Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

MultiOutputClassifier should have `decision_function`

See original GitHub issue

Describe the workflow you want to enable

We should be able to use MultiOutputClassifier as (essentially) a drop-in replacement for the underlying classifiers. Right now it has the functions predict and predict_proba (if defined in the underlying predictors) but missing decision_function (e.g. for svm).

Describe your proposed solution

Use exactly the same code as predict_proba : https://github.com/scikit-learn/scikit-learn/blob/ec48b246785f37f3d976693c03dc5a3702ccedde/sklearn/multioutput.py#L450 but for decision_function

Issue Analytics

State:
Created 2 years ago
Reactions:1
Comments:6 (4 by maintainers)

Top GitHub Comments

1reaction

YannDubscommented, Dec 8, 2021

(A) sounds like the best option for now.

Full generality to avoid code duplication would likely be better in the long run but should probably be discussed in more depth by the core team as this will be a reoccuring issue and the solution should be standardized over all of sklearn.

0reactions

chrittercommented, Dec 8, 2021

If there are no preferences regarding the options, I will submit a PR with option A) as the approach will be in scope of the issue.

Top Results From Across the Web

1.12. Multiclass and multioutput algorithms - Scikit-learn

This section of the user guide covers functionality related to multi-learning problems, including multiclass, multilabel, and multioutput classification and ...

sklearndf.classification.MultiOutputClassifierDF

Compute the decision function for the given inputs. The inputs must have the same features as the inputs used to fit this learner....

Why do predictions and scores return different results in ...

For multilabel classification you should use y_pred_ = np.where(classifier.decision_function(X_test) > 0, 1, 0).

sklearn.multioutput — ibex latest documentation

Note that y doesn't need to contain all labels in `classes`. sample_weight ... def __init__(self, estimator, n_jobs=1): super(MultiOutputClassifier, self).

1.12. Multiclass and multilabel algorithms - 《Scikit-learn 0.22 ...

You don't need to use the sklearn.multiclass moduleunless you want ... the multioutputclassification task with different model formulations.