Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[dask, sklearn] Fix predict proba. (#6566)

Browse Source 
* For sklearn:
  - Handles user defined objective function.
  - Handles `softmax`.

* For dask:
  - Use the implementation from sklearn, the previous implementation doesn't perform any extra handling.
This commit is contained in:
Jiaming Yuan
2021-01-05 08:29:06 +08:00
committed by GitHub
parent 516a93d25c
commit 60cfd14349
5 changed files with 74 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python-package/xgboost/dask.py
View File

@@ -40,6 +40,7 @@ from .training import train as worker_train
from .tracker import RabitTracker, get_host_ip
from .sklearn import XGBModel, XGBRegressorBase, XGBClassifierBase, _objective_decorator
from .sklearn import xgboost_model_doc
from .sklearn import _cls_predict_proba
if TYPE_CHECKING:
@@ -1504,6 +1505,10 @@ class DaskXGBClassifier(DaskScikitLearnBase, XGBClassifierBase):
early_stopping_rounds=early_stopping_rounds,
callbacks=callbacks)
self._Booster = results['booster']
if not callable(self.objective):
self.objective = params["objective"]
# pylint: disable=attribute-defined-outside-init
self.evals_result_ = results['history']
return self
@@ -1554,7 +1559,7 @@ class DaskXGBClassifier(DaskScikitLearnBase, XGBClassifierBase):
data=test_dmatrix,
validate_features=validate_features,
output_margin=output_margin)
return pred_probs
return _cls_predict_proba(self.objective, pred_probs, da.vstack)
# pylint: disable=arguments-differ,missing-docstring
def predict_proba(
@@ -1593,6 +1598,8 @@ class DaskXGBClassifier(DaskScikitLearnBase, XGBClassifierBase):
output_margin=output_margin,
validate_features=validate_features
)
if output_margin:
return pred_probs
if self.n_classes_ == 2:
preds = (pred_probs > 0.5).astype(int)

27
python-package/xgboost/sklearn.py
View File

@@ -819,6 +819,20 @@ class XGBModel(XGBModelBase):
return np.array(json.loads(b.get_dump(dump_format='json')[0])['bias'])
def _cls_predict_proba(objective: Union[str, Callable], prediction: Any, vstack: Callable) -> Any:
if objective == 'multi:softmax':
raise ValueError('multi:softmax objective does not support predict_proba,'
' use `multi:softprob` or `binary:logistic` instead.')
if objective == 'multi:softprob' or callable(objective):
# Return prediction directly if if objective is defined by user since we don't
# know how to perform the transformation
return prediction
# Lastly the binary logistic function
classone_probs = prediction
classzero_probs = 1.0 - classone_probs
return vstack((classzero_probs, classone_probs)).transpose()
@xgboost_model_doc(
"Implementation of the scikit-learn API for XGBoost classification.",
['model', 'objective'], extra_parameters='''
@@ -929,7 +943,9 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
verbose_eval=verbose, xgb_model=model,
callbacks=callbacks)
self.objective = params["objective"]
if not callable(self.objective):
self.objective = params["objective"]
if evals_result:
for val in evals_result.items():
evals_result_key = list(val[1].keys())[0]
@@ -1031,7 +1047,8 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
Returns
-------
prediction : numpy array
a numpy array with the probability of each data example being of a given class.
a numpy array of shape array-like of shape (n_samples, n_classes) with the
probability of each data example being of a given class.
"""
test_dmatrix = DMatrix(X, base_margin=base_margin,
missing=self.missing, nthread=self.n_jobs)
@@ -1040,11 +1057,7 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
class_probs = self.get_booster().predict(test_dmatrix,
ntree_limit=ntree_limit,
validate_features=validate_features)
if self.objective == "multi:softprob":
return class_probs
classone_probs = class_probs
classzero_probs = 1.0 - classone_probs
return np.vstack((classzero_probs, classone_probs)).transpose()
return _cls_predict_proba(self.objective, class_probs, np.vstack)
def evals_result(self):
"""Return the evaluation results.