def make_encoder(self, state, z_size, scope, n_layers, hid_size):
"""
### PROBLEM 3
### YOUR CODE HERE
args:
state: tf variable
z_size: output dimension of the encoder network
scope: scope name
n_layers: number of layers of the encoder network
hid_size: hidden dimension of encoder network
TODO:
1. z_mean: the output of a neural network that takes the state as input,
has output dimension z_size, n_layers layers, and hidden
dimension hid_size
2. z_logstd: a trainable variable, initialized to 0
shape (z_size,)
Hint: use build_mlp
"""
z_mean = build_mlp(state, z_size, scope, n_layers, hid_size)
z_logstd = tf.Variable(tf.zeros(z_size))
return tfp.distributions.MultivariateNormalDiag(
loc=z_mean, scale_diag=tf.exp(z_logstd))
def make_discriminator(self, z, output_size, scope, n_layers, hid_size):
"""
### PROBLEM 3
### YOUR CODE HERE
args:
z: input to to discriminator network
output_size: output dimension of discriminator network
scope: scope name
n_layers: number of layers of discriminator network
hid_size: hidden dimension of discriminator network
TODO:
1. logit: the output of a neural network that takes z as input,
has output size output_size, n_layers layers, and hidden
dimension hid_size
Hint: use build_mlp
"""
logit = build_mlp(z,
output_size,
scope,
n_layers=n_layers,
size=hid_size)
return tfp.distributions.Bernoulli(logit)
def make_discriminator(self, z, output_size, scope, n_layers, hid_size):
"""
args:
z: input to to discriminator network
output_size: output dimension of discriminator network
scope: scope name
n_layers: number of layers of discriminator network
hid_size: hidden dimension of discriminator network
TODO:
1. logit: the output of a neural network that takes z as input,
has output size output_size, n_layers layers, and hidden
dimension hid_size
Hint: use build_mlp
"""
logit = build_mlp(
input_placeholder=z,
output_size=output_size,
scope=scope,
n_layers=n_layers,
size=hid_size,
activation=tf.tanh,
output_activation=None
)
return tfp.distributions.Bernoulli(logit)
def make_encoder(self, state, z_size, scope, n_layers, hid_size):
"""
### PROBLEM 3
### YOUR CODE HERE
args:
state: tf variable
z_size: output dimension of the encoder network
scope: scope name
n_layers: number of layers of the encoder network
hid_size: hidden dimension of encoder network
TODO:
1. z_mean: the output of a neural network that takes the state as input,
has output dimension z_size, n_layers layers, and hidden
dimension hid_size
2. z_logstd: a trainable variable, initialized to 0
shape (z_size,)
Hint: use build_mlp
"""
z_mean = build_mlp(input_placeholder=state,
output_size=z_size,
scope=scope,
n_layers=n_layers,
size=hid_size)
z_logstd = tf.get_variable('log_std',
shape=[z_size],
dtype=tf.float32,
initializer=tf.constant_initializer(0),
trainable=True)
return tfp.distributions.MultivariateNormalDiag(
loc=z_mean, scale_diag=tf.exp(z_logstd))
def make_encoder(self, state, z_size, scope, n_layers, hid_size):
"""
args:
state: tf variable
z_size: output dimension of the encoder network
scope: scope name
n_layers: number of layers of the encoder network
hid_size: hidden dimension of encoder network
TODO:
1. z_mean: the output of a neural network that takes the state as input,
has output dimension z_size, n_layers layers, and hidden
dimension hid_size
2. z_logstd: a trainable variable, initialized to 0
shape (z_size,)
Hint: use build_mlp
"""
z_mean = build_mlp(
input_placeholder=state,
output_size=z_size,
scope=scope,
n_layers=n_layers,
size=hid_size,
activation=tf.tanh,
output_activation=None
)
z_logstd = tf.Variable(initial_value=np.zeros(shape=(int(z_size), ), dtype=np.float32), trainable=True)
return tfp.distributions.MultivariateNormalDiag(loc=z_mean, scale_diag=tf.exp(z_logstd))
