Hi Ilya
Here is a brief description of each annimation, created using your package.
This was a standard signal/background discrimination task in 2-dimensions.
1. fig_ptmet_DT.avi
This shows the tree-structure in the 2 variables. Each frame
corresponds to be a different tree in the sequence of binary
decision trees. This annimation illustrates the fact that a DT
is just a clever way to histogram data (with non-uniform bins)
in n-dimensions. Moreover, a BDT is just an average over these
histograms.
2. fig_BDT_train.avi
This shows the BDT output for signal and background. Each
frame corresponds to the distribution created using a BDT with
increasing numbers of trees. (The number grows from 10 to 1000.)
The point here is the observation that if one has a sufficiently
large number of trees the AdaBoost algorithm will yield perfect
separation on the training sample.
3. fig_BDT_test.avi
Same as 2. except using an independent test sample.