I want to instruct a response prediction model while my sample size is small (18 patients) and about 100 features were extracted for each patient, so I need to use Lasso regression in R software. I entered my data into SPSS software. I need R codes for calling out the data into glmnet package and finding the correlation coefficient between dependant and independant variables. Who can help me?
Thanks so much,
Best Regards,
I worked with LASSO in a recent study (González-Rodríguez and Diéguez-Aranda, 2020) using the package "elasticnet". For doing only LASSO regression with this package you just need to set the lambda parameter = 0. For instance:
library(elasticnet)
cv.lasso.model
Dear Dr. Gonzalez-Rodriguez,
Thanks for your perfect answer. I have some questions:
Do I have to define a matrix for each variable? How can I convert the format of my data in spss, that is similar to data frame, to matrix?
And, do you work on feature selection in lasso, as well?
Thanks a lot.
Best Regards,
Well, I have never worked with SPSS, but I think that the easiest way of doing that could be writing your data as csv file and opening it from R using the function read.csv(). That imports the whole data table (supposing that your data structure is a table comprising a numeric response column "y" and a set of numeric predictor columns "X") as a dataframe object, which is a convenient object for managing data in R.
For instance, if you have your data in a file with name "mydata.csv", you can import it from R:
mydata = read.csv("mydata.csv"),
and supposing now that your response variable is the first column and the predictors are the rest of the columns you can pass the data to the lasso function by doing:
cv.enet(as.matrix( mydata[,-1] ), mydata[,1], K = 10, lambda=0, mode = "fraction", s = seq(0,1,0.01))
Concerning feature selection, it is automatically done depending on the value of relative shrinkage intensity (s) that you chose, depending on the cross-validation scores. A low value of s will provide a model with a lesser amount of predictors, while s=1 will provide a model with all the predictors (no feature selection, then).
Thanks very much. Your guide is very helpful. I will try it.
Best Regards,
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression