VAE添加Concatenate/Add层报Graph disconnected错误求助

我正在为自回归时间序列任务扩展现有VAE网络，引入输入泄漏机制。使用LSTM模块时，无Concatenate或Add层的标准网络可正常运行，但精度不理想。因此我希望实现自回归：生成时序时结合mu、sigma与历史样本。

为此我编写代码添加输入及维度匹配层，但始终报错：

ValueError: Graph disconnected: cannot obtain value for tensor Tensor("input_39:0", shape=(None, 4, 96), dtype=float32) at layer "lstm_29". The following previous layers were accessed without issue: ['repeat_vector_11', 'lstm_30', 'time_distributed_8']

相关代码如下：

from tensorflow import keras
import tensorflow as tf
from tensorflow.keras import layers


class Sampling(layers.Layer):
    """Uses (z_mean, z_log_var) to sample z."""

    def call(self, inputs):
        z_mean, z_log_var = inputs
        batch = tf.shape(z_mean)[0]
        dim = tf.shape(z_mean)[1]
        epsilon = tf.keras.backend.random_normal(shape=(batch, dim),
                                  mean=0., stddev=1.)
        return z_mean + tf.exp(0.5 * z_log_var) * epsilon    


# Build encoder and decoder
latent_dim = 2
input_shape = (4, 96)
encoder_input = keras.Input(shape=input_shape)
enc = layers.LSTM(64)(encoder_input)
z_mean = layers.Dense(latent_dim, name="z_mean")(enc)
z_log_sigma = layers.Dense(latent_dim, name="z_log_var")(enc)
z = Sampling()([z_mean, z_log_sigma])
encoder = keras.Model(encoder_input, [z_mean, z_log_sigma, z], name="encoder")
encoder.summary()


# Extra layers for leaky input to decoder
leak_input_dec_ip = keras.Input(shape=input_shape)
leak_input_dec = layers.LSTM(2)(leak_input_dec_ip)

# Decoder where we have inputs from leaky input and sampling layer
inp_z = keras.Input(shape=(latent_dim,))
dec = layers.Concatenate()([leak_input_dec, inp_z]) # or Add
dec = layers.RepeatVector(96)(inp_z)
dec = layers.LSTM(64, return_sequences=True)(dec)
out = layers.TimeDistributed(layers.Dense(96))(dec)
decoder = keras.Model([leak_input_dec, inp_z], out) 
decoder.summary()

错误核心 #

报错源于模型计算图断开：你在定义Decoder时，将中间层输出leak_input_dec作为模型输入，但Keras要求Model的输入必须是keras.Input实例。同时代码中拼接泄漏输入后未使用拼接结果，等于白加了泄漏机制的逻辑。

修正步骤 #

1. 把泄漏输入的Input层直接作为Decoder的输入，而非中间层张量
2. 确保拼接后的张量被用于后续网络计算，让泄漏信息真正传入模型
3. 检查维度匹配，保证各层输入输出维度兼容

修正后的代码 #

from tensorflow import keras
import tensorflow as tf
from tensorflow.keras import layers


class Sampling(layers.Layer):
    """Uses (z_mean, z_log_var) to sample z."""

    def call(self, inputs):
        z_mean, z_log_var = inputs
        batch = tf.shape(z_mean)[0]
        dim = tf.shape(z_mean)[1]
        epsilon = tf.keras.backend.random_normal(shape=(batch, dim),
                                  mean=0., stddev=1.)
        return z_mean + tf.exp(0.5 * z_log_var) * epsilon    


# Build encoder and decoder
latent_dim = 2
input_shape = (4, 96)
encoder_input = keras.Input(shape=input_shape)
enc = layers.LSTM(64)(encoder_input)
z_mean = layers.Dense(latent_dim, name="z_mean")(enc)
z_log_sigma = layers.Dense(latent_dim, name="z_log_var")(enc)
z = Sampling()([z_mean, z_log_sigma])
encoder = keras.Model(encoder_input, [z_mean, z_log_sigma, z], name="encoder")
encoder.summary()


# Decoder with leaky input
# 直接将Input层作为模型输入
leak_input_dec_ip = keras.Input(shape=input_shape)
# 处理泄漏输入
leak_processed = layers.LSTM(2)(leak_input_dec_ip)

inp_z = keras.Input(shape=(latent_dim,))
# 拼接泄漏输入处理结果与z
dec_concat = layers.Concatenate()([leak_processed, inp_z])
# 使用拼接后的张量进行后续处理
dec = layers.RepeatVector(96)(dec_concat)
dec = layers.LSTM(64, return_sequences=True)(dec)
out = layers.TimeDistributed(layers.Dense(96))(dec)
# 模型输入为两个Input层实例
decoder = keras.Model([leak_input_dec_ip, inp_z], out) 
decoder.summary()

额外提示 #

• 自回归逻辑在推理阶段需要循环传入历史生成样本：训练时可用真实历史数据，推理时需将上一步生成结果作为泄漏输入传入
• 如果需要逐步依赖前序输出的严格自回归，可考虑用tf.while_loop或递归LSTM结构实现序列的逐步生成

内容的提问来源于stack exchange，提问作者ikespand