平行地球: 用Keras多元感知器(MLP)辨識Mnist手寫數字影像

我是用google的Colab上的TPU跑的
Mnist是一堆灰階的手寫數字影像，利用AI來訓練圖片上的數字正不正確
利用MLP的功能把Mnist的資料來做比對，把圖片區分為點，再來抓取特徵比對

!pip install --upgrade tensorflow keras #更新tensorflow跟keras，更新完要登出再登入

import numpy as np #載入相關模組
import pandas as pd
from keras.utils import np_utils
np.random.seed(10)

Using TensorFlow backend.

from keras.datasets import mnist #匯入mnist資料，如果沒有就下載
(X_train_image, y_train_label), (X_test_image, y_test_label) = mnist.load_data() #讀取Mnist資料

print('train data=', len(X_train_image))
print('test data=', len(X_test_image))
train data= 60000 test data= 10000 #輸出訓練資料有60000筆，測試資料有10000筆

print('train_image=', X_train_image.shape)
print('train label=', y_train_label.shape)
train_image= (60000, 28, 28) train label= (60000,) #28X28的圖片，60000筆數字影像，60000數字影像真實的值

import matplotlib.pyplot as plt #利用matplotlib來畫圖，建立函數
def plot_image(image):
fig = plt.gcf()
fig.set_size_inches(2,2)
plt.imshow(image, cmap='binary')
plt.show()

plot_image(X_train_image[0]) #把函數plot_image的框架套進來輸出

y_train_label[0] #第0筆資料的真實值
5

def plot_images_labels_predict(images, labels, prediction, idx, num=10): #建立函數
fig = plt.gcf()
fig.set_size_inches(12, 14)
if num > 25: num = 25
for i in range(0, num):
ax=plt.subplot(5,5, 1+i)
ax.imshow(images[idx], cmap='binary')
title = "l=" + str(labels[idx])
if len(prediction) > 0:
title = "l={},p={}".format(str(labels[idx]), str(prediction[idx]))
else:
title = "l={}".format(str(labels[idx]))
ax.set_title(title, fontsize=10)
ax.set_xticks([]); ax.set_yticks([])
idx+=1
plt.show()

plot_images_labels_predict(X_train_image, y_train_label, [], 0, 10) #畫圖

print('test_image=', X_test_image.shape) #查看測試筆數
print('test_label=', y_test_label.shape)
test_image= (10000, 28, 28) test_label= (10000,)

plot_images_labels_predict(X_test_image, y_test_label, [], 0, 10) #畫圖

x_Train = X_train_image.reshape(60000, 784).astype('float32') #轉換為784個32位元的浮點數
x_Test = X_test_image.reshape(10000, 784).astype('float32')

xTrain= (60000, 784) xTest= (10000, 784)

X_train_image[0]
#查看第0筆資料的內容，就是數字5，每個數字都是灰階的圖片，所以顏色由深到淺，就是255到0

x_Train_norm = x_Train/255 #再除以255，就會有小數點，差異化就會變大，特徵越明顯
x_Test_norm = x_Test/255

x_Train_norm[0]

y_train_label[:10]
array([5, 0, 4, 1, 9, 2, 1, 3, 1, 4], dtype=uint8)

y_TrainOneHot = np_utils.to_categorical(y_train_label)
y_TestOneHot = np_utils.to_categorical(y_test_label)

y_TrainOneHot[:10] #輸出前10筆資料

0000010000 =5 1對下來就是5
1000000000 =0
0123456789

等一下要建立訓練模型
輸入層是784，就是剛剛要放進來訓練的資料。
隱藏層等一下要建，常理來說數值是越大越好，但是會有過度擬合的問題，運算也需要時間。
輸出層是10，因為只有0到9個數字

from tensorflow.python.keras.layers import Dense
from tensorflow.python.keras import Sequential #匯入相關模組

model = Sequential()
#建立模型，輸入層神經元784個，隱藏層神經元256個，輸出層神經元10個

model.add(Dense(units=256, input_dim=784, kernel_initializer='normal', activation='relu'))
model.add(Dense(units=10, kernel_initializer='normal', activation='softmax'))

print(model.summary()) #查看模型摘要
200960=784*256+256
2570=256*10+10
203530=200960+2570

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
train_history = model.fit(x=x_Train_norm, y=y_TrainOneHot, validation_split=0.2, epochs=10, batch_size=200, verbose=2)
#訓練模型，48000=60000*0.8 80%作為訓練資料
12000=60000*0.2 20%作為驗證資料
10次的訓練週期
每批200筆訓練資料，48000/200=240批
Loss誤差越小，準確率越高

import matplotlib.pyplot as plt #畫圖
def show_train_history(train_history, train, validation):
plt.plot(train_history.history[train])
plt.plot(train_history.history[validation])
plt.title('Train History')
plt.ylabel(train)
plt.xlabel('Epoch')
plt.legend(['train', 'validation'], loc='upper left')
plt.show()

show_train_history(train_history, 'accuracy', 'val_accuracy')