View source on GitHub |
Applies dropout both in training and serving, with variable dropout rate.
tf_agents.keras_layers.permanent_variable_rate_dropout.PermanentVariableRateDropout(
rate, permanent=False, **kwargs
)
Initialize this layer the same was as keras.layers.Dropout
, with two notable
differences:
--The parameter rate
can also be a callable.
--The extra boolean parameter permanent
. If set to true, dropout will be
applied both in training and inference.