Question

我在时间序列数据上实现了keras seq2seq模型。我认为这不是编码问题，但可以在这里找到代码（向下滚动查看帖子）或查看此处模型仍然可以对远离训练数据的数据提供良好的结果。为什么会这样？我已经使用了不同的隐藏单位大小，时期，批量大小和时间步长参数，但我训练的范围为-0.2到0.2，并且没想到模型可以预测值，直到-0.8到0.8后退，这正在发生。因此，它不仅可以从数据中推广，更可以预测，这不应该发生。

我试过这个Seq2Seq：

num_encoder_tokens = 1
num_decoder_tokens = 1
encoder_seq_length = None
decoder_seq_length = None
batch_size = #100
epochs = #1
hidden_units= #30
timesteps= #20

#Input Data 
 input_seqs=train #data with sliding window, shape(None,20,1)
 target_seqs=reversed_train #first reversed than sliding window added 
 with same shape as above

 #define training encoder
 encoder_inputs = Input(shape=(None, num_encoder_tokens))
 encoder = LSTM(hidden_units, return_state=True)
 encoder_outputs, state_h, state_c = encoder(encoder_inputs)
 encoder_states = [state_h, state_c]
 #define training decoder
 decoder_inputs = Input(shape=(None,num_decoder_tokens))
 decoder_lstm = LSTM(hidden_units, return_sequences=True, 
 return_state=True)

 decoder_outputs, _, _ = decoder_lstm(decoder_inputs, 
 initial_state=encoder_states)
 decoder_dense = Dense(num_encoder_tokens, activation='tanh')
 decoder_outputs = decoder_dense(decoder_outputs)
 model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

 #training
 model.compile(optimizer='adam', loss='mse')
 model.fit([input_seqs, target_seqs], target_seqs,batch_size=batch_size, 
  epochs=epochs,validation_split=0.2)

 #Testdata
 target_seqs=test # different data as target with same shape

 #define inference encoder
 encoder_model = Model(encoder_inputs, encoder_states)
 #define inference decoder
 decoder_state_input_h = Input(shape=(hidden_units,))
 decoder_state_input_c = Input(shape=(hidden_units,))
 decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]

 decoder_outputs, state_h, state_c = decoder_lstm(decoder_inputs, 
 initial_state=decoder_states_inputs)
 decoder_states = [state_h, state_c]
 decoder_outputs = decoder_dense(decoder_outputs)
 decoder_model = Model([decoder_inputs] + decoder_states_inputs, 
 [decoder_outputs] + decoder_states)


 #Initalize states for decoder 
 states_values = encoder_model.predict(input_seqs)

 #empty target
 target_seq = np.zeros((1, 1, num_decoder_tokens))
 #predict
 output=list()
 for t in range(timesteps):
    output_tokens, h, c  = decoder_model.predict([target_seqs] + states_values)
    output.append(output_tokens[0,0,:])
    states_values = [h,c]
    target_seq = output_tokens

这个更简单的模型有或没有repeatvector不会消磨，我也会遇到同样的问题：

from keras.models import Sequential, Model
from keras.layers import LSTM, RepeatVector

model = Sequential()
model.add(LSTM(40, input_shape=(timesteps, n_features)))
model.add(RepeatVector(timesteps))
model.add(LSTM(40, return_sequences=True))
model.add(Dense(1, activation='tanh'))
model.compile(loss='mse', optimizer='adam', metrics=['mae'])
history=model.fit(train, train, epochs=200, shuffle=False)

lstm seq2seq预测输出不如预期

0 个答案: