lenskit.flexmf.FlexMFNCFConfig ============================== .. py:class:: lenskit.flexmf.FlexMFNCFConfig(/, **data) :canonical: lenskit.flexmf._ncf.FlexMFNCFConfig Bases: :py:obj:`lenskit.flexmf._implicit.FlexMFImplicitConfig` Configuration for NCF (Neural Collaborative Filtering) Scorer. It inherits common training options and implicit-feedback settings from :class:`FlexMFImplicitConfig`. The inherited ``embedding_size`` field is used as the GMF embedding size. :Stability: Experimental .. py:attribute:: mlp_embedding_size :type: pydantic.PositiveInt :value: 8 The size of the MLP embedding space. .. py:attribute:: mlp_layer_sizes :type: list[pydantic.PositiveInt] The sizes of the MLP hidden layers. .. py:attribute:: negative_count :type: pydantic.PositiveInt :value: 4 The number of negative items to sample for each positive item. NCF defaults to 4, following the original NCF paper. .. py:attribute:: user_bias :type: bool | None :value: False Whether to learn a user bias term, added to the NCF output score. NCF defaults to ``False`` to match the original model, since the final prediction layer already has a global bias; set to ``True`` (or ``None`` for the implicit-feedback default) to enable it. .. py:attribute:: item_bias :type: bool :value: False Whether to learn an item bias term, added to the NCF output score. NCF defaults to ``False`` to match the original model.