scikit-activeml 0.1.1

Our package scikit-activeml is a Python library for active learning on top of SciPy and scikit-learn.

scikit-activeml is a Python module for active learning on top of SciPy and scikit-learn. It is distributed under the 3-Clause BSD licence.

The project was initiated in 2020 by the Intelligent Embedded Systems Group at University Kassel.

Installation

The easiest way of installing scikit-activeml is using pip

pip install -U scikit-activeml

Example

The following code implements an active learning cycle with 20 iterations using a logistic regression classifier and uncertainty sampling. To use other classifiers, you can simply wrap classifiers from scikit-learn or use classifiers provided by scikit-activeml. Note that the main difficulty using active learning with scikit-learn is the ability to handle unlabeled data, which we denote as a specific value (MISSING_LABEL) in the label vector y. More query strategies can be found in the documentation.

import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.datasets import make_classification
from skactiveml.pool import UncertaintySampling
from skactiveml.utils import unlabeled_indices, MISSING_LABEL
from skactiveml.classifier import SklearnClassifier

# Generate data set.
X, y_true = make_classification(random_state=0)
y = np.full(shape=y_true.shape, fill_value=MISSING_LABEL)

# Create classifier and query strategy.
clf = SklearnClassifier(LogisticRegression(), classes=np.unique(y_true))
qs = UncertaintySampling(method='entropy')

# Execute active learning cycle.
n_cycles = 20
for c in range(n_cycles):
     clf.fit(X, y)
     unlbld_idx = unlabeled_indices(y)
     X_cand = X[unlbld_idx]
     query_idx = unlbld_idx[qs.query(X_cand=X_cand, clf=clf)]
     y[query_idx] = y_true[query_idx]
print(f'Accuracy: {clf.fit(X, y).score(X, y_true)}')

Development

More information are available in the Developer’s Guide.

Documentation

The documentation is available here: https://scikit-activeml.readthedocs.io