多次運行，對CNN上圖像分類的准確性有影響嗎？

Question

我是初學者。 我通過以下程序在 Jupyter 筆記本上使用 MNIST 嘗試在 CNN model 上進行圖像分類。 從“model.fit_generator”連續運行兩次（或多次）時，第二次運行的准確性受第一次運行結果的影響？ 看到顯示的結果，我認為第二次運行似乎已經有了更好的准確性。 請給我一些建議...

# import library
import keras
import matplotlib.pyplot as plt
import numpy as np
import os
import pandas as pd
import seaborn as sn
import shutil
import tensorflow as tf
from datetime import datetime, timedelta, timezone
from keras import backend as ke
from keras.callbacks import Callback, ModelCheckpoint, EarlyStopping
from keras.datasets import mnist
from keras.layers import Dense, Dropout, Activation, Flatten, Conv2D, MaxPooling2D, BatchNormalization
from keras.models import Sequential
from keras.optimizers import RMSprop
from keras.preprocessing.image import ImageDataGenerator
from keras.utils import np_utils
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import train_test_split
from tqdm import tqdm

# MNIST
mnist=keras.datasets.mnist
(x_train,y_train),(x_test,y_test)=mnist.load_data()
(x_train,y_train),(x_test,y_test)=(x_train[:80],y_train[:80]),(x_test[:20], y_test[:20])
#(x_train,y_train),(x_test,y_test)=(x_train[:160],y_train[:160]),(x_test[:40], y_test[:40])
#(x_train,y_train),(x_test,y_test)=(x_train[:800],y_train[:800]),(x_test[:200], y_test[:200])
#(x_train,y_train),(x_test,y_test)=(x_train[:8000],y_train[:8000]),(x_test[:2000], y_test[:2000])
x_train=x_train.reshape(x_train.shape[0],28,28,1)
x_test=x_test.reshape(x_test.shape[0],28,28,1)
x_train=x_train/255
x_test=x_test/255
print("x_train",x_train.shape)
print("x_test",x_test.shape)

# model
model = Sequential()
model.add(Conv2D(64, (3, 3), input_shape=(28,28,1), padding='same'))
BatchNormalization(axis=-1)
model.add(Activation('relu'))
model.add(Conv2D(64, (3, 3), padding='same'))
BatchNormalization(axis=-1)
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.20))
model.add(Conv2D(64, (3, 3), padding='same'))
BatchNormalization(axis=-1)
model.add(Activation('relu'))
model.add(Conv2D(64, (3, 3), padding='same'))
BatchNormalization(axis=-1)
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.20))
model.add(Conv2D(128, (3, 3), padding='same'))
BatchNormalization(axis=-1)
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dense(10, activation='softmax'))
model.summary()

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

# model fit
model.fit(x_train,y_train,epochs=40)

# evoluate for test data
loss,acc=model.evaluate(x_test,y_test,verbose=2)
print('loss:','{:.3f}'.format(loss),'accuracy:','{:.3f}'.format(acc))

# ImageDataGenerator
datagen = ImageDataGenerator( 
    featurewise_center=False,
    samplewise_center=False, 
    featurewise_std_normalization=False, 
    samplewise_std_normalization=False, 
    zca_whitening=False,  
    rotation_range=10, 
    width_shift_range=0.1, 
    height_shift_range=0.1,  
    zoom_range=[2.0,0.1], 
    horizontal_flip=False, 
    vertical_flip=False)  
datagen.fit(x_train) 

datagent = ImageDataGenerator(
    featurewise_center=False,
    samplewise_center=False,  
    featurewise_std_normalization=False, 
    samplewise_std_normalization=False,  
    zca_whitening=False,  
    rotation_range=10, 
    width_shift_range=0.1,  
    height_shift_range=0.1,
    zoom_range=[2.0,0.1],
    horizontal_flip=False,
    vertical_flip=False)
datagent.fit(x_test)

# parameter
epochs = 100
iteration_train = 5
iteration_test = 2
batch_size_train = int(x_train.shape[0] / iteration_train)
batch_size_test = int(x_test.shape[0] / iteration_test)

gen_train_flow = datagen.flow(x_train, y_train, batch_size=batch_size_train) 
gen_test_flow  = datagent.flow(x_test, y_test, batch_size=batch_size_test) 
history = model.fit(gen_train_flow,
                    steps_per_epoch=iteration_train,
                    epochs=epochs,
                    validation_data=gen_test_flow,
                    validation_steps=iteration_test)#,
                    #callbacks=callbacks)

# evoluate for test data
loss,acc=model.evaluate(x_test,y_test,verbose=2)
print('loss:','{:.3f}'.format(loss),'accuracy:','{:.3f}'.format(acc))

# graph for training
acc=history.history['accuracy']#acc
val_acc=history.history['val_accuracy']#val_acc
epochs=range(1,len(acc)+1)
plt.plot(epochs,acc,'b',label='Training accuracy')
plt.plot(epochs,val_acc,'r',label='Val accuracy')
plt.legend()
plt.show()

Answer 1

你訓練你的 CNN 越多，它就越會開始學習噪音。

因此，它在訓練准確度上的表現會非常好，但驗證准確度會下降 go，因為它不能很好地概括它沒有看到的數據。

查看更多信息： 偏差-方差權衡