ds-utils-transformers

Multi-label binarization integrates into scikit-learn pipelines using a wrapper transformer that exposes get_feature_names_out and outputs dense matrices. It wraps sklearn's MultiLabelBinarizer to ensure compatibility within standard pipeline and ColumnTransformer workflows.

Sentence embeddings can be used in a scikit-learn ColumnTransformer via a wrapper around sentence-transformers models. This wrapper supports lazy loading and NaN handling, allowing text features to be combined with numeric ones inside standard pipelines.

Combining text embeddings and multi-label features for classification is done by placing a binarizer transformer and an embedding transformer in a pipeline or ColumnTransformer. This setup feeds both binary label matrices and sentence embeddings directly into a classifier.

Scikit-learn's MultiLabelBinarizer does not natively support get_feature_names_out for pipelines. A custom wrapper is required to expose this method and ensure the transformer outputs dense matrices compatible with pipeline feature extraction.

Sentence embeddings handle missing values and lazy loading in sklearn pipelines through a dedicated sentence-transformers wrapper. It manages NaN values during transformation and defers model loading until needed, maintaining standard scikit-learn API compatibility.

Custom transformers for multi-label sets and text are needed in NLP pipelines when extracting machine-learning-ready features. They provide a consistent scikit-learn interface for generating binary label matrices and sentence embeddings required by downstream classifiers.