Keras_深度学习_MNIST数据集手写数字识别之各种调参

注：这里的代码是听台大李宏毅老师的ML课程敲的相应代码。

• 先各种import

import numpy as np
np.random.seed(1337)

# https://keras.io/
!pip install -q keras
import keras

from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation
from keras.layers import Convolution2D, MaxPooling2D, Flatten
from keras.optimizers import SGD, Adam
from keras.utils import np_utils
from keras.datasets import mnist
#categorical_crossentropy

• 再定义函数load_data（）

def load_data():
  (x_train, y_train), (x_test, y_test) = mnist.load_data()
  
  number = 10000
  x_train = x_train[0 : number]
  y_train = y_train[0 : number]
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.astype('float32')
  x_test = x_train.astype('float32')
  
  #convert class vectors to binary class matrices
  
  y_train = np_utils.to_categorical(y_train, 10)
  y_test = np_utils.to_categorical(y_test, 10)
  x_train = x_train
  x_test = x_test
  
  #x_test = np.random.normal(x_test)
  
  x_train = x_train/255
  x_test = x_test/255
  
  return (x_train, y_train), (x_test, y_test)

• 第一次运行测试

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'mse', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\nTest Acc:', result[1])

运行结果如下：

说明training的时候就没有train好。

• 修改1：把loss由mse改为categorical_crossrntropy

#修改1：把loss由mse改为categorical_crossrntropy

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))

model.add(Dense(units = 10, activation = 'softmax'))

#这里做了修改
model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy']) 

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\nTest Acc:', result[1])

运行结果如下：

training的结果由11%提升到85%(test acc却更小了)。

• 修改2：在修改1的基础上，fit时的batch_size由100调整为10000（这样GPU能发挥它平行运算的特点，会计算的很快）

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 10000, epochs = 20) #这里做了修改

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\nTest Acc:', result[1])

运行结果如下：

performance很差。

• 修改3：在lossfunc用crossentropy的基础上，fit时的batch_size由10000调整为1（这样GPU就不能发挥它平行运算的效能，会计算的很慢）

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 1, epochs = 20) #这里做了修改

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\nTest Acc:', result[1])

没有等到运行结果出来……

• 修改4：在修改1的基础上，改成deep，再加10层

#修改4：现在改成deep，再加10层

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))
model.add(Dense(units = 689, activation = 'sigmoid'))

#这里做了修改
for i in range(10):
  model.add(Dense(units = 689, activation = 'sigmoid'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

运行结果如下：

还是training就很差。

• 修改5：在修改4的基础上，把sigmoid都改成relu

#修改5：在修改4的基础上，把sigmoid都改成relu

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))

for i in range(10):
  model.add(Dense(units = 689, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

运行结果如下：

training效果非常好，但是test效果依然很差劲。

• 修改6：在修改5的基础上。load_data函数中，第26行和第27行，是除以255做normalize，如果把这两行注释掉：

# 修改6：load_data函数中，第26行和第27行，是除以255做normalize，如果把这两行注释掉，会发现又做不起来了
def load_data():
  (x_train, y_train), (x_test, y_test) = mnist.load_data()
  
  number = 10000
  x_train = x_train[0 : number]
  y_train = y_train[0 : number]
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.astype('float32')
  x_test = x_train.astype('float32')
  
  #convert class vectors to binary class matrices
  
  y_train = np_utils.to_categorical(y_train, 10)
  y_test = np_utils.to_categorical(y_test, 10)
  x_train = x_train
  x_test = x_test
  
  #x_test = np.random.normal(x_test)
  
#   x_train = x_train/255 #这里做了修改
#   x_test = x_test/255
  
  return (x_train, y_train), (x_test, y_test)



(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))

for i in range(10):
  model.add(Dense(units = 689, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

运行结果如下：

training效果又变得很好。

• 修改7:取消修改6中的操作，并把添加的10层注释掉再来一次

# 修改7:取消修改6中的操作，并把添加的10层注释掉再来一次
def load_data():
  (x_train, y_train), (x_test, y_test) = mnist.load_data()
  
  number = 10000
  x_train = x_train[0 : number]
  y_train = y_train[0 : number]
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.astype('float32')
  x_test = x_train.astype('float32')
  
  #convert class vectors to binary class matrices
  
  y_train = np_utils.to_categorical(y_train, 10)
  y_test = np_utils.to_categorical(y_test, 10)
  x_train = x_train
  x_test = x_test
  
  #x_test = np.random.normal(x_test)
  
  x_train = x_train/255
  x_test = x_test/255
  
  return (x_train, y_train), (x_test, y_test)



(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))

# for i in range(10):
#   model.add(Dense(units = 689, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

model.compile(loss = 'categorical_crossentropy', optimizer = SGD(lr = 0.1), metrics = ['accuracy'])

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

运行结果如下：

training效果又变得很好。

• 修改8: 换一下gradient descent strategy ,把SGD换为adam

# 修改8: 换一下gradient descent strategy ,把SGD换为adam

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))

# for i in range(10):
#   model.add(Dense(units = 689, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

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

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

此时虽然最后收敛得差不多，但是最后上升的速度加快了。

• 修改9：在testing set每个image故意加上noise。 

#修改9：在testing set每个image故意加上noise。 现在效果更差了。。。
def load_data():
  (x_train, y_train), (x_test, y_test) = mnist.load_data()
  
  number = 10000
  x_train = x_train[0 : number]
  y_train = y_train[0 : number]
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.reshape(number, 28*28)
  x_test = x_test.reshape(x_test.shape[0], 28*28)
  
  x_train = x_train.astype('float32')
  x_test = x_train.astype('float32')
  
  #convert class vectors to binary class matrices
  
  y_train = np_utils.to_categorical(y_train, 10)
  y_test = np_utils.to_categorical(y_test, 10)
  x_train = x_train
  x_test = x_test
  
  #x_test = np.random.normal(x_test)
  
  x_train = x_train/255
  x_test = x_test/255
  
  x_test = np.random.normal(x_test)
  
  return (x_train, y_train), (x_test, y_test)

(x_train, y_train), (x_test, y_test) = load_data()

model = Sequential()

model.add(Dense(input_dim = 28*28, units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))
model.add(Dense(units = 689, activation = 'relu'))

# for i in range(10):
#   model.add(Dense(units = 689, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

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

model.fit(x_train, y_train, batch_size = 100, epochs = 20)

result = model.evaluate(x_train, y_train, batch_size = 10000)
print('\nTrain Acc:', result[1])  

result = model.evaluate(x_test, y_test, batch_size = 10000)
print('\n Test Acc:', result[1])

运行结果如下：

现在效果更差了。。。