检查输入时出错：预期 conv1d_57_input 有 3 个维度，但得到了形状为 (152, 64) 的数组

Question

我收到此错误：

ValueError: 检查输入时出错：预期 conv1d_57_input 有 3 个维度，但得到的数组形状为 (152, 64)。

我的代码：

model = Sequential()

model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(152,64)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Dropout(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dense(4, activation='softmax'))

model.summary()

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

history = model.fit(trainingMatrix, labelTraining, validation_data=(validationMatrix, labelValidation), epochs=3)

变量描述：

trainingMatrix.shape = (152,64); 行与具有特征的样本和列相关联。

是重塑问题吗？

编辑：

我做了以下更改：

trainingMatrix = np.expand_dims(trainingMatrix, axis=3)
validationMatrix = np.expand_dims(validationMatrix, axis=3)

model = Sequential()
model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(64,1)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Dropout(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dense(4, activation='softmax'))
model.summary()

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

history = model.fit(trainingMatrix, labelTraining, validation_data=(validationMatrix, labelValidation), epochs=3)

我收到这个新错误：检查目标时出错：预期 dense_28 的形状为 (1,) 但得到的数组的形状为 (4,)

我的总结：

_________________________________________________________________
Layer (type)                 Output Shape              Param    
=================================================================
conv1d_51 (Conv1D)           (None, 62, 64)            256       
_________________________________________________________________
conv1d_52 (Conv1D)           (None, 60, 64)            12352     
_________________________________________________________________
dropout_15 (Dropout)         (None, 60, 64)            0         
_________________________________________________________________
max_pooling1d_15 (MaxPooling (None, 30, 64)            0         
_________________________________________________________________
flatten_16 (Flatten)         (None, 1920)              0         
_________________________________________________________________
dense_27 (Dense)             (None, 100)               192100    
_________________________________________________________________
dense_28 (Dense)             (None, 4)                 404       
=================================================================
Total params: 205,112
Trainable params: 205,112
Non-trainable params: 0

新代码和新错误：

trainingMatrix = np.expand_dims(trainingMatrix, axis=0)
validationMatrix = np.expand_dims(validationMatrix, axis=0)

model = Sequential()

model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(152,64,1)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Dropout(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dense(4, activation='softmax'))
model.summary()

ValueError: Input 0 is in compatible with layer conv1d_57: expected ndim=3, found ndim=4

下面的方案可行，但命中率太低。有没有人推荐一个配置来改进？我没有达到超过 20% 的准确率。 （使用 MLP 我得到了 90%）

trainingMatrix = np.expand_dims(trainingMatrix, axis=3)
validationMatrix = np.expand_dims(validationMatrix, axis=3)

model = Sequential()

model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(64,1)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Dropout(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dense(4, activation='softmax'))

model.summary()

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

history = model.fit(trainingMatrix, labelTraining, validation_data=(validationMatrix, labelValidation), epochs=1000)

我的 labelTraining 是：

1 0 0 0
1 0 0 0
...
0 1 0 0
0 1 0 0
...
0 0 0 1

还好吗？

Answer 1

感谢大家的帮助。遵循以 97% 的准确率运行的代码。

trainingMatrix = np.expand_dims(trainingMatrix, axis=3)
validationMatrix = np.expand_dims(validationMatrix, axis=3)

model = Sequential()

model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(64,1)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(4, activation='softmax'))

model.summary()

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

history = model.fit(trainingMatrix, labelTraining, batch_size=batchSize, epochs=epochs, verbose=1, validation_data=(validationMatrix, labelValidation))

Answer 2

sparse_categorical_crossentropy & categorical_crossentropy

sparse_categorical_crossentropy：

在目标列包含多标签值且具有单个目标列时使用{表示标签可以大于1}

分类交叉熵

用于当目标 col 包含多标签值并且具有多目标 col{ 这意味着每个 col 目标值都是二进制的}

将 sparse_categorical_crossentropy 替换为 categorical_crossentropy