python - 具有整数序列的 Keras 示例字级模型给出 `expected ndim=3, found ndim=4`

Question

我正在尝试在 their blog 上实现 Keras 字级示例列在奖励部分 -> 如果我想使用带有整数序列的字级模型怎么办？

我已经用名称标记了各层，以帮助我稍后将这些层从加载的模型重新连接到推理模型。我想我已经遵循了他们的示例模型:

# Define an input sequence and process it - where the shape is (timesteps, n_features)
encoder_inputs = Input(shape=(None, src_vocab), name='enc_inputs')
# Add an embedding layer to process the integer encoded words to give some 'sense' before the LSTM layer
encoder_embedding = Embedding(src_vocab, latent_dim, name='enc_embedding')(encoder_inputs)
# The return_state constructor argument configures a RNN layer to return a list where the first entry is the outputs
# and the next entries are the internal RNN states. This is used to recover the states of the encoder.
encoder_outputs, state_h, state_c = LSTM(latent_dim, return_state=True, name='encoder_lstm')(encoder_embedding)
# We discard `encoder_outputs` and only keep the states.
encoder_states = [state_h, state_c]

# Set up the decoder, using `encoder_states` as initial state of the RNN.
decoder_inputs = Input(shape=(None, target_vocab), name='dec_inputs')
decoder_embedding = Embedding(target_vocab, latent_dim, name='dec_embedding')(decoder_inputs)
# The return_sequences constructor argument, configuring a RNN to return its full sequence of outputs (instead of
# just the last output, which the defaults behavior).
decoder_lstm = LSTM(latent_dim, return_sequences=True, name='dec_lstm')(decoder_embedding, initial_state=encoder_states)
decoder_outputs = Dense(target_vocab, activation='softmax', name='dec_outputs')(decoder_lstm)
# Put the model together
model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

但我明白了

ValueError: Input 0 is incompatible with layer encoder_lstm: expected ndim=3, found ndim=4

上线

encoder_outputs, state_h, state_c = LSTM(...

我错过了什么？或者博客上的示例是否假设我跳过了一个步骤？

更新:

我正在训练:

X = [source_data, target_data]
y = offset_data(target_data)
model.fit(X, y, ...)

更新 2:

所以，我还没有完全做到这一点。我的 decoder_lstm 和 decoder_outputs 像上面一样定义，并修复了输入。当我从 h5 文件加载模型并构建推理模型时，我尝试使用

连接到训练 模型

decoder_inputs = model.input[1]  # dec_inputs (Input(shape=(None,)))
# decoder_embedding = model.layers[3]  # dec_embedding (Embedding(target_vocab, latent_dim)) 
target_vocab = model.output_shape[2]
decoder_state_input_h = Input(shape=(latent_dim,), name='input_3')  # named to avoid conflict
decoder_state_input_c = Input(shape=(latent_dim,), name='input_4')
decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]
# Use decoder_lstm from the training model
# decoder_lstm = LSTM(latent_dim, return_sequences=True)
decoder_lstm = model.layers[5] # dec_lstm
decoder_outputs, state_h, state_c = decoder_lstm(decoder_inputs, initial_state=decoder_states_inputs)

但我收到错误

ValueError: Input 0 is incompatible with layer dec_lstm: expected ndim=3, found ndim=2

尝试传递 decoder_embedding 而不是 decoder_inputs 也会失败。

我正在尝试改编 lstm_seq2seq_restore.py 的示例但它不包括嵌入层的复杂性。

更新 3:

当我使用 decoder_outputs, state_h, state_c = detector_lstm(decoder_embedding, ...) 构建推理模型时，我已确认 decoder_embedding 是 type 的对象嵌入但我得到:

ValueError: Layer dec_lstm was called with an input that isn't a symbolic tensor. Received type: <class 'keras.layers.embeddings.Embedding'>. Full input: [<keras.layers.embeddings.Embedding object at 0x1a1f22eac8>, <tf.Tensor 'input_3:0' shape=(?, 256) dtype=float32>, <tf.Tensor 'input_4:0' shape=(?, 256) dtype=float32>]. All inputs to the layer should be tensors.

该模型的完整代码位于 Bitbucket .

Answer 1

问题出在Input层的输入形状上。嵌入层接受整数序列作为输入，对应于句子中的单词索引。由于这里句子的单词数不固定，因此必须将Input层的输入形状设置为(None,)。

我认为您误认为我们的模型中没有嵌入层，因此模型的输入形状为 (timesteps, n_features) 以使其与LSTM 层。

更新:

您需要首先将 decoder_inputs 传递到嵌入层，然后将生成的输出张量传递到 decoder_lstm 层，如下所示:

decoder_inputs = model.input[1] # (Input(shape=(None,)))
# pass the inputs to the embedding layer
decoder_embedding = model.get_layer(name='dec_embedding')(decoder_inputs) 

# ...

decoder_lstm = model.get_layer(name='dec_lstm') # dec_lstm
decoder_outputs, state_h, state_c = decoder_lstm(decoder_embedding, ...)

更新2:

在训练时，创建decoder_lstm层时需要设置return_state=True:

decoder_lstm, _, _ = LSTM(latent_dim, return_sequences=True, return_state=True, name='dec_lstm')(decoder_embedding, initial_state=encoder_states)