python - LSTM val_loss 没有从 inf 改进

Question

我正在尝试教一个模型来预测股票价格，由于一个热编码，我的数据框有很多列，这是我的模型代码。

seq_len = 128
opt = keras.optimizers.Adam(learning_rate=0.001)

def create_model():
  in_seq = Input(shape = (seq_len, 143))
      
  x = Bidirectional(LSTM(128, return_sequences=True))(in_seq)
  x = Bidirectional(LSTM(128, return_sequences=True))(x)
  x = Bidirectional(LSTM(64, return_sequences=True))(x) 
          
  avg_pool = GlobalAveragePooling1D()(x)
  max_pool = GlobalMaxPooling1D()(x)
  conc = concatenate([avg_pool, max_pool])
  conc = Dense(64, activation="relu")(conc)
  out = Dense(1, activation="linear")(conc)      

  model = Model(inputs=in_seq, outputs=out)
  model.compile(loss="mse", optimizer= opt, metrics=['mae', 'mape'])    
  return model

model = create_model()

#model.summary()
#callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=2)
callback = tf.keras.callbacks.ModelCheckpoint('Bi-LSTM.hdf5', monitor='val_loss', save_best_only=True, verbose=1)

model.fit(X_train, y_train,
              batch_size=2048,
              verbose=2,
              callbacks=[callback],
              epochs=200,
              #shuffle=True,
              validation_data=(X_val, y_val),)    

model = tf.keras.models.load_model('/content/Bi-LSTM.hdf5')

###############################################################################
'''Calculate predictions and metrics'''

#Calculate predication for training, validation and test data
train_pred = model.predict(X_train)
val_pred = model.predict(X_val)
test_pred = model.predict(X_test)

#Print evaluation metrics for all datasets
train_eval = model.evaluate(X_train, y_train, verbose=0)
val_eval = model.evaluate(X_val, y_val, verbose=0)
test_eval = model.evaluate(X_test, y_test, verbose=0)
print(' ')
print('Evaluation metrics')
print('Training Data - Loss: {:.4f}, MAE: {:.4f}, MAPE: {:.4f}'.format(train_eval[0], train_eval[1], train_eval[2]))
print('Validation Data - Loss: {:.4f}, MAE: {:.4f}, MAPE: {:.4f}'.format(val_eval[0], val_eval[1], val_eval[2]))
print('Test Data - Loss: {:.4f}, MAE: {:.4f}, MAPE: {:.4f}'.format(test_eval[0], test_eval[1], test_eval[2]))

但在训练期间，结果是

Epoch 000xx: val_loss did not improve from inf
Epoch x/xxx
x/x - 19s - loss: nan - mae: nan - mape: nan - val_loss: nan - val_mae: nan - val_mape: nan

所以我的问题是，

1. 让模型训练更长的时间会改善结果吗？
2. 我可以修复这个模型还是必须更改为另一个模型？

请让我知道我错在哪里。

标签： pythontensorflowmachine-learningkeraslstm