def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
kernel_size=3,
n_attn_heads=16,
dropout=0.9,
):
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
encoder_embedded = tf.layers.dense(self.X, size_layer)
encoder_embedded += position_encoding(encoder_embedded)
e = tf.identity(encoder_embedded)
for i in range(num_layers):
dilation_rate = 2**i
pad_sz = (kernel_size - 1) * dilation_rate
with tf.variable_scope('block_%d' % i):
encoder_embedded += cnn_block(encoder_embedded, dilation_rate,
pad_sz, size_layer, kernel_size)
encoder_output, output_memory = encoder_embedded, encoder_embedded + e
g = tf.identity(encoder_embedded)
for i in range(num_layers):
dilation_rate = 2**i
pad_sz = (kernel_size - 1) * dilation_rate
with tf.variable_scope('decode_%d' % i):
attn_res = h = cnn_block(encoder_embedded, dilation_rate,
pad_sz, size_layer, kernel_size)
C = []
for j in range(n_attn_heads):
h_ = tf.layers.dense(h, size_layer // n_attn_heads)
g_ = tf.layers.dense(g, size_layer // n_attn_heads)
zu_ = tf.layers.dense(encoder_output,
size_layer // n_attn_heads)
ze_ = tf.layers.dense(output_memory,
size_layer // n_attn_heads)
d = tf.layers.dense(h_, size_layer // n_attn_heads) + g_
dz = tf.matmul(d, tf.transpose(zu_, [0, 2, 1]))
a = tf.nn.softmax(dz)
c_ = tf.matmul(a, ze_)
C.append(c_)
c = tf.concat(C, 2)
h = tf.layers.dense(attn_res + c, size_layer)
h = tf.nn.dropout(h, keep_prob=dropout)
encoder_embedded += h
encoder_embedded = tf.sigmoid(encoder_embedded[-1])
self.logits = tf.layers.dense(encoder_embedded, output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
def __init__(self, name, input_size, output_size, size_layer,
learning_rate):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
self.hidden_layer = tf.placeholder(tf.float32,
(None, 2 * size_layer))
self.REWARD = tf.placeholder(tf.float32, (None, 1))
feed_critic = tf.layers.dense(self.X,
size_layer,
activation=tf.nn.relu)
cell = tf.nn.rnn_cell.LSTMCell(size_layer, state_is_tuple=False)
self.rnn, self.last_state = tf.nn.dynamic_rnn(
inputs=self.X,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
feed_critic = tf.layers.dense(
self.rnn[:, -1], output_size, activation=tf.nn.relu) + self.Y
feed_critic = tf.layers.dense(feed_critic,
size_layer // 2,
activation=tf.nn.relu)
self.logits = tf.layers.dense(feed_critic, 1)
self.cost = tf.reduce_mean(tf.square(self.REWARD - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
def __init__(self,
learning_rate,
num_layers,
size,
size_layer,
forget_bias=0.8):
def lstm_cell(size_layer):
return tf.nn.rnn_cell.LSTMCell(size_layer, state_is_tuple=False)
rnn_cells = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple=False)
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, size))
drop = tf.contrib.rnn.DropoutWrapper(rnn_cells,
output_keep_prob=forget_bias)
self.hidden_layer = tf.placeholder(tf.float32,
(None, num_layers * 2 * size_layer))
self.outputs, self.last_state = tf.nn.dynamic_rnn(
drop, self.X, initial_state=self.hidden_layer, dtype=tf.float32)
self.logits = tf.layers.dense(
self.outputs[-1],
size,
kernel_initializer=tf.glorot_uniform_initializer())
self.cost = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(labels=self.Y,
logits=self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
def __init__(self, state_size, window_size, trend, skip):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.skip = skip
tf.reset_default_graph()
self.INITIAL_FEATURES = np.zeros((4, self.state_size))
self.X = tf.placeholder(tf.float32, (None, None, self.state_size))
self.Y = tf.placeholder(tf.float32, (None, self.OUTPUT_SIZE))
cell = tf.nn.rnn_cell.LSTMCell(self.LAYER_SIZE, state_is_tuple=False)
self.hidden_layer = tf.placeholder(tf.float32,
(None, 2 * self.LAYER_SIZE))
self.rnn, self.last_state = tf.nn.dynamic_rnn(
