我使用keras
训练一个CNN,基本误差是维数不匹配。
原因是,调试后:
print("Before")
print(TX.shape)
print(TeX.shape)
X_train = TX.reshape(1000, 1, img_rows, img_cols)
X_test = TeX.reshape(430, 1, img_rows, img_cols)
print("After")
print(TX.shape)
print(TeX.shape)
产生产出:
Using Theano backend.
Using gpu device 0: GeForce GTX 750 Ti (CNMeM is disabled, CuDNN not available)
Before
(1000, 27, 36)
(430, 27, 36)
After
(1000, 27, 36)
(430, 27, 36)
如果需要,我的模型总结如下:
____________________________________________________________________________________________________
图层(类型)输出形状Param #连接到
convolution2d_1 (Convolution2D) (无,32,25,34) 320 convolution2d_input_1
activation_1 (激活)(无,32,25,34) 0 convolution2d_1
convolution2d_2 (Convolution2D) (无,32,23,32) 9248 activation_1
activation_2 (激活)(无,32,23,32) 0 convolution2d_2
convolution2d_3 (Convolution2D) (无,32,21,30) 9248 activation_2
activation_3 (激活)(无,32,21,30) 0 convolution2d_3
maxpooling2d_1 (MaxPooling2D) (无,32,10,15) 0 activation_3
dropout_1 (Dropout) (无,32,10,15) 0 maxpooling2d_1
flatten_1 (平坦)(无,4800) 0 dropout_1
dense_1 (稠密)(无,128个) 614528 flatten_1
activation_4 (激活)(无,128个)0 dense_1
dropout_2 (Dropout) (无,128个)0 activation_4
dense_2 (稠密)(无,26) 3354 dropout_2
activation_5 (激活)(无,26) 0 dense_2
共: 636698人
发布于 2016-04-29 05:25:16
您正在将经过整形的数组分配给一个新变量,但是仍然要打印旧变量的形状:
X_train = TX.reshape(..)
你必须使用:
print(X_train.shape)
https://stackoverflow.com/questions/36939414
复制