I know how to "extract" coefficient from cv.glmnet. But is there a way to change the coefficient of a fitted object? I know it's kind of a hack, but I want to manually tweak the coefficients of a fitted object just for my own experiment.
Extracting coefficient variable names from glmnet into a data.frame
The coefficients for glmnet are stored under $glmnet.fit$beta :
library(glmnet)
library(Matrix)
fit = cv.glmnet(x=as.matrix(mtcars[,-1]),y=mtcars[,1])
head(fit$glmnet.fit$beta)
6 x 79 sparse Matrix of class "dgCMatrix"
[[ suppressing 79 column names ‘s0’, ‘s1’, ‘s2’ ... ]]
cyl . -0.01192151 -0.1447790 -0.2658654 -0.376195 -0.4770942 -0.5686616
disp . . . . . . .
hp . . . . . . .
drat . . . . . . .
wt . -0.45776130 -0.7006176 -0.9218541 -1.123436 -1.3065806 -1.4739808
qsec . . . . . . .
It's one column for every lambda tested and when you call coefficients(), by default you pull out the column corresponding to fit$lambda.1se. Let's say we want to change that column:
coefficients(fit,s=fit$lambda.1se)
11 x 1 sparse Matrix of class "dgCMatrix"
1
(Intercept) 33.940487806
cyl -0.843038418
disp .
hp -0.006965929
drat .
wt -2.365917424
qsec .
vs .
am .
gear .
carb .
wh = which(fit$lambda==fit$lambda.1se)
fit$glmnet.fit$beta[,wh] = runif(nrow(fit$glmnet.fit$beta))
coefficients(fit,s=fit$lambda.1se)
11 x 1 sparse Matrix of class "dgCMatrix"
1
(Intercept) 33.94048781
cyl 0.45636267
disp 0.28286532
hp 0.04186184
drat 0.55084730
wt 0.35273817
qsec 0.96165338
vs 0.79227125
am 0.01036681
gear 0.47738589
carb 0.17170791
And other columns and the intercept remains unchanged:
11 x 1 sparse Matrix of class "dgCMatrix"
1
(Intercept) 36.38102033
cyl -0.87610600
disp .
hp -0.01377978
drat .
wt -2.75934250
qsec .
vs .
am 0.16806977
gear .
carb -0.01384960
It is really weird to do it. If you want to change the coefficients for prediction purposes, you can always pull out coefficients, alter and multiply them with your data matrix again.