inputs=self.X,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
tensor_action, tensor_validation = tf.split(self.rnn[:, -1], 2, 1)
feed_action = tf.layers.dense(tensor_action, self.OUTPUT_SIZE)
feed_validation = tf.layers.dense(tensor_validation, 1)
self.logits = feed_validation + tf.subtract(
feed_action, tf.reduce_mean(feed_action, axis=1, keep_dims=True))
self.cost = tf.reduce_sum(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(
learning_rate=self.LEARNING_RATE).minimize(self.cost)
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
def __init__(self, state_size, window_size, trend, skip):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.skip = skip
self.X = tf.placeholder(tf.float32, (None, self.state_size))
self.REWARDS = tf.placeholder(tf.float32, (None))
self.ACTIONS = tf.placeholder(tf.int32, (None))
feed_forward = tf.layers.dense(self.X,
self.LAYER_SIZE,
activation=tf.nn.relu)
self.logits = tf.layers.dense(feed_forward,
self.OUTPUT_SIZE,
activation=tf.nn.softmax)
input_y = tf.one_hot(self.ACTIONS, self.OUTPUT_SIZE)
loglike = tf.log((input_y * (input_y - self.logits) + (1 - input_y) *
(input_y + self.logits)) + 1)
rewards = tf.tile(tf.reshape(self.REWARDS, (-1, 1)),
[1, self.OUTPUT_SIZE])
self.cost = -tf.reduce_mean(loglike * (rewards + 1))
self.optimizer = tf.train.AdamOptimizer(
learning_rate=self.LEARNING_RATE).minimize(self.cost)
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
forget_bias = 0.1,
):
def lstm_cell(size_layer):
return tf.nn.rnn_cell.GRUCell(size_layer)
rnn_cells = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple = False,
)
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
drop = tf.contrib.rnn.DropoutWrapper(
rnn_cells, output_keep_prob = forget_bias
)
self.hidden_layer = tf.placeholder(
tf.float32, (None, num_layers * size_layer)
)
self.outputs, self.last_state = tf.nn.dynamic_rnn(
drop, self.X, initial_state = self.hidden_layer, dtype = tf.float32
)
self.logits = tf.layers.dense(self.outputs[-1], output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost
)
def __init__(self, state_size, window_size, trend, skip, batch_size):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.skip = skip
self.action_size = 3
self.batch_size = batch_size
self.memory = deque(maxlen=1000)
self.inventory = []
self.gamma = 0.95
self.epsilon = 0.5
self.epsilon_min = 0.01
self.epsilon_decay = 0.999
tf.reset_default_graph()
self.sess = tf.InteractiveSession()
self.X = tf.placeholder(tf.float32, [None, self.state_size])
self.Y = tf.placeholder(tf.float32, [None, self.action_size])
feed = tf.layers.dense(self.X, 512, activation=tf.nn.relu)
tensor_action, tensor_validation = tf.split(feed, 2, 1)
feed_action = tf.layers.dense(tensor_action, self.action_size)
feed_validation = tf.layers.dense(tensor_validation, 1)
self.logits = feed_validation + tf.subtract(
feed_action, tf.reduce_mean(feed_action, axis=1, keep_dims=True))
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.GradientDescentOptimizer(1e-5).minimize(
self.cost)
self.sess.run(tf.global_variables_initializer())
def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
forget_bias = 0.1,
lambda_coeff = 0.5
):
def lstm_cell(size_layer):
return tf.nn.rnn_cell.GRUCell(size_layer)
rnn_cells = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple = False,
)
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
drop = tf.contrib.rnn.DropoutWrapper(
rnn_cells, output_keep_prob = forget_bias
)
self.hidden_layer = tf.placeholder(
tf.float32, (None, num_layers * size_layer)
)
_, last_state = tf.nn.dynamic_rnn(
drop, self.X, initial_state = self.hidden_layer, dtype = tf.float32
)
self.z_mean = tf.layers.dense(last_state, size)
self.z_log_sigma = tf.layers.dense(last_state, size)
