automl.refit() XGBoostError: base_score must be in (0,1) for logistic loss

[zygmuntz]

I get the following error when running refit( x_train, y_train ):

In [18]: automl.refit( x_train, y_train )
---------------------------------------------------------------------------
XGBoostError                              Traceback (most recent call last)
<ipython-input-18-16f19027783b> in <module>()
----> 1 automl.refit( x_train, y_train )

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/estimators.pyc in refit(self, X, y)
	46 
	47         """
---> 48         return self._automl.refit(X, y)
	49 
	50     def fit_ensemble(self, y, task=None, metric=None, precision='32',

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/estimators.pyc in refit(self, X, y)
	46 
	47         """
---> 48         return self._automl.refit(X, y)
	49 
	50     def fit_ensemble(self, y, task=None, metric=None, precision='32',

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/automl.pyc in refit(self, X, y)
	424                 # this updates the model inplace, it can then later be used in
	425                 # predict method
--> 426                 model.fit(X.copy(), y.copy())
	427 
	428         self._can_predict = True

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/pipeline/base.pyc in fit(self, X, y, fit_params, init_params)
	61         X, fit_params = self.pre_transform(X, y, fit_params=fit_params,
	62                                           init_params=init_params)
---> 63         self.fit_estimator(X, y, **fit_params)
	64         return self
	65 

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/pipeline/base.pyc in fit_estimator(self, X, y, **fit_params)
	136         if fit_params is None:
	137             fit_params = {}
--> 138         self.pipeline_.steps[-1][-1].fit(X, y, **fit_params)
	139         return self
	140 

/usr/local/anaconda/lib/python2.7/site-packages/autosklearn/pipeline/components/classification/xgradient_boosting.pyc in fit(self, X, y)
	127                 seed=self.seed
	128                 )
--> 129         self.estimator.fit(X, y)
	130 
	131         return self

/usr/local/anaconda/lib/python2.7/site-packages/xgboost/sklearn.pyc in fit(self, X, y, sample_weight, eval_set, eval_metric, early_stopping_rounds, verbose)
	341                               early_stopping_rounds=early_stopping_rounds,
	342                               evals_result=evals_result, feval=feval,
--> 343                               verbose_eval=verbose)
	344 
	345         if evals_result:

/usr/local/anaconda/lib/python2.7/site-packages/xgboost/training.pyc in train(params, dtrain, num_boost_round, evals, obj, feval, maximize, early_stopping_rounds, evals_result, verbose_eval, learning_rates, xgb_model)
	119     if not early_stopping_rounds:
	120         for i in range(num_boost_round):
--> 121             bst.update(dtrain, i, obj)
	122             nboost += 1
	123             if len(evals) != 0:

/usr/local/anaconda/lib/python2.7/site-packages/xgboost/core.pyc in update(self, dtrain, iteration, fobj)
	692 
	693         if fobj is None:
--> 694             _check_call(_LIB.XGBoosterUpdateOneIter(self.handle, iteration, dtrain.handle))
	695         else:
	696             pred = self.predict(dtrain)

/usr/local/anaconda/lib/python2.7/site-packages/xgboost/core.pyc in _check_call(ret)
	95     """
	96     if ret != 0:
---> 97         raise XGBoostError(_LIB.XGBGetLastError())
	98 
	99 

XGBoostError: base_score must be in (0,1) for logistic loss

[mfeurer]

This is related to #163. Although I don't know what one can do if boosting fails with such an error. Is this issue reproducible?

[mfeurer]

Closing this because XGBoost will be removed from auto-sklearn due to issue #271.