the same kind of seamless experience the users can experience when developing deep neural network models using PyTorch.