Ola Zytek
Ola Zytek
Realapp objects should support taking different algorithm transformers by explanation type - after all, this was the original purpose of having algorithm transformers. Some things we need to think about:...
Data imputation can be necessary for training and using ML models, but can lead to interpretability challenges if users don't realize that the values were imputed, and are therefore not...
The current testing code coverage is only about 50%. While `pyreal` does not currently have a required coverage threshold, there are likely important blocks of code being missed right now....
Right now, `Explainers` call `convert_columns_to_interpretable` on explanations to use their feature dictionary. This doesn't always make sense, since this method doesn't apply to all explanations. We should move this function...
The housing demo tutorial (for developers) needs to be finished up, with more explanation examples.
Based on problems encountered in Sibyl-API, it looks like KernelSHAP explainers are very large. We should investigate how large they are and if there is any way to reduce their...
RealApps may hold multiple SHAP explainers, such as local and global explainers. Fitting SHAP explainers can be a very slow process, and we may be able to reuse this work...
For this issue, Decision Tree explainers should be supported in RealApp, and the `plot_tree` function should be updated to have the consistent visualization format. For the explanation format, consider: https://scikit-learn.org/stable/auto_examples/tree/plot_unveil_tree_structure.html
Explanation algorithms like SHAP work particularly well when presented alongside a corresponding PDP/ICE plot (generating either in the traditional way or itself using a SHAP-based algorithm). This allows us to...
For this issue, we will add the top-level functions that automatically handle everything necessary to generate some visualization: ``` python # Curated list of explanations pyreal.visualize_model_explanation(realApp) pyreal.visualize_prediction_explanation(realApp, X) pyreal.visualize_prediction_explanation(realApp, X,...