Master TODO

[bogdan-kulynych]

Tasks

README.md

• Blurb should list tasks.
• Links to Papers / bibs - 1. Statistical Inference. 2. Responsiveness Verification, 3. Recourse Verification
• Installation can be list high level requirements: Python v 3.??, SCIP/CPLEX. Add Link to How to Obtain CPLEX to docs.
• Update Quickstart Task to Classification with Continuous Features and Causal Constraints
  • Implement Auditor to Test/Measure Responsiveness
  • Implement Filtering for Check Feasibility of Counterfactual Explanations from other Methods
• Add Contributing Section see example

Docs

• Sphinx to Generate Docs from docstrings

MIP Solver

• Ability to Edit SCIP Solver Parameters
• Tests/Code for MIP Solver Selection
  • Import should check that SCIP/CPLEX are installed and throw warning if not.
  • Default Solver Selection load SCIP if scip_requested or (not cplex_installed) else load CPLEX
  • Import throws warning if CPLEX Community Edition is installed. In this case, the default solver should be SCIP

Release

• Add Test to Check Installation
• Some Test for Docs
• Add Redirects actionable-recourse
• GIF of Functionality?

Later

• Functionality for Regression
• LIst of Good Issues to Get Started
• Update sample size calculation to match paper #4

List of Tasks

What you can do with this package:

1. Test how the predictions change with respect to complex interaction models

• Specification is Instance Level.
• API makes this Fast/Easy/Versatile
• Exact Control over Actionability of Each Feature/Multiple Features, Causal Models, Probabilistic Changes

2. Build and Deploy Safer Models

• Test if Models Assign Fixed Predictions (e.g., in lending)
• Measure Responsiveness (e.g., to gaming)

3. Provide Better Explanations

• Recourse Actions
• Feature Responsiveness Scores