epsilon = tf.random_normal(tf.shape(self.z_log_sigma))
self.z_vector = self.z_mean + tf.exp(self.z_log_sigma)
with tf.variable_scope('decoder', reuse = False):
rnn_cells_dec = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)], state_is_tuple = False
)
drop_dec = tf.contrib.rnn.DropoutWrapper(
rnn_cells_dec, output_keep_prob = forget_bias
)
x = tf.concat([tf.expand_dims(self.z_vector, axis=0), self.X], axis = 1)
self.outputs, self.last_state = tf.nn.dynamic_rnn(
drop_dec, self.X, initial_state = last_state, dtype = tf.float32
)
self.logits = tf.layers.dense(self.outputs[-1], output_size)
self.lambda_coeff = lambda_coeff
self.kl_loss = -0.5 * tf.reduce_sum(1.0 + 2 * self.z_log_sigma - self.z_mean ** 2 -
tf.exp(2 * self.z_log_sigma), 1)
self.kl_loss = tf.scalar_mul(self.lambda_coeff, self.kl_loss)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits) + self.kl_loss)
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost
)
def __init__(self, input_size, output_size, layer_size, learning_rate):
self.X = tf.placeholder(tf.float32, (None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
feed_forward = tf.layers.dense(self.X,
layer_size,
activation=tf.nn.relu)
self.logits = tf.layers.dense(feed_forward, output_size)
self.cost = tf.reduce_sum(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(
learning_rate=learning_rate).minimize(self.cost)
def __init__(self, name, input_size, output_size, size_layer, learning_rate):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
self.REWARD = tf.placeholder(tf.float32, (None, 1))
feed_critic = tf.layers.dense(self.X, size_layer, activation = tf.nn.relu)
feed_critic = tf.layers.dense(feed_critic, output_size, activation = tf.nn.relu) + self.Y
feed_critic = tf.layers.dense(feed_critic, size_layer//2, activation = tf.nn.relu)
self.logits = tf.layers.dense(feed_critic, 1)
self.cost = tf.reduce_mean(tf.square(self.REWARD - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(self.cost)
def __init__(self, input_size, output_size, layer_size, learning_rate):
self.X = tf.placeholder(tf.float32, (None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
feed = tf.layers.dense(self.X, layer_size, activation=tf.nn.relu)
tensor_action, tensor_validation = tf.split(feed, 2, 1)
feed_action = tf.layers.dense(tensor_action, output_size)
feed_validation = tf.layers.dense(tensor_validation, 1)
self.logits = feed_validation + tf.subtract(
feed_action, tf.reduce_mean(feed_action, axis=1, keep_dims=True))
self.cost = tf.reduce_sum(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(
learning_rate=learning_rate).minimize(self.cost)
def __init__(self, input_size, output_size, layer_size, learning_rate, name):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
cell = tf.nn.rnn_cell.LSTMCell(layer_size, state_is_tuple = False)
self.hidden_layer = tf.placeholder(tf.float32, (None, 2 * layer_size))
self.rnn,self.last_state = tf.nn.dynamic_rnn(inputs=self.X,cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
self.logits = tf.layers.dense(self.rnn[:,-1], output_size)
self.cost = tf.reduce_sum(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate = learning_rate).minimize(self.cost)
def __init__(self, name, input_size, output_size, size_layer):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, None, input_size))
self.hidden_layer = tf.placeholder(tf.float32,
(None, 2 * size_layer))
cell = tf.nn.rnn_cell.LSTMCell(size_layer, state_is_tuple=False)
self.rnn, self.last_state = tf.nn.dynamic_rnn(
inputs=self.X,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
self.logits = tf.layers.dense(self.rnn[:, -1], output_size)
def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
forget_bias=0.1,
):
def lstm_cell(size_layer):
return tf.nn.rnn_cell.BasicRNNCell(size_layer)
with tf.variable_scope('forward', reuse=False):
