scikit-activeml

scikit-activeml implements active learning algorithms to select informative samples for labeling and improve model performance in data-scarce bioinformatics tasks such as genomics and proteomics.

Key Features:

• Foundation: Built on the SciPy and scikit-learn frameworks.
• Integration with Scikit-Learn: Integrates with scikit-learn classifiers, including Logistic Regression, to apply active learning to existing models.
• Handling Unlabeled Data: Represents unlabeled instances by a designated MISSING_LABEL value in the label vector y_true.
• Query Strategies: Implements a range of query strategies to identify the most informative data points for labeling.
• Customizable Active Learning Cycles: Supports iterative active learning cycles configurable with different classifiers.

Scientific Applications:

• Genomics: Reduce labeling effort for genomic datasets where experimental validation is costly by prioritizing informative samples.
• Proteomics: Prioritize labels in proteomic datasets to improve model accuracy with fewer labeled instances.
• Large-scale biological data analysis: Improve model performance under limited labeling resources in other large-scale bioinformatics analyses.

Methodology:

Uses uncertainty sampling to iteratively select and query labels for samples about which the model is least certain; employs active learning cycles, the MISSING_LABEL convention in y_true to mark unlabeled instances, and scikit-learn classifiers (e.g., Logistic Regression).