def __init__(self,
input_shape,
output_dim,
name='CategoricalMLPRegressor',
hidden_sizes=(32, 32),
hidden_nonlinearity=tf.nn.tanh,
hidden_w_init=tf.initializers.glorot_uniform(),
hidden_b_init=tf.zeros_initializer(),
output_nonlinearity=tf.nn.softmax,
output_w_init=tf.initializers.glorot_uniform(),
output_b_init=tf.zeros_initializer(),
optimizer=None,
optimizer_args=None,
tr_optimizer=None,
tr_optimizer_args=None,
use_trust_region=True,
max_kl_step=0.01,
normalize_inputs=True,
layer_normalization=False):
super().__init__(input_shape, output_dim, name)
self._use_trust_region = use_trust_region
self._max_kl_step = max_kl_step
self._normalize_inputs = normalize_inputs
with tf.compat.v1.variable_scope(self._name, reuse=False) as vs:
self._variable_scope = vs
if optimizer_args is None:
optimizer_args = dict()
if tr_optimizer_args is None:
tr_optimizer_args = dict()
if optimizer is None:
self._optimizer = make_optimizer(LbfgsOptimizer,
**optimizer_args)
else:
self._optimizer = make_optimizer(optimizer, **optimizer_args)
if tr_optimizer is None:
self._tr_optimizer = make_optimizer(ConjugateGradientOptimizer,
**tr_optimizer_args)
else:
self._tr_optimizer = make_optimizer(tr_optimizer,
**tr_optimizer_args)
self._first_optimized = False
self.model = NormalizedInputMLPModel(
input_shape,
output_dim,
hidden_sizes=hidden_sizes,
hidden_nonlinearity=hidden_nonlinearity,
hidden_w_init=hidden_w_init,
hidden_b_init=hidden_b_init,
output_nonlinearity=output_nonlinearity,
output_w_init=output_w_init,
output_b_init=output_b_init,
layer_normalization=layer_normalization)
self._initialize()
def __init__(self,
input_shape,
output_dim,
name='ContinuousMLPRegressor',
hidden_sizes=(32, 32),
hidden_nonlinearity=tf.nn.tanh,
hidden_w_init=tf.glorot_uniform_initializer(),
hidden_b_init=tf.zeros_initializer(),
output_nonlinearity=None,
output_w_init=tf.glorot_uniform_initializer(),
output_b_init=tf.zeros_initializer(),
optimizer=None,
optimizer_args=None,
normalize_inputs=True):
super().__init__(input_shape, output_dim, name)
self._normalize_inputs = normalize_inputs
with tf.compat.v1.variable_scope(self._name, reuse=False) as vs:
self._variable_scope = vs
if optimizer_args is None:
optimizer_args = dict()
if optimizer is None:
optimizer = LbfgsOptimizer(**optimizer_args)
else:
optimizer = optimizer(**optimizer_args)
self._optimizer = optimizer
self.model = NormalizedInputMLPModel(
input_shape=input_shape,
output_dim=output_dim,
hidden_sizes=hidden_sizes,
hidden_nonlinearity=hidden_nonlinearity,
hidden_w_init=hidden_w_init,
hidden_b_init=hidden_b_init,
output_nonlinearity=output_nonlinearity,
output_w_init=output_w_init,
output_b_init=output_b_init)
self._initialize()