scikit-multilearn

0.2.0 verified Mon Apr 27 auth: no python maintenance

A BSD-licensed library for multi-label classification built on top of scikit-learn. Current version is 0.2.0. The project appears to be in maintenance mode with no recent releases; last PyPI release was in 2018.

pip install scikit-multilearn

Common errors

error ModuleNotFoundError: No module named 'skmultilearn' ↓

cause Trying to import the module before installing or using wrong name.

fix

Run 'pip install scikit-multilearn' then use 'import skmultilearn'.

error TypeError: init() got an unexpected keyword argument 'require_dense' ↓

cause Using an older version of scikit-multilearn (<0.2.0) where require_dense was not introduced.

fix

Upgrade to 0.2.0: 'pip install scikit-multilearn==0.2.0'.

error ValueError: Input data is not a dense numpy array or a sparse scipy array ↓

cause passing a list or other format where dense array expected.

fix

Convert input to numpy array: 'X = np.array(X)'.

Warnings

deprecated The library has not been updated since 2018; consider using scikit-multilearn v2 (if available) or alternatives like 'skmultilearn' fork. ↓

fix Check for newer forks or use 'pip install scikit-multilearn==0.2.0' (still old).

gotcha Many methods require `require_dense` parameter to be set to [True, True] for classifiers that expect dense arrays, otherwise confusing errors occur. ↓

fix Always pass `require_dense=[True, True]` to adapters like BinaryRelevance when using sklearn classifiers.

gotcha The module name in imports is 'skmultilearn', not 'scikit_multilearn' or 'sklearn_multilearn'. ↓

fix Use 'import skmultilearn' or 'from skmultilearn import ...'.

breaking Version 0.2.0 dropped Python 2 support; Python 3.5+ required. ↓

fix Upgrade to Python 3.5 or later.

Imports

BinaryRelevance
```
from skmultilearn.problem_transform import BinaryRelevance
```
Common mistake: importing from 'sklearn.multilearn' or 'scikit_multilearn'.

LabelPowerset

wrong

from skmultilearn import LabelPowerset

correct

from skmultilearn.problem_transform import LabelPowerset

LabelPowerset is in the problem_transform submodule, not top-level.

ClassifierChain

wrong

from skmultilearn.classifier_chain import ClassifierChain

correct

from skmultilearn.problem_transform import ClassifierChain

Old import path; use 'problem_transform'.

LabelSpaceClustererBase

wrong

from skmultilearn.cluster.base import LabelSpaceClustererBase

correct

from skmultilearn.cluster import LabelSpaceClustererBase

Internal path changed; use 'cluster' directly.

Quickstart

Quick example using BinaryRelevance with a RandomForest base classifier.

import numpy as np
from skmultilearn.problem_transform import BinaryRelevance
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_multilabel_classification

X, Y = make_multilabel_classification(n_samples=100, n_features=20, n_classes=5, random_state=42)
classifier = BinaryRelevance(classifier=RandomForestClassifier(), require_dense=[True, True])
classifier.fit(X, Y)
predictions = classifier.predict(X)
print(predictions.shape)