Posteriors from binary:logistic about xgboost HOT 12 CLOSED

This is strange, the option is sufficient for getting the posterior probability. Maybe you can post a minimal script that reproduce the error.

from xgboost.

Hey

This is the Higgs script that you posted previously, slightly modified:

# setup parameters for xgboost
param = {}
# use logistic regression loss, use raw prediction before logistic transformation
# since we only need the rank
param['objective'] = 'binary:logistic'
# scale weight of positive examples
param['scale_pos_weight'] = sum_wneg/sum_wpos
param['bst:eta'] = 0.1 
param['bst:max_depth'] = 6
param['bst:subsample'] = 0.4
param['eval_metric'] = 'auc'
param['silent'] = 1
param['nthread'] = 16
param['booster_type'] = 0 #0 = trees, 1=linear

Thanks for the quick reply.

j

from xgboost.

This is because the xgboost does not remember objective in model in past version, and when you load it with empty parameter, it does not remember the option.

I have made a fix to the problem, check out the most recent version and see if it fixes your problem.

from xgboost.

You mean I have to reload param when I predict?

I will check the fix and let you know - appreciate the help.

j

from xgboost.

In the old version, when you load the model, you create a Booster with param that contains the same objective. The newest version should solve the problem

from xgboost.

I downloaded the new version, hit make (needed to remove -fopenmp flag), but libxboostpy.so was not generated so I got a traceback when xgboost was imported.

I kept the old version in xgboost_archive directory.

Any suggestions?

j

from xgboost.

If this is a compilation issue, pls report the compiler message. you need
to make in python folder

On Friday, August 15, 2014, HoD3521 [email protected] wrote:

I downloaded the new version, hit make (needed to remove -fopenmp flag),
but libxboostpy.so was not generated so I got a traceback when xgboost was
imported.

I kept the old version in xgboost_archive directory.

Any suggestions?

j

—
Reply to this email directly or view it on GitHub
https://github.com/tqchen/xgboost/issues/25#issuecomment-52375332.

Sincerely,

Tianqi Chen
Computer Science & Engineering, University of Washington

from xgboost.

$ /usr/bin/python -u higgs-numpy.py
../../python
../../python/libxgboostpy.so
Traceback (most recent call last):
File "higgs-numpy.py", line 15, in
import xgboost as xgb
File "../../python/xgboost.py", line 20, in
xglib = ctypes.cdll.LoadLibrary(XGBOOST_PATH)
File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/ctypes/init.py", line 443, in LoadLibrary
return self._dlltype(name)
File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/ctypes/init.py", line 365, in init
self._handle = _dlopen(self._name, mode)
OSError: dlopen(../../python/libxgboostpy.so, 6): image not found

from xgboost.

Let me try making in python folder and get back to you.

from xgboost.

making in the python folder worked - thanks.

from xgboost.

And the posteriors look correct. Really appreciate the help.

j

from xgboost.

I am glad it works out for you. Thank you for using xgboost

from xgboost.