Ranges of output values of different explainers and visualization of relevances and used colormaps

[elboyran]

The RISE relevances are always positive.

Then why do we display also blue in the color-map for example in the MNIST tutorial and generic tutorial? They don't look scaled properly!

Is the KernelSHAP inside the dashboard calculated properly? Again only positive values, but blue in the dashboard example. In the notebook, it looks correct.

Became more obvious by adding the heatmap colormap explanation to the dashboard Image page (still inside the work in progress PR #855 ):

Also, RISE relevances for the Tabular sunshine prediction regression task are not scaled correctly (we do not have such a tutorial). The RISE relevances for row 10 are not scaled 0 to 1.

The last relevances are calculated in dashboard/_models_tabular.py according to Laura.

Probably other errors like these can be spotted in the dashboard....

UPDATE: will not be addressed in this project, but included in the next one (if funded)!