def __init__(self, vmin, vmax, num_atoms, dense_kwargs=None):
super().__init__()
if dense_kwargs is None:
dense_kwargs = models.default_dense_kwargs()
self.distributional_layer = tf.keras.layers.Dense(
num_atoms, **dense_kwargs)
self.values = tf.cast(tf.linspace(vmin, vmax, num_atoms), tf.float32)
def __init__(self,
loc_activation='tanh',
dense_loc_kwargs=None,
scale_activation='softplus',
scale_min=1e-4,
scale_max=1,
dense_scale_kwargs=None,
distribution=tfp.distributions.MultivariateNormalDiag):
super().__init__()
self.loc_activation = loc_activation
if dense_loc_kwargs is None:
dense_loc_kwargs = models.default_dense_kwargs()
self.dense_loc_kwargs = dense_loc_kwargs
self.scale_activation = scale_activation
self.scale_min = scale_min
self.scale_max = scale_max
if dense_scale_kwargs is None:
dense_scale_kwargs = models.default_dense_kwargs()
self.dense_scale_kwargs = dense_scale_kwargs
self.distribution = distribution
def __init__(self, activation='tanh', dense_kwargs=None):
super().__init__()
self.activation = activation
if dense_kwargs is None:
dense_kwargs = models.default_dense_kwargs()
self.dense_kwargs = dense_kwargs
def __init__(self, dense_kwargs=None):
super().__init__()
if dense_kwargs is None:
dense_kwargs = models.default_dense_kwargs()
self.v_layer = tf.keras.layers.Dense(1, **dense_kwargs)