def apply_OutputNN(inputs, activation_type):
activation_fun = nn_utils.get_activation_fun(activation_type)
with tf.variable_scope('hidden1') as scope:
weights = tf.get_variable("weights")
biases = tf.get_variable("biases")
hidden1 = activation_fun(tf.matmul(inputs, weights) + biases)
with tf.variable_scope('output') as scope:
weights = tf.get_variable("weights")
return tf.matmul(hidden1, weights)
def apply_OutputNN(inputs, hidden_activation_type, output_activation_type):
'''
output_activation_type:
e.g. use None (==linear) for regression problems
'''
hidden_activation_type = nn_utils.get_activation_fun(hidden_activation_type)
output_activation_type = nn_utils.get_activation_fun(output_activation_type)
with tf.variable_scope('hidden1') as scope:
weights = tf.get_variable("weights")
biases = tf.get_variable("biases")
hidden1 = hidden_activation_type(tf.matmul(inputs, weights) + biases)
with tf.variable_scope('output') as scope:
weights = tf.get_variable("weights")
out = tf.matmul(hidden1, weights)
if output_activation_type is not None:
return output_activation_type(out)
return out