ValueError: Input 0 is in compatible with layer model: expected shape=(None, 14999, 7), found shape=(None, 7)

Question

我正在尝试将 Conv1D 层应用于具有数字数据集的分类模型。我的模型的神经网络如下：

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv1D(8,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu',input_shape = (14999,7)))
model.add(tf.keras.layers.Conv1D(16,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu'))
model.add(tf.keras.layers.MaxPooling1D(2))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Conv1D(32,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu'))
model.add(tf.keras.layers.Conv1D(64,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu'))
model.add(tf.keras.layers.MaxPooling1D(2))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Conv1D(128,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu'))
model.add(tf.keras.layers.Conv1D(256,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu'))
model.add(tf.keras.layers.MaxPooling1D(2))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(512,activation = 'relu'))
model.add(tf.keras.layers.Dense(128,activation = 'relu'))
model.add(tf.keras.layers.Dense(32,activation = 'relu'))
model.add(tf.keras.layers.Dense(3, activation = 'softmax'))

而模型的输入形状为 (14999, 7)。

model.summary() 给出以下输出

Model: "sequential_8"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d_24 (Conv1D)           (None, 14997, 8)          176       
_________________________________________________________________
conv1d_25 (Conv1D)           (None, 14995, 16)         400       
_________________________________________________________________
max_pooling1d_10 (MaxPooling (None, 7497, 16)          0         
_________________________________________________________________
dropout_9 (Dropout)          (None, 7497, 16)          0         
_________________________________________________________________
conv1d_26 (Conv1D)           (None, 7495, 32)          1568      
_________________________________________________________________
conv1d_27 (Conv1D)           (None, 7493, 64)          6208      
_________________________________________________________________
max_pooling1d_11 (MaxPooling (None, 3746, 64)          0         
_________________________________________________________________
dropout_10 (Dropout)         (None, 3746, 64)          0         
_________________________________________________________________
conv1d_28 (Conv1D)           (None, 3744, 128)         24704     
_________________________________________________________________
conv1d_29 (Conv1D)           (None, 3742, 256)         98560     
_________________________________________________________________
max_pooling1d_12 (MaxPooling (None, 1871, 256)         0         
_________________________________________________________________
dropout_11 (Dropout)         (None, 1871, 256)         0         
_________________________________________________________________
flatten_3 (Flatten)          (None, 478976)            0         
_________________________________________________________________
dense_14 (Dense)             (None, 512)               245236224 
_________________________________________________________________
dense_15 (Dense)             (None, 128)               65664     
_________________________________________________________________
dense_16 (Dense)             (None, 32)                4128      
_________________________________________________________________
dense_17 (Dense)             (None, 3)                 99        
=================================================================
Total params: 245,437,731
Trainable params: 245,437,731
Non-trainable params: 0

模型拟合代码为：

model.compile(loss = 'sparse_categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])
history = model.fit(xtrain_scaled, ytrain_scaled, epochs = 30, batch_size = 5, validation_data = (xval_scaled, yval_scaled))

在执行时，我遇到以下错误：

ValueError: Input 0 is incompatible with layer model: expected shape=(None, 14999, 7), found shape=(None, 7)

谁能帮忙解决这个问题？

Answer 1

TL;DR：

改变

model.add(tf.keras.layers.Conv1D(8,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu',input_shape = (14999,7)))

到

model.add(tf.keras.layers.Conv1D(8,kernel_size = 3, strides = 1,padding = 'valid', activation = 'relu',input_shape = (7)))

完整答案：

假设：我猜您选择在输入形状中写入 14999 的原因是因为这是您的批量大小/训练数据的总大小

这个问题：

• Tensorflow 假设输入形状不包括批量大小
  • 通过指定 2D input_shape，您可以让 Tensorflow 期待形状为 (Batch_size, 14999, 7) 的 3D 输入。但是你的输入显然是大小(Batch_size, 7)

解决方案：

将形状从 (14999, 7) 更改为 (7)

• TF 现在将期待您提供的相同形状

PS：不要担心告诉你的模型你在数据集中有多少训练示例。 TF Keras 假设它可以显示无限的训练示例。