Explainer Wrapper for captum TracInCP

[dilyabareeva]

• open a separate branch
• write an explainer wrapper for captum TracInCP https://github.com/pytorch/captum/blob/master/captum/influence/_core/tracincp.py
• has to follow the CaptumInfluence interface: all abstract method should be implemented
• self-influence method implementation should use the captum TracInCP self-influence method
• TracInCP requires generation of checkpoints for calculation - add checkpoints fixture to conftest.py for the MNIST model
• add tests to test_captum_influence.py - we aim for close to 100% coverage on new code
• run make styleto make sure all the lining and styling checks pass
• run pytest to see if all tests pass
• create a Pull Request into main branch

Info:

• the paper: https://arxiv.org/abs/2002.08484