結合實用數據分析該書，整理了下代碼，記錄以作備忘和分享：
注：其中用到mlpy（機器學習庫）,安裝會出現問題，可參考文末引用文章的處理方法。
1 # -*- coding: utf-8 -*-
2 """
3 Created on Wed Oct 17 21:14:44 2018 4
5 @author: Luove 6 """
7 # KRR適合分類和回歸訓練集很少時，非線性方法
8 import os 9 import numpy as np 10 import matplotlib.pyplot as plt 11 import dateutil.parser as dparser # dateutil模塊主要有兩個函數，parser和rrule。parser是根據字符串解析成datetime，而rrule是則是根據定義的規則來生成datetime;https://blog.csdn.net/cherdw/article/details/55224341
12 from pylab import * # 將matplotlib和numpy封裝在一起，模擬MATLAB編程環境
13 from sklearn.cross_validation import train_test_split 14 from sklearn import linear_model 15 from sklearn import datasets 16 import mlpy 17 from mlpy import KernelRidge 18
19 # np.hamming 漢明窗，構造一個函數（僅處理窗內數據）。這個函數在某一區間有非零值，而在其余區間皆為0.漢明窗就是這樣的一種函數
20 # 階梯圖，又叫瀑布圖，可以用於企業成本、銷售等數據的變化和構成情況的分析；plot.step()
21 x1 = np.linspace(1,100,500) 22 x2 = np.linspace(1,100,50) 23 y1 = np.cos(x1) 24 y2 = np.cos(x2) 25
26 axs1 = plt.subplot(211) 27 axs2 = plt.subplot(212) 28 axs1.step(x1,y1) 29 axs2.step(x2,y2) 30 plt.show() 31
32
33 goldfile = "D:\Analyze\Python Matlab\Python\BookCodes\PDA_Book-master\PDA_Book-master\Chapter7\Gold.csv"
34 # tsa,時間序列分析，將時間序列平滑化，（本身包含：趨勢T，季節性/周期性S，波動性V）
35 def smooth(x,window_length): 36 s = np.r_[2*x[0]-x[window_length-1::-1], x, 2*x[-1]-x[-1:-window_length:-1]] 37 w = np.hamming(window_length) 38 y = np.convolve(w/w.sum(), s, mode='same') # 卷積函數，移動平均濾波（平滑方法）,第一個參數長度要大於等於第二參數長度，否則會交換位置；mode={'full','same','valid'}，默認full
39 return y[window_length:-window_length+1] 40
41 # 金價走勢,注意下面dtype變化：日期用object,值用None（各列內容識別，）
42 x = np.genfromtxt(goldfile,dtype='object',delimiter=',',skip_header=1,usecols=(0),converters={0:dparser.parse}) # 第一列日期，dateutil.parser.parse,字符串中解析出日期
43 y = np.genfromtxt(goldfile,dtype=None,delimiter=',',skip_header=1,usecols=(1)) # 獲取第二列
44 y_smoothed = smooth(y,len(y)) 45 plt.step(x,y,'r*',label='raw data') 46 plt.step(x,y_smoothed,label='smoothed data') 47 plt.legend() 48 #x = [2,3,9,634,32,4,676,4,234,43,7,-13,0]
49 #x = np.array(x)
50 #np.round(smooth(x,len(x)))
51 #[ 33., 80., 124., 165., 189., 199., 192., 169., 137., 104., 66., 35., 16.]
52 #plt.plot(x)
53 #plt.plot(np.round(smooth(x,len(x)))) # 加載pylab,不必plt.show()?
54 ##plt.show()
55 #window_length=x.shape[0]
56
57 house = datasets.load_boston() 58 houseX = house.data[:,np.newaxis] # 添加一個新軸，添加一維度，由(506, 13)轉成(506, 1，13)
59 houseX_temp = houseX[:,:,2] 60
61 x_train,xtest,ytrain,ytest=train_test_split(houseX_temp,house.target,test_size=1.0/3) 62 lreg = linear_model.LinearRegression() 63 lreg.fit(x_train,ytrain) 64 plt.scatter(xtest,ytest,color='green') 65 plt.plot(xtest,lreg.predict(xtest),color='blue',linewidth=2) 66
67 np.random.seed(0) 68 targetvalues = np.genfromtxt(goldfile,skip_header=1,dtype=None,delimiter=',',usecols=(1)) # usecols篩選感興趣列
69 type(targetvalues) 70 trainingpoints = np.arange(125).reshape(-1,1) # transform ,轉換成一列，行自適應
71 testpoint = np.arange(126).reshape(-1,1) 72 knl = mlpy.kernel_gaussian(trainingpoints,trainingpoints,sigma=1) # 訓練核矩陣,對稱半正定,(125, 125)
73 knltest = mlpy.kernel_gaussian(testpoint,trainingpoints,sigma=1) # 測試核矩陣,(126, 125)
74
75 knlridge = KernelRidge(lmb=0.01) 76 knlridge.learn(knl,targetvalues) 77 resultpoints = knlridge.pred(knltest) 78
79 fig = plt.figure(1) 80 plt.plot(trainingpoints,targetvalues,'o') 81 plt.plot(testpoint,resultpoints) 82 #plt.show()
83 len(resultpoints) 84 resultpoints[-5:-1] 85
86 # 采用平滑后的數據,即smooth后的targetvalues
87 targetvalues_smoothed = smooth(targetvalues,len(targetvalues)) 88 knlridge.learn(knl,targetvalues_smoothed) 89 resultpoints_smoothed = knlridge.pred(knltest) 90 plt.step(trainingpoints,targetvalues_smoothed,'o') 91 plt.step(testpoint,resultpoints_smoothed) 92 #plt.show()
93 len(resultpoints_smoothed) 94 resultpoints_smoothed[-5:-1] # 平滑前126期預測值：1389.8;平滑后126期預測值1388.6
95 #x = np.arange(0, 2, 0.05).reshape(-1, 1) # training points
96 #y = np.ravel(np.exp(x)) + np.random.normal(1, 0.2, x.shape[0]) # target values
97 #xt = np.arange(0, 2, 0.01).reshape(-1, 1) # testing points
98 #K = mlpy.kernel_gaussian(x, x, sigma=1) # training kernel matrix
99 #Kt = mlpy.kernel_gaussian(xt, x, sigma=1) # testing kernel matrix
100 #krr = KernelRidge(lmb=0.01)
101 #krr.learn(K, y)
102 #yt = krr.pred(Kt)
103 #fig = plt.figure(1)
104 #plot1 = plt.plot(x[:, 0], y, 'o')
105 #plot2 = plt.plot(xt[:, 0], yt)
106 #plt.show()
其中，mlpy.KernelRidge模型參數lmb（正則化參數），設定越小，擬合趨勢和原趨勢基本一致，如下圖：分別是lmb=0.01,lmb=1(默認)
而正則化參數意義文檔中解釋不清，詳細可參考引用的文章，解釋比較好，摘取部門截圖如下：
Ref:
Windows下Python模塊-----mlpy（機器學習庫）的安裝（本文未按此操作，有用的可以給咱交流下啊）
pip安裝MLPY庫 （安裝推薦按此操作）
《實用數據分析》：文中數據及mlpy文檔需要可自取：https://github.com/Luove/Data