evalutating only the first n classifiers in an ensemble

[dmlc]

evalutating only the first n classifiers in an ensemble

[tqchen]

If you mean evaluating the results and print it into screen, you can
directly put 'test set' into the watchlist(note xgboost support watch
multiple test sets), this will give you the loss of testset of entire
training phase.

On Mon, Jun 2, 2014 at 3:07 AM, andreh7 [email protected] wrote:

is it possible to use only the first n classifiers (e.g. trees) in the
ensemble ? Is it possible to do this from python ?

(this would be interesting in order to see at which point the 'loss' on
the test set increases again or becomes larger than the loss on the
training set)

—
Reply to this email directly or view it on GitHub
https://github.com/tqchen/xgboost/issues/12.

[andreh7]

thanks a lot for the quick reply ! I was actually more thinking of getting the values programmatically but I'll start by what you suggested. Thanks !

[tqchen]

Another possible way, if you want to do it in python way, is to take a look at train function in
https://github.com/tqchen/xgboost/blob/master/python/xgboost.py
tweak it a bit (adding prediction, evaluation after each update, which is not hard),
you should remember to add your testing data to cachelist by

 bst = Booster(params, [dtrain]+[dtest] )

So that it caches test prediction as well and make it faster to repeatily predict dtest

[tqchen]

OK there is another request for this feature in #63 , and it is now supported by the most recent version in master. In python, you can do bst.predict(data, ntree_limit=10) to use first 10 trees.