Question

我有一个简单的序列模型，我正在尝试向该模型添加一个变体层，我的代码是：

encoder_inputs = Input(shape=(None,)) 
encoder_emb = Embedding(input_dim=vocab_in_size, output_dim=embedding_dim)
encoder_lstm =LSTM(units=units,return_state=True)
encoder_outputs, state_h, state_c = encoder_lstm(encoder_emb(encoder_inputs))


################## VAE ##################################

latent_dim =256
# output layer for mean and log variance

z_mu = Dense(2)(state_h)  #remplacer h
z_log_var = Dense(2)(state_h)

def sampling(args):
    batch = K.shape(z_mu)[0]
    dim = K.int_shape(z_mu)[1]
    z_mean, z_log_sigma = args
    epsilon = K.random_normal(shape=(batch,dim),
                              mean=0., stddev=1.)
    return z_mean + K.exp(z_log_sigma/2) * epsilon

# note that "output_shape" isn't necessary with the TensorFlow backend
# so you could write `Lambda(sampling)([z_mean, z_log_sigma])`
z = Lambda(sampling, output_shape=(2,))([z_mu, z_log_var])

decoder_inputs = Input(shape=(None,))
    inp2=decoder_inputs
    decoder_emb = Embedding(input_dim=vocab_out_size, output_dim=embedding_dim)(decoder_inputs)
    hidden=Dense(units,activation='relu')(z)
    state_1 = Dense(units)(hidden)
    state_2 = Dense(units)(hidden)   
    decoder_lstm= LSTM(units=units, return_sequences=True,activation='linear')
    decoder_out=decoder_lstm(decoder_emb,[state_1, state_2])

    class VariationalLayer(Layer):
        """
        `Bottle Neck Layer`. This layer provides the kl-divergence loss function
        """

        def __init__(self, **kwargs):
            self.is_placeholder = True
            self.target = Input(shape=(None,units))  # original time sequence shift once
            super(VariationalLayer, self).__init__(**kwargs)


        def vae_loss(self,x, x_decoded_mean):
            x_decoded_mean=tf.reshape(x_decoded_mean,[-1,256])
            self.xent_loss = K.mean(metrics.mean_squared_error(self.target, x_decoded_mean), axis=1)  # sum error over time
            self.kl_loss = -0.5 * K.sum(1 + z_log_var - K.square(z_mu) - K.exp(z_log_var), axis=-1)
            return K.mean(self.xent_loss + self.kl_loss)

        def call(self, inputs):
            x = inputs[0]
            x_decoded_mean = inputs[1]
            loss = self.vae_loss(x, x_decoded_mean)
            self.add_loss(loss, inputs=inputs)
            # We won't actually use the output.
            return x

    # build model graph
    vae_layer = VariationalLayer(name='loss_layer')
    y = vae_layer([encoder_inputs, decoder_out])
    vae = Model([encoder_inputs,vae_layer.target],y)
    vae.compile(optimizer="rmsprop", loss=None)

我收到此错误：

ValueError：图形已断开：无法获得张量的值层上的Tensor（“ input_112：0”，shape =（?,？），dtype = float32） “ input_112”。访问以下先前的层时没有问题：['input_111'，'embedding_77'，'lstm_77'，'dense_192'， 'dense_193'，'lambda_40'，'dense_194']

我是深度学习的初学者，我无法理解错在哪里。

图形断开：无法获得张量值

0 个答案: