lenskit.sklearn.nmf

Nonnegative matrix factorization for implicit feedback.

This module contains a non-negative factorization implicit-feedback scorer built on sklearn.decomposition.non_negative_factorization().

Classes

NMFConfig(*[, embedding_size, beta_loss, ...])	Configuration for NMFScorer.
NMFScorer([config])	Non-negative matrix factorization for implicit feedback using SciKit-Learn's sklearn.decomposition.non_negative_factorization().

class lenskit.sklearn.nmf.NMFConfig(*, embedding_size=None, beta_loss='frobenius', max_iter=200, alpha_W=0.0, alpha_H='same', l1_ratio=0.0)

Bases: EmbeddingSizeMixin, BaseModel

Configuration for NMFScorer. See the documentation for sklearn.decomposition.non_negative_factorization() for the configuration options.

Parameters:

• embedding_size (Annotated[int, Gt(gt=0)] | None)

• beta_loss (Literal['frobenius', 'kullback-leibler', 'itakura-saito'])

• max_iter (Annotated[int, Gt(gt=0)])

• alpha_W (float)

• alpha_H (float | Literal['same'])

• l1_ratio (float)

embedding_size: PositiveInt | None

The dimension of user and item embeddings (number of latent features to learn).

model_config: ClassVar[ConfigDict] = {'extra': 'forbid'}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

class lenskit.sklearn.nmf.NMFScorer(config=None, **kwargs)

Bases: Component[ItemList, …], Trainable

Non-negative matrix factorization for implicit feedback using SciKit-Learn’s sklearn.decomposition.non_negative_factorization(). It computes the user and item embedding matrices using an indicator matrix as the input.

Stability:

Caller (see Stability Levels).

Parameters:

• config (NMFConfig)

• kwargs (Any)

train(data, options=TrainingOptions(retrain=True, device=None, rng=None))

Train the model to learn its parameters from a training dataset.

Parameters:

• data (Dataset) – The training dataset.

• options (TrainingOptions) – The training options.