rnn_cells_forward = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple=False,
)
self.X_forward = tf.placeholder(tf.float32, (None, None, size))
drop_forward = tf.contrib.rnn.DropoutWrapper(
rnn_cells_forward, output_keep_prob=forget_bias)
self.hidden_layer_forward = tf.placeholder(
tf.float32, (None, num_layers * size_layer))
self.outputs_forward, self.last_state_forward = tf.nn.dynamic_rnn(
drop_forward,
self.X_forward,
initial_state=self.hidden_layer_forward,
dtype=tf.float32,
)
with tf.variable_scope('backward', reuse=False):
rnn_cells_backward = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple=False,
)
self.X_backward = tf.placeholder(tf.float32, (None, None, size))
drop_backward = tf.contrib.rnn.DropoutWrapper(
rnn_cells_backward, output_keep_prob=forget_bias)
self.hidden_layer_backward = tf.placeholder(
tf.float32, (None, num_layers * size_layer))
self.outputs_backward, self.last_state_backward = tf.nn.dynamic_rnn(
drop_backward,
self.X_backward,
initial_state=self.hidden_layer_backward,
dtype=tf.float32,
)
self.outputs = self.outputs_backward - self.outputs_forward
self.Y = tf.placeholder(tf.float32, (None, output_size))
self.logits = tf.layers.dense(self.outputs[-1], output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
def __init__(self, name, input_size, output_size, size_layer):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, None, input_size))
self.hidden_layer = tf.placeholder(tf.float32, (None, 2 * size_layer))
cell = tf.nn.rnn_cell.LSTMCell(size_layer, state_is_tuple = False)
self.rnn,self.last_state = tf.nn.dynamic_rnn(inputs=self.X, cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
tensor_action, tensor_validation = tf.split(self.rnn[:,-1],2,1)
feed_action = tf.layers.dense(tensor_action, output_size)
feed_validation = tf.layers.dense(tensor_validation, 1)
self.logits = feed_validation + tf.subtract(feed_action,
tf.reduce_mean(feed_action,axis=1,keep_dims=True))
def __init__(self, state_size, window_size, trend, skip):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.skip = skip
tf.reset_default_graph()
self.X = tf.placeholder(tf.float32, (None, self.state_size))
self.Y = tf.placeholder(tf.float32, (None, self.state_size))
self.ACTION = tf.placeholder(tf.float32, (None))
self.REWARD = tf.placeholder(tf.float32, (None))
self.batch_size = tf.shape(self.ACTION)[0]
with tf.variable_scope('curiosity_model'):
action = tf.reshape(self.ACTION, (-1,1))
state_action = tf.concat([self.X, action], axis=1)
save_state = tf.identity(self.Y)
feed = tf.layers.dense(state_action, 32, activation=tf.nn.relu)
self.curiosity_logits = tf.layers.dense(feed, self.state_size)
self.curiosity_cost = tf.reduce_sum(tf.square(save_state - self.curiosity_logits), axis=1)
self.curiosity_optimizer = tf.train.RMSPropOptimizer(self.LEARNING_RATE) .minimize(tf.reduce_mean(self.curiosity_cost))
total_reward = tf.add(self.curiosity_cost, self.REWARD)
with tf.variable_scope("q_model"):
with tf.variable_scope("eval_net"):
x_action = tf.layers.dense(self.X, 128, tf.nn.relu)
self.logits = tf.layers.dense(x_action, self.OUTPUT_SIZE)
with tf.variable_scope("target_net"):
y_action = tf.layers.dense(self.Y, 128, tf.nn.relu)
y_q = tf.layers.dense(y_action, self.OUTPUT_SIZE)
q_target = total_reward + self.GAMMA * tf.reduce_max(y_q, axis=1)
action = tf.cast(self.ACTION, tf.int32)
action_indices = tf.stack([tf.range(self.batch_size, dtype=tf.int32), action], axis=1)
q = tf.gather_nd(params=self.logits, indices=action_indices)
self.cost = tf.losses.mean_squared_error(labels=q_target, predictions=q)
self.optimizer = tf.train.RMSPropOptimizer(self.LEARNING_RATE).minimize(
self.cost, var_list=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "q_model/eval_net"))
