Guillaume VIGNAL

Shap has done breaking changes in 0.45.0 version ; shapash's test don't pass

Shap has done breaking changes in 0.45.0 version. Meanwhile we fix it in Shapash, we limit the shap version less than 0.45.0 strictely. Fixes #533

Rather than computing predictions and probabilities when needed in different plot or when launching the webapp, we made the choice to do it during the compile step. Fixes #531

Shap has done breaking changes in 0.45.0 version. Meanwhile we fix it in Shapash, we limit the shap version less than 0.45.0 strictly.

By default in the report we have only 200 points into the different plots (in order to have a "light" report in terms of size). In our case to be...

Actually, if the prediction values are not given to the smart_explainer, predictions and probabilities are computed during the launching of the app, or some plot visualization. It would be better...

In the Contribution plot for categorical, a sampling is done in the webapp randomly. The consequence is that for some modalities of the variable can disappear due to this behavior....

There are elements in local plot which are unclear. The term Response which says that the class 1 is Survival and not that the true value for this id is...

The representation of the density of the points is not proportional to the number of points. Let's try to make it proportional to test if the visualization is better. 

# Description In this feature we add : - jittering points on violin plot - smart selection to keep diversity and represent all the classes for discrete variables - a...