class VAE(keras.Model):

    def __init__(self):
        super(VAE, self).__init__()

        # Encoder
        self.fc1 = layers.Dense(128)
        self.fc2 = layers.Dense(64)
        self.fc3 = layers.Dense(z_dim) # get mean prediction
        self.fc4 = layers.Dense(z_dim)

        # Decoder
        self.fc5 = layers.Dense(64)
        self.fc6 = layers.Dense(128)
        self.fc7 = layers.Dense(96)

    def encoder(self, x):
        h = tf.nn.relu(self.fc1(x))
        h = tf.nn.relu(self.fc2(h))
        # get mean
        mu = self.fc3(h)
        # get variance
        log_var = self.fc4(h)

        return mu, log_var

    def decoder(self, z):
        out = tf.nn.relu(self.fc5(z))
        out = tf.nn.relu(self.fc6(out))
        out = self.fc7(out)

        return out

    def reparameterize(self, mu, log_var):
        eps = tf.random.normal(log_var.shape)
        std = tf.exp(log_var*0.5)
        z = mu + std * eps

        return z

    def call(self, inputs, training=None):
        # [b, 96] => [b, z_dim], [b, z_dim]
        mu, log_var = self.encoder(inputs)
        # reparameterization trick
        z = self.reparameterize(mu, log_var)
        x_hat = self.decoder(z)

        return x_hat, mu, log_var

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48

模型调用对应input

vae=VAE()
xhat,mu1,var=vae(data.values)

• 1
• 2