LSTM Keras - 什么是正确的输入形状

Question

实际上，我正在尝试在 Keras 和 Tensorflow 中构建 LSTM 模型。我的数据集有大约 3200 个项目，具有 4 个特征和 3 个标签。

X Shape: (3200, 4)
Y Shape: (3200, 3)

如果我想要大约 5 次步骤，那么我是否必须像这样重塑：

n_time_steps= 5
n_features = 4
X_train = X_train.reshape((-1, n_time_steps, n_features))

所以我得到了这些形状：

X Shape: (640, 5, 4)
Y Shape: (3200, 3)

我有点困惑，因为 640 =！ 3200 个数据点......但模型编译和拟合没有任何错误。但是 acc 和 loss 是疯狂的。 当我也尝试重塑 Y_train Y Shape: (640, 5, 3) throws

不兼容的形状：[10,3] 与 [10,5,3] [[节点子（定义于：12）]] [操作：__inference_train_function_74818 函数调用栈： train_function

这是我的模型

opt = 'adam'
model = keras.Sequential()
model.add(layers.LSTM(128, input_shape=(n_time_steps,4)))
model.add(layers.Dropout(0.2))
model.add(layers.Dense(3 ,activation="sigmoid"))
model.compile(optimizer=opt,loss=hn_multilabel_loss,metrics=['accuracy','mae'])
model.summary()

history = model.fit(X_train, Y_train,batch_size = 10, epochs=10, validation_split = 0.1)

有人知道如何创建一个具有 5 个时间步长和 4 个特征的 LSTM 吗？什么是正确的输入和输出形状？

谢谢大家！

Answer 1

您可以使用此函数将 2D 数据集转换为具有可自定义时间步数的数据集：

def multivariate_data(dataset, target, start_index, end_index, history_size,
                      target_size, step, single_step=False):
  data = []
  labels = []

  start_index = start_index + history_size
  if end_index is None:
    end_index = len(dataset) - target_size

  for i in range(start_index, end_index):
    indices = range(i-history_size, i, step)
    data.append(dataset[indices])

    if single_step:
      labels.append(target[i+target_size])
    else:
      labels.append(target[i:i+target_size])

  return np.array(data), np.array(labels)

我成功完成了你的任务（我简化了一点）：

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

X_train = np.random.rand(3200, 4)
y_train = np.random.randint(0, 2, (3200, 3))

def multivariate_data(dataset, target, start_index, end_index, history_size,
                      target_size, step, single_step=False):
  data, labels = [], []
  start_index = start_index + history_size
  if end_index is None:
    end_index = len(dataset) - target_size
  for i in range(start_index, end_index):
    indices = range(i-history_size, i, step)
    data.append(dataset[indices])
    if single_step:
      labels.append(target[i+target_size])
    else:
      labels.append(target[i:i+target_size])

  return np.array(data), np.array(labels)

X_train, y_train = multivariate_data(X_train, y_train, 0, 3200, 5, 0, 1, True)

n_time_steps, n_features = 5, 4

model = tf.keras.Sequential()
model.add(layers.LSTM(128, input_shape=(n_time_steps,4)))
model.add(layers.Dense(3))
model.compile(optimizer='adam',loss='mae')

history = model.fit(X_train, y_train, batch_size = 10, epochs=1)

输出：

  10/3195 [..............................] - ETA: 16:12 - loss: 0.3244
 120/3195 [>.............................] - ETA: 1:19 - loss: 0.2725 
 230/3195 [=>............................] - ETA: 40s - loss: 0.2536 
 330/3195 [==>...........................] - ETA: 27s - loss: 0.2545
 440/3195 [===>..........................] - ETA: 20s - loss: 0.2597