Ola Zytek
Ola Zytek
A lot of transformers may need to save some information about the data they transformed in order to transform the corresponding explanations. In some case, this is a single value...
This issue extends #204, with updated support for multivariate time series explanations. For this issue, the occlusion time series explainer should work on data that includes multiple signals, and include...
With the move to gitbooks, our existing testing protocol does not cover code in the quickstart. We should find a new way to test this code.
In `GlobalFeatureImportance`, the SHAP value computation currently has the additivity check turned off (`self.explainer.shap_values(x_model_np, check_additivity=False)`), to prevent an error that was happening with `TreeExplainer`. This should be investigated further with...
For now, Pyreal is slower than alternatives in many cases. This is partly because Pyreal maintains the original data and copies it to run transforms and partially due to unknown...
The benchmarking system addressed in #62 only runs on classification models. This should be updated to also include regression models. This will likely require downloading new datasets.
Copying data when transforming can be a resource-intensive process. To lay the groundwork for future optimizations, an "inplace" flag should be added to the transformation processes that allows the transformation...
As Pyreal grows, so will the time taken for the benchmarking process. We intend to run this process using supercomputing resources where possible. For this issue, we need to add...
The `_helper_summed_values` function in `transformer/one_hot_encode/OneHotEncoder` should be moved to the `AdditiveFeatureContributions/Importances` class, or directly to the `DataFrameExplanation` class, so it can be used by future transformers.
PR #87 added the function `evaluate_variation` to `Explainer` objects. This function should also be added to the official API documentation.