Repository navigation
Expand file tree
/
Copy pathLasso_Regression_sample.R
More file actions
114 lines (64 loc) · 2.94 KB
/
Copy pathLasso_Regression_sample.R
File metadata and controls
114 lines (64 loc) · 2.94 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
# Lasso (Least Absolute Shrinkage and Selection Operator)
# Method for automatically penalizing extra features (can set coefficient of feature to be zero)
# performs both variable selection and regularization
# ridge regression can't zero coefficients
# Another type of regularization method is ElasticNet, it is hybrid of lasso and ridge regression both.
## It is trained with L1 and L2 prior as regularizer.
### A practical advantage of trading-off between Lasso and Ridge is that, it allows Elastic-Net to inherit some of Ridge's stability under rotation
# in regular regression we just minimize SSE whereas in lasso regression we minimize SSE + lambda*modulus of beta
# overfitting can be adressed by 2 options 1) reduce no of features 2) regularization - keep all features but reduce magnitude of parameters
# installing glmnet package
# loading glmnet package
library(glmnet)
# glmnet perfroms elasticnet regression whihc is combination lasso and ridge we can only perform lasso by just specifying lasso
# glmnet works with matrices
library(dplyr)
setwd('D:/Float/20160712')
load('usum.Rdata')
# used complete.cases to keep only rows which are complete
usum1 = usum[,-c(1:4,6:13)]
# counting number of nas in each column so as to eliminate those which has significant nas
na_count <- apply(usum1, 2, function(x) sum(length(which(is.na(x)))))
#column nos with na's less than 100
Col <- match(names(na_count[na_count <= 40]),names(usum1))
usum1 = usum1[,Col]
# removing rows with NAs
usum2 = usum1[complete.cases(usum1),]
# removing high negative balace
usum2 <- usum2[usum2$accounts.balances.checking > -5000, ]
X = usum2[,-1]
X = as.matrix(X)
Y = usum2$RVbankCard
#a <- summary(lm(formula = RVbankCard ~ ., data = usum2))
#b <- a$coefficients
# Cross-validation to find optimal lambda
CV = cv.glmnet(x = X,y = Y,family = "gaussian",type.measure = "deviance", alpha = 1, nlambda = 100)
plot(CV)
# To know the attributes
attributes(CV)
# calcualte the minimum lambda
min_lambda = CV$lambda.min
# Construct lasso model
lasso.model = glmnet(x= X,y = Y,family = "gaussian",type.gaussian = "covariance", alpha = 1, lambda = min_lambda)
attributes(lasso.model)
coeff <- lasso.model$beta[,1]
sign_coeff <- coeff[coeff!=0]
print(sign_coeff)
# model using logistic regression
X = usum2[,-1]
X = as.matrix(X)
Y = usum2$RVbankCard
Y[Y < 660] = 0
Y[Y >= 660] = 1
# Cross-validation to find optimal lambda
CV = cv.glmnet(x = X,y = Y,family = "binomial",type.measure = "class", alpha = 1, nlambda = 100)
plot(CV)
# calcualte the minimum lambda
min_lambda = CV$lambda.min
# Construct lasso model
lasso.model = glmnet(x= X,y = Y,family = "binomial",type.logistic = "modified.Newton", alpha = 1, lambda = min_lambda)
attributes(lasso.model)
coeff <- lasso.model$beta[,1]
sign_coeff <- coeff[coeff!=0]
print(sign_coeff)
write.csv(sign_coeff,'coeff_lass.csv')