dropout¶
- probly.transformation.dropout(base: T, p: float = 0.25, rng_collection: str = 'dropout', rngs: Rngs | RngStream | int = 1, shared_mask: bool = False) T[source]¶
Create a Dropout predictor from a base predictor based on [GG16].
- Parameters:
base – The base model to be used for dropout.
p – The probability of dropping out a neuron. Default is 0.25.
rng_collection – Optional rng collection name for flax layer initialization. Default is “dropout”.
rngs – Optional rngs for flax layer initialization. Default is 1.
shared_mask – If True, insert shared-mask dropout layers that draw one mask per forward pass shared across the batch (torch backend only). Shared masking applies only to the dropout layers this transform inserts; any pre-existing dropout keeps its standard per-element masks. Default is False.
- Returns:
The DropOut predictor.
