iXAI
iXAI copied to clipboard
mmschlk
Fast and incremental explanations for online machine learning models. Works best with the river framework.
Hello, I have run the sample code and already when initializing the IncrementalPFI the passed loss function (as in the example Accuracy()) is checked with the method _validate_loss_function_. In the...
iPFI and iSAGE attribute all importance to categorical variables and return ~0 fi scores for all numerical variables. The issue can be reproduced using this code. ``` # Load imports...
Fixes 1. Handles unseen labels in multiclass classification for PDP 2. Added option to not update storage in PDP 3. Added is_classification flag in PDP
Code to reproduce the issue ``` from river import metrics from river.utils import Rolling from river.ensemble import AdaptiveRandomForestClassifier from river.datasets import ImageSegments from river import preprocessing from river import compose...
Create default loss function that calculates the absolute deviation between the labels and predictions (like a norm in PFI). This can be the default loss for all explainers with a...
Add all doc dependencies to setup.py (only top level imports no secondary dependencies). In setup.py line docs_packages is a list of all required packages in typical req.txt format (version numbers...
Create pre-commit hooks that run static tests and then all tests as a github action workflow on push
Add unit tests for all tracker objects in `utils.tracker`
The examples currently sitting in the root of the projects should be moved into notebooks and run to showcase the explanations. For this the visualization (#29) and the validators (#24)...