Pie Charts

[SpisakJ]

Thank you for this great tool.
I think it would be even better with an option to display all nodes as piecharts.
Since different depths can lead to different configurations and i might want to use high depth for classification but only visually show some low depths. In these cases i would like to turn the bar plots to pie charts for lower depths where nodes are not leafs.

[pplonski]

Thank you @SpisakJ! Nice suggestion. May I ask what is your use case and what algorithm are you visualizing?

[SpisakJ]

Im visualising a DecisionTreeClassifier from sklearn.tree.
I used the decision tree as an analysis tool to get an idea about the internal workings of a neural network.
So i basically end up with two classes and many features, I trained the tree up to depth 7 to get high purity in the leafs but also want to show a smaller tree in the paper since of course the full depth tree is very hard to show. With supertree i had the option of couse to manually explore the many branches which made it possible to get a good idea even from the full tree.

Now for the tree of only depth 3 (taking the existing depth 7 tree but only looking at the first three depths) the visualisation only has the bar plots where its very hard to see how many samples of each class are in the node unless you hover over it for the text description. So here i would have liked the option to see the piecharts that are visualised at the leafs in depath 7 for these prior nodes.

With your project i think that would be somwhere in the supertree.min.js file but with the compressed names I couldnt get a clear idea on how to do it. For now i just ended up creating some plotting functions for it myself.

Roughly like this:

                                     (X|Y) (all samples pie chart)
                                    /      \  (edges)
                       (X_1|Y_1)   (X_2|Y_2)  (piecharts after first split, instead of having the decision boundary visualised with bar plots)
                     /         \     /             \       (more edges)
               (X_3|Y_3)    ....    ....           .....    (leaf pie charts, as they already exist) 

[pplonski]

Wow! analyzing NN with decision tree is mind blowing idea

We kept js compressed because we want to sell license for uncompressed code ... but we are thinking about changing it, and keep all open source. Stay tuned