t_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='q_model/target_net')
e_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='q_model/eval_net')
self.target_replace_op = [tf.assign(t, e) for t, e in zip(t_params, e_params)]
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
def __init__(self, name, input_size, output_size, size_layer, learning_rate):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, input_size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
self.REWARD = tf.placeholder(tf.float32, (None, 1))
feed_critic = tf.layers.dense(self.X, size_layer, activation = tf.nn.relu)
tensor_action, tensor_validation = tf.split(feed_critic,2,1)
feed_action = tf.layers.dense(tensor_action, output_size)
feed_validation = tf.layers.dense(tensor_validation, 1)
feed_critic = feed_validation + tf.subtract(feed_action,tf.reduce_mean(feed_action,axis=1,keep_dims=True))
feed_critic = tf.nn.relu(feed_critic) + self.Y
feed_critic = tf.layers.dense(feed_critic, size_layer//2, activation = tf.nn.relu)
self.logits = tf.layers.dense(feed_critic, 1)
self.cost = tf.reduce_mean(tf.square(self.REWARD - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(self.cost)
def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
forget_bias = 0.1,
):
def lstm_cell(size_layer):
return tf.nn.rnn_cell.LSTMCell(size_layer, state_is_tuple = False)
backward_rnn_cells = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple = False,
)
forward_rnn_cells = tf.nn.rnn_cell.MultiRNNCell(
[lstm_cell(size_layer) for _ in range(num_layers)],
state_is_tuple = False,
)
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
drop_backward = tf.contrib.rnn.DropoutWrapper(
backward_rnn_cells, output_keep_prob = forget_bias
)
forward_backward = tf.contrib.rnn.DropoutWrapper(
forward_rnn_cells, output_keep_prob = forget_bias
)
self.backward_hidden_layer = tf.placeholder(
tf.float32, shape = (None, num_layers * 2 * size_layer)
)
self.forward_hidden_layer = tf.placeholder(
tf.float32, shape = (None, num_layers * 2 * size_layer)
)
self.outputs, self.last_state = tf.nn.bidirectional_dynamic_rnn(
forward_backward,
drop_backward,
self.X,
initial_state_fw = self.forward_hidden_layer,
initial_state_bw = self.backward_hidden_layer,
dtype = tf.float32,
)
self.outputs = tf.concat(self.outputs, 2)
self.logits = tf.layers.dense(self.outputs[-1], output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost
)
def __init__(self, state_size, window_size, trend, skip):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.INITIAL_FEATURES = np.zeros((4, self.state_size))
self.skip = skip
tf.reset_default_graph()
self.actor = Actor('actor-original', self.state_size, self.OUTPUT_SIZE,
self.LAYER_SIZE)
self.actor_target = Actor('actor-target', self.state_size,
self.OUTPUT_SIZE, self.LAYER_SIZE)
self.critic = Critic('critic-original', self.state_size,
self.OUTPUT_SIZE, self.LAYER_SIZE,
self.LEARNING_RATE)
self.critic_target = Critic('critic-target', self.state_size,
self.OUTPUT_SIZE, self.LAYER_SIZE,
self.LEARNING_RATE)
self.grad_critic = tf.gradients(self.critic.logits, self.critic.Y)
self.actor_critic_grad = tf.placeholder(tf.float32,
[None, self.OUTPUT_SIZE])
weights_actor = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
scope='actor')
self.grad_actor = tf.gradients(self.actor.logits, weights_actor,
-self.actor_critic_grad)
grads = zip(self.grad_actor, weights_actor)
self.optimizer = tf.train.AdamOptimizer(
self.LEARNING_RATE).apply_gradients(grads)
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
def __init__(self, name, input_size, output_size, size_layer):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, input_size))
feed_actor = tf.layers.dense(self.X, size_layer, activation = tf.nn.relu)
tensor_action, tensor_validation = tf.split(feed_actor,2,1)
