ML之PLiR之Glmnet：利用Glmnet算法求解ElasticNet回归类型问题(实数值评分预测)

ML之PLiR之Glmnet算法：利用Glmnet算法求解ElasticNet回归类型问题(实数值评分预测)

---

输出结果

0 2
1 2
2 2
3 3
4 3
5 3
6 3
7 3
8 3
9 2
10 2
11 2
12 2
13 3
14 3
15 2
16 2
17 2
18 2
19 2
20 2
21 2
22 2
23 2
24 2
25 2
26 2
27 2
28 2
29 3
30 3
31 3
32 2
33 3
34 2
35 2
36 2
37 2
38 2
39 2
40 2
41 2
42 1
43 1
44 2
45 2
46 2
47 1
48 2
49 1
50 1
51 1
52 1
53 1
54 1
55 1
……

95 1
96 1
97 1
98 1
99 1
['"alcohol"', '"volatile acidity"', '"sulphates"', '"total sulfur dioxide"', '"chlorides"', '"fixed acidity"', '"pH"', '"free sulfur dioxide"', '"residual sugar"', '"citric acid"', '"density"']

1、Glmnet算法

实现代码

#calculate starting value for lambda
lam = maxXY/alpha

#this value of lambda corresponds to beta = list of 0's
#initialize a vector of coefficients beta
beta = [0.0] * ncols

#initialize matrix of betas at each step
betaMat = []
betaMat.append(list(beta))

#begin iteration
nSteps = 100
lamMult = 0.93 #100 steps gives reduction by factor of 1000 in
               # lambda (recommended by authors)
nzList = []

for iStep in range(nSteps):
    #make lambda smaller so that some coefficient becomes non-zero
    lam = lam * lamMult

    deltaBeta = 100.0
    eps = 0.01
    iterStep = 0
    betaInner = list(beta)
    while deltaBeta > eps:
        iterStep += 1
        if iterStep > 100: break

        #cycle through attributes and update one-at-a-time
        #record starting value for comparison
        betaStart = list(betaInner)
        for iCol in range(ncols):

            xyj = 0.0
            for i in range(nrows):
                #calculate residual with current value of beta
                labelHat = sum([xNormalized[i][k]*betaInner[k]
                                for k in range(ncols)])
                residual = labelNormalized[i] - labelHat

                xyj += xNormalized[i][iCol] * residual

            uncBeta = xyj/nrows + betaInner[iCol]
            betaInner[iCol] = S(uncBeta, lam * alpha) / (1 +
                                            lam * (1 - alpha))

        sumDiff = sum([abs(betaInner[n] - betaStart[n])
                       for n in range(ncols)])
        sumBeta = sum([abs(betaInner[n]) for n in range(ncols)])
        deltaBeta = sumDiff/sumBeta
    print(iStep, iterStep)
    beta = betaInner

    #add newly determined beta to list
    betaMat.append(beta)

    #keep track of the order in which the betas become non-zero
    nzBeta = [index for index in range(ncols) if beta[index] != 0.0]
    for q in nzBeta:
        if (q in nzList) == False:
            nzList.append(q)