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.