Python 检查输入时出错:预期conv1d_57_输入有3个维度,但得到了形状为(152,64)的数组
我得到了这个错误: ValueError:检查输入时出错:预期conv1d_57_输入有3个维度,但得到了形状为(152,64)的数组 我的代码:Python 检查输入时出错:预期conv1d_57_输入有3个维度,但得到了形状为(152,64)的数组,python,keras,cnn,Python,Keras,Cnn,我得到了这个错误: ValueError:检查输入时出错:预期conv1d_57_输入有3个维度,但得到了形状为(152,64)的数组 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')) mode
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)
变量说明:
trainingMatrix.shape=(152,64)代码>行与具有要素的样本和列相关联
这是一个重塑的问题吗
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)
我得到了这个新错误:检查目标时出错:预期密集型_28具有形状(1),但得到了具有形状(4)的数组
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:输入0与层conv1d_57不兼容:预期ndim=3,发现ndim=4
下面的解决方案有效,但命中率太低。是否有人建议对配置进行改进?我没有达到超过20%的准确率。(使用MLP我得到了90%)
我的labelTraining是:
1 0 0 0
1 0 0 0
...
0 1 0 0
0 1 0 0
...
0 0 0 1
可以吗?谢谢大家的帮助。按照代码工作,准确率为97%
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))
谢谢大家的帮助。按照代码工作,准确率为97%
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))
稀疏的\u分类的\u交叉熵和分类的\u交叉熵
稀疏\u分类\u交叉熵:
当目标列包含多标签值且具有单个目标列{表示标签可以大于1}时使用
范畴交叉熵
当目标列包含多标签值且具有多目标列{表示每个列目标值为二进制}时使用
用分类交叉熵替换稀疏分类交叉熵稀疏分类交叉熵和分类交叉熵
稀疏\u分类\u交叉熵:
当目标列包含多标签值且具有单个目标列{表示标签可以大于1}时使用
范畴交叉熵
当目标列包含多标签值且具有多目标列{表示每个列目标值为二进制}时使用
用分类交叉熵替换稀疏分类交叉熵152
您的样本数正确吗?在这种情况下,将输入形状更改为input\u shape=(64,)
。您不应该在输入形状中给出样本数,Keras会自动进行。您是否尝试过将其形状改为152,64,1或1152,64?@VivekMehta我根据您的建议更改了代码,但我收到了其他错误:输入0与层conv1d_1不兼容:预期ndim=3,发现ndim=2@3NiGMa在这两种情况下我都会遇到这个错误:输入0与层conv1d_6不兼容:预期ndim=3,发现ndim=4152
您的样本数正确吗?在这种情况下,将输入形状更改为input\u shape=(64,)
。您不应该在输入形状中给出样本数,Keras会自动进行。您是否尝试过将其形状改为152,64,1或1152,64?@VivekMehta我根据您的建议更改了代码,但我收到了其他错误:输入0与层conv1d_1不兼容:预期ndim=3,发现ndim=2@3NiGMa在这两种情况下我都会遇到此错误:输入0与层conv1d_6不兼容:预期ndim=3,发现ndim=4