Python bese_net Beispiele

Beispiel #1

Datei: qr_dqn.py Projekt: q-learning-trader/tensorflow-rl

def build_model(n=200, dim=(30, 4)):
    inputs = tf.keras.layers.Input(dim)

    x = base.bese_net(inputs)

    out = output(x, n)

    return tf.keras.Model(inputs, out)

Beispiel #2

Datei: qr_dqn.py Projekt: zhenpingfeng/tensorflow2-reinforcement-learning

def build_model(n=200, dim=(130, 4)):
    inputs = tf.keras.layers.Input(dim, name="inputs")

    x = base.bese_net(inputs)
    x = tf.keras.layers.Flatten()(x)

    out = output(x, n)

    return tf.keras.Model(inputs, out)

Beispiel #3

Datei: neural_evolution_ac.py Projekt: q-learning-trader/tensorflow-rl

def ne_critic(dim=(10,4)):
    states = tf.keras.layers.Input(dim, name="states")
    action = tf.keras.layers.Input((2,), name="action")

    x = base.bese_net(states)

    x_action = tf.keras.layers.Concatenate()([x, action])
    q1 = output(x_action, "q1")
    q2 = output(x_action, "q2")
    v = output(x, "v")

Beispiel #4

Datei: neural_evolution_ac.py Projekt: q-learning-trader/tensorflow-rl

def critic(dim):
    states = tf.keras.layers.Input(dim, name="states")
    action = tf.keras.layers.Input((2,), name="action")

    x = base.bese_net(states)

    x_action = tf.keras.layers.Concatenate()([x, action])
    q1 = output(x_action, "q1")
    q2 = output(x_action, "q2")
    v = output(x, "v")

    return tf.keras.Model([states, action], [q1, q2, v])

Beispiel #5

Datei: td3.py Projekt: zhenpingfeng/tensorflow2-reinforcement-learning

def critic(input_shape):
    s = tf.keras.layers.Input(input_shape, name="s")
    a = tf.keras.layers.Input((2,), name="a")

    x = base.bese_net(s)
    f = tf.keras.layers.Flatten()(x)

    x = tf.keras.layers.Concatenate()([f, a])
    q1 = q(x, "q1")
    q2 = q(x, "q2")

    return tf.keras.Model([s, a], [q1, q2])

Beispiel #6

Datei: rainbow.py Projekt: zhenpingfeng/tensorflow2-reinforcement-learning

def build_model(n=200, dim=(130, 4)):
    inputs = tf.keras.layers.Input(dim, name="inputs")

    x = base.bese_net(inputs)
    x = tf.keras.layers.Flatten()(x)
    x = IndependentDense(512, "elu")(x)

    advantage = output(x, n)
    value = IndependentDense(n)(x)
    value = tf.reshape(value, (-1,1,n))
    sub = advantage - tf.reshape(tf.reduce_mean(advantage, 1), (-1,1,n))
    out = tf.keras.layers.Add(name="q")([value, sub])

    return tf.keras.Model(inputs, out)

Beispiel #7

Datei: dqn.py Projekt: zhenpingfeng/tensorflow2-reinforcement-learning

def build_model(dim=(130, 4)):
    inputs = tf.keras.layers.Input(dim)

    x = base.bese_net(inputs)
    x = tf.keras.layers.Flatten()(x)
    # x = tf.keras.layers.GRU(128)(x)

    x = tf.keras.layers.Dense(512, "elu")(x)
    out = tf.keras.layers.Dense(3, name="q")(x)
    # v = tf.keras.layers.Dense(328, "elu", kernel_initializer="he_normal")(x)
    # v = tf.keras.layers.Dense(1, name="v")(v)
    # a = tf.keras.layers.Dense(328, "elu", kernel_initializer="he_normal")(x)
    # a = tf.keras.layers.Dense(2, name="a")(a)
    # out = v + tf.subtract(a, tf.reduce_mean(a, axis=1, keepdims=True))

    return tf.keras.Model(inputs, out)

Beispiel #8

Datei: dqn.py Projekt: q-learning-trader/tensorflow-rl

def build_model(dim=(30, 4)):
    inputs = tf.keras.layers.Input(dim)

    x = base.bese_net(inputs)

    tensor_action, tensor_validation = tf.split(x, 2, 1)
    v = tf.keras.layers.Dense(
        328, "elu", kernel_initializer="he_normal")(tensor_validation)
    # v = tf.keras.layers.BatchNormalization()(v)
    v = tf.keras.layers.Dense(1, name="v")(v)
    a = tf.keras.layers.Dense(328, "elu",
                              kernel_initializer="he_normal")(tensor_action)
    # a = tf.keras.layers.BatchNormalization()(a)
    a = tf.keras.layers.Dense(2, name="a")(a)
    out = v + tf.subtract(a, tf.reduce_mean(a, axis=1, keepdims=True))

    return tf.keras.Model(inputs, out)