def test_different_mlp(self):
# We create another mlp with different name
with tf.variable_scope("MLP"):
self.mlp_different_copy = mlp(
input_var=self._input,
output_dim=self._output_shape,
hidden_sizes=(32, 32),
hidden_nonlinearity=self.hidden_nonlinearity,
name="mlp2")
# Initialize the new mlp variables
self.sess.run(tf.global_variables_initializer())
# We modify the weight of the default mlp and feed
# The another mlp created should output different result
with tf.variable_scope("MLP", reuse=True):
w = tf.get_variable("mlp1/hidden_0/kernel")
self.sess.run(w.assign(w + 1))
mlp_output = self.sess.run(
self.mlp_f, feed_dict={self._input: self.obs_input})
mlp_output2 = self.sess.run(
self.mlp_different_copy,
feed_dict={self._input: self.obs_input})
np.not_equal(mlp_output, mlp_output2)
def test_layer_normalization(self):
# Create a mlp with layer normalization
with tf.variable_scope("MLP"):
self.mlp_f_w_n = mlp(
input_var=self._input,
output_dim=self._output_shape,
hidden_sizes=(32, 32),
hidden_nonlinearity=self.hidden_nonlinearity,
name="mlp2",
layer_normalization=True)
# Initialize the new mlp variables
self.sess.run(tf.global_variables_initializer())
with tf.variable_scope("MLP", reuse=True):
h1_w = tf.get_variable("mlp2/hidden_0/kernel")
h1_b = tf.get_variable("mlp2/hidden_0/bias")
h2_w = tf.get_variable("mlp2/hidden_1/kernel")
h2_b = tf.get_variable("mlp2/hidden_1/bias")
out_w = tf.get_variable("mlp2/output/kernel")
out_b = tf.get_variable("mlp2/output/bias")
beta_1 = tf.get_variable("mlp2/LayerNorm/beta")
gamma_1 = tf.get_variable("mlp2/LayerNorm/gamma")
beta_2 = tf.get_variable("mlp2/LayerNorm_1/beta")
gamma_2 = tf.get_variable("mlp2/LayerNorm_1/gamma")
# First layer
y = tf.matmul(self._input, h1_w) + h1_b
y = self.hidden_nonlinearity(y)
mean, variance = tf.nn.moments(y, [1], keep_dims=True)
normalized_y = (y - mean) / tf.sqrt(variance + 1e-12)
y_out = normalized_y * gamma_1 + beta_1
# Second layer
y = tf.matmul(y_out, h2_w) + h2_b
y = self.hidden_nonlinearity(y)
mean, variance = tf.nn.moments(y, [1], keep_dims=True)
normalized_y = (y - mean) / tf.sqrt(variance + 1e-12)
y_out = normalized_y * gamma_2 + beta_2
# Output layer
y = tf.matmul(y_out, out_w) + out_b
out = self.sess.run(y, feed_dict={self._input: self.obs_input})
mlp_output = self.sess.run(
self.mlp_f_w_n, feed_dict={self._input: self.obs_input})
np.testing.assert_array_almost_equal(out, mlp_output)
def setUp(self):
super(TestNetworks, self).setUp()
self.obs_input = np.array([[1, 2, 3, 4]])
input_shape = self.obs_input.shape[1] # 4
self.hidden_nonlinearity = tf.nn.relu
self._input = tf.placeholder(
tf.float32, shape=(None, input_shape), name="input")
self._output_shape = 2
# We build a default mlp
with tf.variable_scope("MLP"):
self.mlp_f = mlp(
input_var=self._input,
output_dim=self._output_shape,
hidden_sizes=(32, 32),
hidden_nonlinearity=self.hidden_nonlinearity,
name="mlp1")
self.sess.run(tf.global_variables_initializer())
def test_multiple_same_mlp(self):
# We create another mlp with the same name, trying to reuse it
with tf.variable_scope("MLP", reuse=True):
self.mlp_same_copy = mlp(
input_var=self._input,
output_dim=self._output_shape,
hidden_sizes=(32, 32),
hidden_nonlinearity=self.hidden_nonlinearity,
name="mlp1")
# We modify the weight of the default mlp and feed
# The another mlp created should output the same result
with tf.variable_scope("MLP", reuse=True):
w = tf.get_variable("mlp1/hidden_0/kernel")
self.sess.run(w.assign(w + 1))
mlp_output = self.sess.run(
self.mlp_f, feed_dict={self._input: self.obs_input})
mlp_output2 = self.sess.run(
self.mlp_same_copy, feed_dict={self._input: self.obs_input})
np.testing.assert_array_almost_equal(mlp_output, mlp_output2)