序言：常言道，在實(shí)踐中學(xué)習(xí)，在理論中提升。輾輾轉(zhuǎn)轉(zhuǎn)看了幾本神經(jīng)網(wǎng)絡(luò)的書，磕磕絆絆地了解了一些概念，對(duì)于神經(jīng)網(wǎng)絡(luò)還是懵懵懂懂。在matlab里跑代碼，純調(diào)用庫函數(shù)，真的是閉著眼睛開車。幸運(yùn)的是，趕巧遇到了以下兩本神作，對(duì)神經(jīng)網(wǎng)絡(luò)的本質(zhì)算是一知半解。

• Python神經(jīng)網(wǎng)絡(luò)編程，塔里克.拉希德著，人民郵電出版社出版。

• 深度學(xué)習(xí)圖解，安德魯.特拉斯克著，清華大學(xué)出版社出版。

對(duì)于神經(jīng)網(wǎng)絡(luò)來說，選擇一個(gè)好的評(píng)估函數(shù)、訓(xùn)練函數(shù)、傳遞函數(shù)后，使用合適的大量的訓(xùn)練數(shù)據(jù)訓(xùn)練則成為關(guān)鍵。學(xué)習(xí)的關(guān)鍵在于調(diào)整網(wǎng)絡(luò)的權(quán)值使誤差最小。

看書至今，由于懶，代碼不曾寫過一行。今天是周日，桂林天寒地凍，山里風(fēng)大。看到了《深度學(xué)習(xí)圖解》的第6章，勾起了興趣，瞧一瞧書本上的樣例程序，照葫蘆畫瓢，越畫越像。

拋去所有背景知識(shí)，直接上代碼。在書本的范例基礎(chǔ)上，引入了函數(shù)。模塊化編程，簡(jiǎn)單就是美。

# streetlightlearning.py

# Create a backprogation network to learn a streetlights

# author: andrew W.Trask & Icefish

import numpy as np

weights = np.array([0.5, 0.48, -0.7])

alpha = 0.1

# define input as streetlights and output as walk_vs_stop

streetlights = np.array([[1, 0, 1],

                        [0, 1, 1],

                        [0, 0, 1],

                        [1, 1, 1],

                        [0, 1, 1],

                        [1, 0, 1]])

walk_vs_stop = np.array([[0],

                        [1],

                        [0],

                        [1],

                        [1],

                        [0]])

##input_data = streetlights[0]

##goal_prediction = walk_vs_stop[0]

##

##for iteration in range(1000):

##    prediction = input_data.dot(weights)

##    error = (goal_prediction - prediction)**2

##    delta = prediction - goal_prediction

##    weights = weights - (alpha*(input_data*delta))

##    print("Error:"+str(error)+"Prediction:"+str(prediction))

# 訓(xùn)練函數(shù)，迭代次數(shù)可自定義

def train(iteration_times):

    global weights

    global streetlights

    global walk_vs_stop



    for iteration in range(iteration_times):

        error_for_all_lights = 0

        for row_index in range(len(walk_vs_stop)):

            input_data = streetlights[row_index]

            goal_prediction = walk_vs_stop[row_index]

            prediction = input_data.dot(weights) # output of the network

            error = (prediction - goal_prediction)**2

            error_for_all_lights = error_for_all_lights + error

            delta = prediction - goal_prediction

            weights = weights - (alpha*(input_data*delta))

            print("Prediction:"+str(prediction))

    print("Error:"+str(error_for_all_lights)+"\n")

def sim(streetlights_input):

    global weights;

    return streetlights_input.dot(weights)

def main():

    input_str = input("請(qǐng)輸入網(wǎng)絡(luò)訓(xùn)練迭代次數(shù):")

    iteration_times = int(input_str)

    if iteration_times>0 :

        train(iteration_times)

    else:

        print("迭代次數(shù)輸入有誤，程序終止！")

        return

    #訓(xùn)練完成，提示用戶使用訓(xùn)練好的網(wǎng)絡(luò)

    while(True):

        a,b,c = input("輸入信號(hào)燈狀態(tài)，用逗號(hào),隔開:").split(',')

        a = int(a)

        b = int(b)

        c = int(c)

        input_data = np.array([a,b,c])

        prediction = sim(input_data)

        status = int(round(prediction,0))

        print("Prediction:"+str(status)+"\n")

        if status == 1:

            print("walk now is safe.\n")

        else:

            print("stop,or walk to hell.\n")



if __name__ == '__main__':

    main()

運(yùn)行OK