Arnaud Joly
Arnaud Joly
Hi @mchangun, I think this would a nice addition and that you would be interested in #399 and #970.
The best way to add new features is to make a pull request. Please have a look to the contributing guidelines http://scikit-learn.org/dev/developers/index.html. Be aware that adding new features require more...
> Is this not implemented already? http://scikit-learn.org/dev/modules/multiclass.html lists KNN as being multi-label. Yes, the binary relevance (one-vs-rest) is already implemented. However, ML-KNN propose to compute to perform prediction in a...
The weighted median will discard such value since they will have a weight of 0.
> But even with your code it seems to me that diff[i] becomes zero if h_pred[i] = 0. It will give you the infinity.
Sure, this will also work. Note that the sign might be flipped. However, this doesn't matter.
> Remove constant feature caching should it prove to not actually be helpful (it wasn't in my original tests) This part is useful for tree built in depth-first wise with...
> Another thing that I would like to do, once @arjoly is done with #5203 and his next PR, is to split this huge _tree.pyx file into several files, so...
I would add to your list all the criterion optimizations that you have done in #5041: - pre-computing w \* y_ik \* y_ik and w \* y_ik for MSE and...
I am updating the list to add the test about importance computations.