model.load_weights() 给出不正确的结果

Question

我使用以下 CNN 模型训练 MNIST 数据并将权重保存为 mnist_weights.h5 以重现结果。

import keras
from __future__ import print_function
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
import numpy as np
from sklearn.model_selection import train_test_split

batch_size = 128
num_classes = 3
epochs = 4

# input image dimensions
img_rows, img_cols = 28, 28

#Just for reducing data set 
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x1_train=x_train[y_train==0]; y1_train=y_train[y_train==0]
x1_test=x_test[y_test==0];y1_test=y_test[y_test==0]
x2_train=x_train[y_train==1];y2_train=y_train[y_train==1]
x2_test=x_test[y_test==1];y2_test=y_test[y_test==1]
x3_train=x_train[y_train==2];y3_train=y_train[y_train==2]
x3_test=x_test[y_test==2];y3_test=y_test[y_test==2]

X=np.concatenate((x1_train,x2_train,x3_train,x1_test,x2_test,x3_test),axis=0)
Y=np.concatenate((y1_train,y2_train,y3_train,y1_test,y2_test,y3_test),axis=0)

# the data, shuffled and split between train and test sets
x_train, x_test, y_train, y_test = train_test_split(X,Y)

if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Conv2D(1, kernel_size=(2, 2),
                 activation='relu',
                 input_shape=input_shape))
model.add(MaxPooling2D(pool_size=(16,16)))
model.add(Flatten())
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])

model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          verbose=1,
          validation_data=(x_test, y_test))

model.save_weights('mnist_weights.h5')

现在我正在使用相同的模型和相同的数据集来重现结果，所以我从上面的代码中保存了一个负载权重。（代码如下）

import keras
from __future__ import print_function
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
import numpy as np
from sklearn.model_selection import train_test_split

batch_size = 128
num_classes = 3
epochs = 1

# input image dimensions
img_rows, img_cols = 28, 28

#Just for reducing data set 
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x1_train=x_train[y_train==0]; y1_train=y_train[y_train==0]
x1_test=x_test[y_test==0];y1_test=y_test[y_test==0]
x2_train=x_train[y_train==1];y2_train=y_train[y_train==1]
x2_test=x_test[y_test==1];y2_test=y_test[y_test==1]
x3_train=x_train[y_train==2];y3_train=y_train[y_train==2]
x3_test=x_test[y_test==2];y3_test=y_test[y_test==2]

X=np.concatenate((x1_train,x2_train,x3_train,x1_test,x2_test,x3_test),axis=0)
Y=np.concatenate((y1_train,y2_train,y3_train,y1_test,y2_test,y3_test),axis=0)

# the data, shuffled and split between train and test sets
x_train, x_test, y_train, y_test = train_test_split(X,Y)

if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Conv2D(1, kernel_size=(2, 2),
                 activation='relu',
                 input_shape=input_shape,trainable=False))
model.add(MaxPooling2D(pool_size=(16,16)))
model.add(Flatten())
model.add(Dense(num_classes, activation='softmax',trainable=False))
model.load_weights('mnist_weights.h5')
model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])

model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          verbose=1,
          validation_data=(x_test, y_test))

model.save_weights('mnist_weights1.h5')

最后，当我检查两个代码的准确性时，两者都不同，为什么会这样，当我提供相同的模型和相同的权重时。（我正在使用 1 个纪元和 trainable=False）

Answer 1

准确率不同，因为数据集的拆分方式不同。

x_train, x_test, y_train, y_test = train_test_split(X,Y)

如果您向 train_test_split 提供 random_state 参数，您应该会看到两个代码片段给出相同的 val_acc.

model.load_weights() 给出不正确的结果

model.load_weights() giving incorrect results

keras

keras-layer

keras-2