feed_action = tf.layers.dense(tensor_action, output_size)
feed_validation = tf.layers.dense(tensor_validation, 1)
self.logits = feed_validation + tf.subtract(feed_action,
tf.reduce_mean(feed_action,axis=1,keep_dims=True))
def __init__(self,
size_layer,
embedded_size,
learning_rate,
size,
output_size,
num_blocks=2,
num_heads=8,
min_freq=50):
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
encoder_embedded = tf.layers.dense(self.X, embedded_size)
encoder_embedded = tf.nn.dropout(encoder_embedded, keep_prob=0.8)
x_mean = tf.reduce_mean(self.X, axis=2)
en_masks = tf.sign(x_mean)
encoder_embedded += sinusoidal_position_encoding(
self.X, en_masks, embedded_size)
for i in range(num_blocks):
with tf.variable_scope('encoder_self_attn_%d' % i,
reuse=tf.AUTO_REUSE):
encoder_embedded = multihead_attn(queries=encoder_embedded,
keys=encoder_embedded,
q_masks=en_masks,
k_masks=en_masks,
future_binding=False,
num_units=size_layer,
num_heads=num_heads)
with tf.variable_scope('encoder_feedforward_%d' % i,
reuse=tf.AUTO_REUSE):
encoder_embedded = pointwise_feedforward(encoder_embedded,
embedded_size,
activation=tf.nn.relu)
self.logits = tf.layers.dense(encoder_embedded[-1], output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
def __init__(self,
input_,
dimension=2,
learning_rate=0.01,
hidden_layer=256,
epoch=20):
input_size = input_.shape[1]
self.X = tf.placeholder("float", [None, input_.shape[1]])
weights = {
'encoder_h1':
tf.Variable(tf.random_normal([input_size, hidden_layer])),
'encoder_h2':
tf.Variable(tf.random_normal([hidden_layer, dimension])),
'decoder_h1':
tf.Variable(tf.random_normal([dimension, hidden_layer])),
'decoder_h2':
tf.Variable(tf.random_normal([hidden_layer, input_size])),
}
biases = {
'encoder_b1': tf.Variable(tf.random_normal([hidden_layer])),
'encoder_b2': tf.Variable(tf.random_normal([dimension])),
'decoder_b1': tf.Variable(tf.random_normal([hidden_layer])),
'decoder_b2': tf.Variable(tf.random_normal([input_size])),
}
first_layer_encoder = tf.nn.sigmoid(
tf.add(tf.matmul(self.X, weights['encoder_h1']),
biases['encoder_b1']))
self.second_layer_encoder = tf.nn.sigmoid(
tf.add(tf.matmul(first_layer_encoder, weights['encoder_h2']),
biases['encoder_b2']))
first_layer_decoder = tf.nn.sigmoid(
tf.add(tf.matmul(self.second_layer_encoder, weights['decoder_h1']),
biases['decoder_b1']))
second_layer_decoder = tf.nn.sigmoid(
tf.add(tf.matmul(first_layer_decoder, weights['decoder_h2']),
biases['decoder_b2']))
self.cost = tf.reduce_mean(tf.pow(self.X - second_layer_decoder, 2))
self.optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(
self.cost)
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
for i in range(epoch):
last_time = time.time()
_, loss = self.sess.run([self.optimizer, self.cost],
feed_dict={self.X: input_})
if (i + 1) % 10 == 0:
print('epoch:', i + 1, 'loss:', loss, 'time:',
time.time() - last_time)
def __init__(self, name, input_size, output_size, size_layer):
with tf.variable_scope(name):
self.X = tf.placeholder(tf.float32, (None, input_size))
feed_actor = tf.layers.dense(self.X, size_layer, activation = tf.nn.relu)
self.logits = tf.layers.dense(feed_actor, output_size)
def __init__(self, state_size, window_size, trend, skip):
self.state_size = state_size
self.window_size = window_size
self.half_window = window_size // 2
self.trend = trend
self.skip = skip
tf.reset_default_graph()
self.INITIAL_FEATURES = np.zeros((4, self.state_size))
self.X = tf.placeholder(tf.float32, (None, None, self.state_size))
self.Y = tf.placeholder(tf.float32, (None, None, self.state_size))
self.hidden_layer = tf.placeholder(tf.float32,
(None, 2 * self.LAYER_SIZE))
self.ACTION = tf.placeholder(tf.float32, (None))
self.REWARD = tf.placeholder(tf.float32, (None))
self.batch_size = tf.shape(self.ACTION)[0]
