def __init__(self, config, name=None):
super(LNormPolicy, self).__init__(name=name or self.__class__.__name__)
self._config = config
hidden_widths = config.hidden_widths
self.p = config.p
self.q = config.q
if config.embed:
transformation_layers = [layers_lib.soft_hot_layer(**config.embed)]
else:
transformation_layers = []
self._body = tf.keras.Sequential(transformation_layers + [
tf.keras.layers.Dense(w, activation='relu') for w in hidden_widths
] + [tf.keras.layers.Dense(1, activation=None)])
def __init__(self, config, name=None):
super(ExponentialFamilyPolicy,
self).__init__(name=name or self.__class__.__name__)
self._config = config
hidden_widths = config.hidden_widths
self._dist = DISTS[config.dist_name]
params = self._dist.params
self._log_pdf = self._dist.log_pdf
self._mode = self._dist.mode
if config.embed:
transformation_layers = [layers_lib.soft_hot_layer(**config.embed)]
else:
transformation_layers = []
self._body = tf.keras.Sequential(transformation_layers + [
tf.keras.layers.Dense(w, activation='relu') for w in hidden_widths
] + [
tf.keras.layers.Dense(len(params), activation=None),
tf.keras.layers.Lambda(lambda x: tf.stack( # pylint: disable=g-long-lambda
[
tf.math.softplus(x[Ellipsis, i]) if is_non_neg else x[
Ellipsis, i] for i, is_non_neg in enumerate(params)
], -1))
])