- fixed bug if both eval_metrics xgb-param and
metrics param of cv function have been set
- cv early stopping output looks now like the one of xgb.train
- ensures same behavior for verbose_eval=0 and verbose_eval=False
- fix printing last eval message if early_stopping_rounds is set, but xgb
runs to the end
- best_ntree_limit as new booster atrribute added
- usage of bst.best_ntree_limit in python doc added
- fixed wrong 'best_iteration' after training continuation
- allows feval to return a list of tuples (name, error/score value)
- changed behavior for multiple eval_metrics in conjunction with
early_stopping: Instead of raising an error, the last passed evel_metric
(or last entry in return value of feval) is used for early stopping
- allows list of eval_metrics in dict-typed params
- unittest for new features / behavior
documentation updated
- example for assigning a list to 'eval_metric'
- note about early stopping on last passed eval metric
- info msg for used eval metric added
Made changes to training.py to make sure all eval_metric information get passed to evals_result. Previous version lost and mislabeled data in evals_result when using more than one eval_metric.
Structure of eval_metric is now:
eval_metric[evals][eval_metric] = list of metrics
Example:
>>> dtrain = xgb.DMatrix('agaricus.txt.train', silent=True)
>>> dtest = xgb.DMatrix('agaricus.txt.test', silent=True)
>>> param = [('max_depth', 2), ('objective', 'binary:logistic'), ('bst:eta', 0.01), ('eval_metric', 'logloss'), ('eval_metric', 'error')]
>>> watchlist = [(dtest,'eval'), (dtrain,'train')]
>>> num_round = 3
>>> evals_result = {}
>>> bst = xgb.train(param, dtrain, num_round, watchlist, evals_result=evals_result)
>>> print(evals_result['eval']['logloss'])
>>> print(evals_result)
Prints:
['0.684877', '0.676767', '0.668817']
{'train': {'logloss': ['0.684954', '0.676917', '0.669036'], 'error': ['0.04652', '0.04652', '0.04652']}, 'eval': {'logloss': ['0.684877', '0.676767', '0.668817'], 'error': ['0.042831', '0.042831', '0.042831']}}
