如何在 Keras-Python 中输入二维数组？

Question

我对机器学习和 keras 非常陌生，一直在尝试输入数据；我有如下数据：

[[[0.01363717 0.        ]
  [0.01577874 0.        ]
  [0.01463021 0.        ]]

 [[0.01577874 0.        ]
  [0.01463021 0.        ]
  [0.01006721 0.        ]]

 [[0.01463021 0.        ]
  [0.01006721 0.        ]
  [0.00762504 0.        ]]...]

数据的形状是：(1607, 3, 2)。我怎样才能通过：

 [[0.01363717 0.        ]
  [0.01577874 0.        ]
  [0.01463021 0.        ]]

作为 512 个 CuDNNLSTM 单元层的输入？

这是我的整个网络：

def create_model():
    model = Sequential()

    model.add(CuDNNLSTM(512, input_shape=(3,2), return_sequences=True, name='inputlstm1'))
    model.add(Dropout(0.2))

    model.add(CuDNNLSTM(512, return_sequences=True,name='lstm2'))
    model.add(Dropout(0.2))

    model.add(CuDNNLSTM(512, return_sequences=True,name='lstm3'))
    model.add(Dropout(0.2))

    model.add(Dense(32, activation='relu', name='dense1'))
    model.add(Dropout(0.2))

    model.add(Dense(1, activation='softmax', name='denseoutput2'))


    # Compile model
    model.compile(
        loss='mse',
        optimizer='adam',
        metrics=['accuracy'],
    )
    return model 

以及它的拟合：

model=create_model()
history=model.fit(xtrain, ytrain,batch_size=1, epochs=5, validation_data=(xtest, ytest), verbose=1)

Answer 1

keras 层的构建包括指定传递数组的形状，这里要训练的数组形状为 (3,2)，有 1607 个样本，

input_shape = (3,2)
X = LSTM(124, activation = 'sigmoid', name='layer1', dropout = 0.4) (temp)

如果你想使用堆叠的 LSTM，你可以使用这个

input_shape = (3,2)
    X = LSTM(124, activation = 'sigmoid', name='layer1', dropout = 0.4,return_sequences=True) (temp)
    X = LSTM(64, activation = 'sigmoid', name='layer2', dropout = 0.4) (X)

编辑

def create_model():
    model = keras.models.Sequential()

    model.add(keras.layers.CuDNNLSTM(512, input_shape=(3,2), return_sequences=True, name='inputlstm1'))
    model.add(keras.layers.Dropout(0.2))

    model.add(keras.layers.CuDNNLSTM(512, return_sequences=True,name='lstm2'))
    model.add(keras.layers.Dropout(0.2))

    # The last layer of Stacked LSTM need not to return the input sequences
    model.add(keras.layers.CuDNNLSTM(512,name='lstm3'))
    model.add(keras.layers.Dropout(0.2))

    model.add(keras.layers.Dense(32, activation='relu', name='dense1'))
    model.add(keras.layers.Dropout(0.2))

    model.add(keras.layers.Dense(1, activation='softmax', name='denseoutput2'))


    # Compile model
    model.compile(
        loss='mse',
        optimizer='adam',
        metrics=['accuracy'],
    )
    return model