datascope 0.0.1

Measuring data importance over ML pipelines using the Shapley value.

# DataScope: Measuring data importance over ML pipelines using the Shapley value

This is a tool for inspecting ML pipelines by measuring how important is each training data point for predicting the label of a given test data example.

## Development Guide

### Setup

#### Install libGL.so for opencv ` apt-get install ffmpeg libsm6 libxext6 -y `

#### Create a new conda virtual environment ` conda env create -f datascope.yml `

#### Install datascope ` python setup.py install `

#### Install datascope-pipelines

### Repo Structure

This repository has several sections:

• Algorithms for computing the Shapley value are in datascope/algorithms

• Code for inspecting ML pipelines defined with sklearn are in datascope/inspection

• All tests are stored under tests

• Any experiment scripts and Jupyter notebooks are stored under experiments

### Adding Dependencies

For now we can keep all Python dependencies in the requirements.txt. Later on we will separate them by “flavor” (e.g. test dependencies, experiment dependencies).