self.seq_len = tf.shape(self.X)[1]
with tf.variable_scope('curiosity_model'):
action = tf.reshape(self.ACTION, (-1, 1, 1))
repeat_action = tf.tile(action, [1, self.seq_len, 1])
state_action = tf.concat([self.X, repeat_action], axis=-1)
save_state = tf.identity(self.Y)
cell = tf.nn.rnn_cell.LSTMCell(self.LAYER_SIZE,
state_is_tuple=False)
self.rnn, last_state = tf.nn.dynamic_rnn(
inputs=state_action,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
self.curiosity_logits = tf.layers.dense(self.rnn[:, -1],
self.state_size)
self.curiosity_cost = tf.reduce_sum(
tf.square(save_state[:, -1] - self.curiosity_logits), axis=1)
self.curiosity_optimizer = tf.train.RMSPropOptimizer(
self.LEARNING_RATE).minimize(
tf.reduce_mean(self.curiosity_cost))
total_reward = tf.add(self.curiosity_cost, self.REWARD)
with tf.variable_scope("q_model"):
with tf.variable_scope("eval_net"):
cell = tf.nn.rnn_cell.LSTMCell(self.LAYER_SIZE,
state_is_tuple=False)
rnn, self.last_state = tf.nn.dynamic_rnn(
inputs=self.X,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
self.logits = tf.layers.dense(rnn[:, -1], self.OUTPUT_SIZE)
with tf.variable_scope("target_net"):
cell = tf.nn.rnn_cell.LSTMCell(self.LAYER_SIZE,
state_is_tuple=False)
rnn, last_state = tf.nn.dynamic_rnn(
inputs=self.Y,
cell=cell,
dtype=tf.float32,
initial_state=self.hidden_layer)
y_q = tf.layers.dense(rnn[:, -1], self.OUTPUT_SIZE)
q_target = total_reward + self.GAMMA * tf.reduce_max(y_q, axis=1)
action = tf.cast(self.ACTION, tf.int32)
action_indices = tf.stack(
[tf.range(self.batch_size, dtype=tf.int32), action], axis=1)
q = tf.gather_nd(params=self.logits, indices=action_indices)
self.cost = tf.losses.mean_squared_error(labels=q_target,
predictions=q)
self.optimizer = tf.train.RMSPropOptimizer(
self.LEARNING_RATE).minimize(
self.cost,
var_list=tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES, "q_model/eval_net"))
t_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
scope='q_model/target_net')
e_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
scope='q_model/eval_net')
self.target_replace_op = [
tf.assign(t, e) for t, e in zip(t_params, e_params)
]
self.sess = tf.InteractiveSession()
self.sess.run(tf.global_variables_initializer())
def __init__(
self,
learning_rate,
num_layers,
size,
size_layer,
output_size,
kernel_size=3,
n_attn_heads=16,
dropout=0.9,
):
self.X = tf.placeholder(tf.float32, (None, None, size))
self.Y = tf.placeholder(tf.float32, (None, output_size))
encoder_embedded = tf.layers.dense(self.X, size_layer)
e = tf.identity(encoder_embedded)
for i in range(num_layers):
z = layer(
encoder_embedded,
encoder_block,
kernel_size,
size_layer * 2,
encoder_embedded,
)
z = tf.nn.dropout(z, keep_prob=dropout)
encoder_embedded = z
encoder_output, output_memory = z, z + e
g = tf.identity(encoder_embedded)
for i in range(num_layers):
attn_res = h = layer(
encoder_embedded,
decoder_block,
kernel_size,
size_layer * 2,
residual=tf.zeros_like(encoder_embedded),
)
C = []
for j in range(n_attn_heads):
h_ = tf.layers.dense(h, size_layer // n_attn_heads)
g_ = tf.layers.dense(g, size_layer // n_attn_heads)
zu_ = tf.layers.dense(encoder_output,
size_layer // n_attn_heads)
ze_ = tf.layers.dense(output_memory,
size_layer // n_attn_heads)
d = tf.layers.dense(h_, size_layer // n_attn_heads) + g_
dz = tf.matmul(d, tf.transpose(zu_, [0, 2, 1]))
a = tf.nn.softmax(dz)
c_ = tf.matmul(a, ze_)
C.append(c_)
c = tf.concat(C, 2)
h = tf.layers.dense(attn_res + c, size_layer)
h = tf.nn.dropout(h, keep_prob=dropout)
encoder_embedded = h
encoder_embedded = tf.sigmoid(encoder_embedded[-1])
self.logits = tf.layers.dense(encoder_embedded, output_size)
self.cost = tf.reduce_mean(tf.square(self.Y - self.logits))
self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(
self.cost)
