mfumagalli68/xi-method

Xi method

xi-method's Issues

Tests: test code, test code against different OS

Custom separation measure provided by the user

First milestone

Commit: e16adef

The result i get with wine quality dataset it's the same as shown in the paper.

first step

To explore:

Code related:

• Poetry as a tool for packaging code https://github.com/python-poetry/poetry
• License ( what's the most appropriate)

Structure:

• What functions we include in the package beyond the main xi function

Next step

@ValentinaGhidini

List of next steps:

• Fix function to check overlaps of args.
  User can specify if a feature is discrete, give a number of observation in the partition or number of partition.
  Check that we don't have conflict between those choices in terms of features.
• Visualization: extend package functionality to include plotting. See issue #1
• Code for regression
• #4
  Need to think how to do it.
• #7

plot img data

I am not sure about the "df" in input of plot_img(): df should be the heatmap, i.e. a matrix, with the same dimension as the original image and values equal to the explanations for each pixel (in the correct positions!).

https://github.com/mfumagalli68/xi/blob/f27154c96f3911f5d1de8ac0e32b357c81707add/xi/plotting/plot.py#L36

Moreover, specify somewhere that we are using one-channel images (we could extend it to multi-channel images quite easily, but I need to think how to represent the final heatmap).

plot tabular data

The number of default bars should be either k=10 or k=3 (see line 6)

https://github.com/mfumagalli68/xi/blob/f27154c96f3911f5d1de8ac0e32b357c81707add/xi/plotting/plot.py#L6