Skip to content

Latest commit

 

History

History

python

README.md

PxPermute

Overview

The Channel Interpretation Python Package is a tool that allows you to perform channel interpretation using the PxPermute method as well as adapted pixel-wise interpretation methods implemented via captum library. It helps you analyze and interpret the contributions of individual channels in your data.

Installation

  1. First, make sure you have poetry installed.

  2. Set up a virtual environment and install the package dependencies using Poetry:

    poetry install

Usage

For using the package, your data has to have the structure described in scifAI documentation.

  1. Before running the interpretation method on your data, update a configuration file (config.cfg) according to your needs. You can find an example of the configuration file in the repository. The configuration file consists of the following sections:

    • [general] - contains parameters like device, batch_size ad num_workers.
    • [data] - contains information about the data you want to interpret: path to the data file, the channels to interpret with the corresponding names and other parameters to normalize your data.
    • [interpretation] - contains information about the interpretation method you want to use. You can specify the shuffle_times for PxPermute, and some other parameters used for captum interpretation methods.
    • [model] - contains path to the model.

  2. Run the package using the following command:

    poetry run python main.py --method <interpretationn_method> --output-path <path_to_output_file>

    Per default, the output file will be saved in the output folder, and the method will be PxPermute.

  3. The output are plots showing the contributions of individual channels in your data.

  4. Currently, the package works only with ResNet-18 model. To run interpretation methods on your own model, you have to adjust the load_model function in util.py file. The function should return the model.

Interpretation Methods

Choose from a variety of interpretation methods to analyze your data:

  • PxPermute
  • DeepLift
  • GuidedGradCam
  • Saliency
  • DeepLiftShap
  • GradientShap
  • InputXGradient
  • IntegratedGradients
  • GuidedBackprop
  • Deconvolution
  • Occlusion
  • FeaturePermutation
  • ShapleyValueSampling
  • Lime
  • KernelShap
  • LRP