Early stopping needs a default metric - python package

[yychenca]

It appears for early stopping the current version doesn't specify a default metric and therefore if we didn't explicitly define a metric it will fail:

import lightgbm as lgb
from sklearn import datasets, metrics, model_selection

rng = np.random.RandomState(2016)

X, y = datasets.make_classification(n_samples=10000, n_features=100)
x_train, x_test, y_train, y_test = model_selection.train_test_split(X, y, test_size=0.1, random_state=1)
lgb_model = lgb.LGBMClassifier(n_estimators=100).fit(x_train, y_train, [(x_test, y_test)]
#, eval_metric="auc" ## commented out the metric
,early_stopping_rounds=1
)

---

ValueError Traceback (most recent call last)
in ()
10 lgb_model = lgb.LGBMClassifier(n_estimators=100).fit(x_train, y_train, [(x_test, y_test)]
11 #, eval_metric="auc"
---> 12 ,early_stopping_rounds=1
13 )
14 # lgb_model.predict(x_test)

/Users/cchen1/anaconda/lib/python2.7/site-packages/lightgbm-0.1-py2.7.egg/lightgbm/sklearn.pyc in fit(self, X, y, eval_set, eval_metric, early_stopping_rounds, verbose, train_fields, valid_fields, other_params)
333 eval_metric, early_stopping_rounds,
334 verbose, train_fields, valid_fields,
--> 335 other_params)
336 return self
337

/Users/cchen1/anaconda/lib/python2.7/site-packages/lightgbm-0.1-py2.7.egg/lightgbm/sklearn.pyc in fit(self, X, y, eval_set, eval_metric, early_stopping_rounds, verbose, train_fields, valid_fields, other_params)
250 early_stopping_rounds=early_stopping_rounds,
251 evals_result=evals_result, fobj=self.fobj, feval=feval,
--> 252 verbose_eval=verbose, train_fields=train_fields, valid_fields=valid_fields)
253
254 if evals_result:

/Users/cchen1/anaconda/lib/python2.7/site-packages/lightgbm-0.1-py2.7.egg/lightgbm/engine.pyc in train(params, train_data, num_boost_round, valid_datas, valid_names, fobj, feval, init_model, train_fields, valid_fields, early_stopping_rounds, evals_result, verbose_eval, learning_rates, callbacks)
206 begin_iteration=0,
207 end_iteration=num_boost_round,
--> 208 evaluation_result_list=evaluation_result_list))
209 except callback.EarlyStopException:
210 break

/Users/cchen1/anaconda/lib/python2.7/site-packages/lightgbm-0.1-py2.7.egg/lightgbm/callback.pyc in callback(env)
176 """internal function"""
177 if len(best_score) == 0:
--> 178 init(env)
179 for i in range(len(env.evaluation_result_list)):
180 score = env.evaluation_result_list[i][2] * factor_to_bigger_better[i]

/Users/cchen1/anaconda/lib/python2.7/site-packages/lightgbm-0.1-py2.7.egg/lightgbm/callback.pyc in init(env)
158
159 if len(env.evaluation_result_list) == 0:
--> 160 raise ValueError('For early stopping you need at least one set in evals.')
161
162 if verbose:

ValueError: For early stopping you need at least one set in evals.

[guolinke]

@wxchan I think we can add default metric when object is str and eval_metric is none

[wxchan]

@guolinke I have made this change and commited to #97 . You can check that PR now. I also updated the python examples to make it more clear.

Btw, I can't access to Wifi in where I lived these couples of days, so I may reply late some time.

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.