yes

]]>I am a Matlab user. Could you please define the different variables in the function you gave above (i.e. h1[i], h2[i], bins [i],…) such that I can implement it in Matlab? Do the two histograms need to be with the same bin edges or could it be different for each histogram? ]]>

If this situation happens, it should still be possible to generate the prediction intervals I think. Take out the entire vector of responses from the leaf and combine with other vectors from next trees. Do the quantile logic on the concat of all these vectors..

]]>b) the node is not pure, but the feature vector is exactly the same for all responses – very rarely happens with real world datasets.

Are you implying this:

Same input features (X) can result in a wide variety of outputs (Y).

If yes, then I would argue that this does happen in a lot of practical use-cases. [You wont always have access to all the possible features to model some scenario / activity – so using a subset of features would result in this]

]]>