Question

使用时间序列与Keras时出现此错误：

ValueError: Error when checking input: expected lstm_1_input to have 3 dimensions, but got array with shape (31, 3)

这是我的功能：

def CreateModel(shape):
  """Creates Keras Model.

  Args:
    shape: (set) Dataset shape. Example: (31,3).

  Returns:
    A Keras Model.

  Raises:
    ValueError: Invalid shape
  """

  if not shape:
    raise ValueError('Invalid shape')

  logging.info('Creating model')
  model = Sequential()
  model.add(LSTM(4, input_shape=(31, 3)))
  model.add(Dense(1))
  model.compile(loss='mean_squared_error', optimizer='adam')
  return model

主要代码：

print(training_features.shape)
model = CreateModel(training_features.shape)
model.fit(
      training_features,
      training_label,
      epochs=FLAGS.epochs,
      batch_size=FLAGS.batch_size,
      verbose=FLAGS.keras_verbose_level)

完成错误：

Traceback (most recent call last):
  File "<embedded module '_launcher'>", line 149, in run_filename_as_main
  File "<embedded module '_launcher'>", line 33, in _run_code_in_main
  File "model.py", line 300, in <module>
    app.run(main)
  File "absl/app.py", line 433, in run
    _run_main(main, argv)
  File "absl/app.py", line 380, in _run_main
    sys.exit(main(argv))
  File "model.py", line 274, in main
    verbose=FLAGS.keras_verbose_level)
  File "keras/models.py", line 960, in fit
    validation_steps=validation_steps)
  File "keras/engine/training.py", line 1581, in fit
    batch_size=batch_size)
  File "keras/engine/training.py", line 1414, in _standardize_user_data
    exception_prefix='input')
  File "keras/engine/training.py", line 141, in _standardize_input_data
    str(array.shape))
ValueError: Error when checking input: expected lstm_1_input to have 3 dimensions, but got array with shape (31, 3)

代码最初来自here

我试过了：

training_features = numpy.reshape(
      training_features,
      (training_features.shape[0], 1, training_features.shape[1]))

但我明白了：

ValueError: Input 0 is incompatible with layer lstm_1: expected ndim=3, found ndim=4

Answer 1

如果您的原始数据是（31,3），那么我认为您正在寻找的是training_features.shape =（31,3,1）。您可以使用以下行...

training_features = training_features.reshape(-1, 3, 1)

这将简单地向现有数据添加一个新轴（-1只是告诉numpy使用原始数据中的值来计算这个维度。）

您还需要修复模型的输入形状。 31应该是数据中的样本数。这不会包含在Keras input_shape参数中。你应该使用......

model.add(LSTM(4, input_shape=(3, 1)))

Keras会自动将批量大小设置为None，这意味着任何数量的样本都可以与模型一起使用。

输入形状的Keras LSTM输入形状误差

1 个答案: