Add predictor.predict_multi and predictor.predict_proba_multi #2727
Conversation
|
Job PR-2727-bd4488f is done. |
|
Shall we just extend |
That is possible, but it would add a significant amount of complexity to the doc of In general most users wouldn't need to use |
|
Another thought: A different name could be |
| # Using > instead of >= to align with Pandas `.idxmax` logic which picks the left-most column during ties. | ||
| # If this is not done, then predictions can be inconsistent when converting in binary classification from multiclass-form pred_proba and | ||
| # binary-form pred_proba when the pred_proba is 0.5 for positive and negative classes. | ||
| y_pred = [1 if pred > 0.5 else 0 for pred in y_pred_proba] |
There was a problem hiding this comment.
Shall we move 0.5 into a property or method method call? Rationale: at some point we'll add optimization of the threshold so having either method or named property would be a natural place to change instead of looking for magic value here.
There was a problem hiding this comment.
Yes, but this should be a separate PR as it is a major change. I've had this in mind for awhile
| @@ -1491,7 +1491,7 @@ def evaluate_predictions(self, y_true, y_pred, sample_weight=None, silent=False, | |||
| return self._learner.evaluate_predictions(y_true=y_true, y_pred=y_pred, sample_weight=sample_weight, silent=silent, | |||
| auxiliary_metrics=auxiliary_metrics, detailed_report=detailed_report) | |||
|
|
|||
| def leaderboard(self, data=None, extra_info=False, extra_metrics=None, only_pareto_frontier=False, skip_score=False, silent=False): | |||
| def leaderboard(self, data=None, extra_info=False, extra_metrics=None, only_pareto_frontier=False, skip_score=False, silent=False) -> pd.DataFrame: | |||
There was a problem hiding this comment.
nit: add parameter types too
There was a problem hiding this comment.
added, note type hint for data will wait for future PR due to multiple types support
| transform_features=transform_features, | ||
| inverse_transform=inverse_transform) | ||
|
|
||
| def predict_dict(self, data=None, models=None, as_pandas=True, transform_features=True, inverse_transform=True) -> dict: |
There was a problem hiding this comment.
nit: ... -> Dict[str, DataFrame] + add parameter types
There was a problem hiding this comment.
added, note type hint for data and output will wait for future PR due to multiple types support
| X: pd.DataFrame, | ||
| models: List[str], | ||
| record_pred_time: bool = False, | ||
| **kwargs): |
There was a problem hiding this comment.
will wait for future PR, the return type is not trivial as it can be a tuple
Innixma
force-pushed
the
zero_shot_hpo
branch
from
bd4488f
to
0caa888
Compare
Innixma
changed the title
|
Job PR-2727-0caa888 is done. |
|
Note: renamed to |
Issue #, if available:
Description of changes:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.