Hello,

Trying to train a multinomial regression classifier using lasso penalty. Values of 0, 1, and 2 represent the genotypes for a sample of 242 individuals under positions 1 through 49 (labeled as pos1 through pos 49). The genotypes correspond to a specific ancestry.

Libraries: tidyverse, glmnet

Code:

Y <- train %>%

select(ancestry) %>%

na.omit() %>%

as.matrix()

head(Y)

X <-train %>%

select(pos1:pos49)%>%

as.matrix()

head(X)

#Specifying the set of tuning parameter (휆) values

lambdas <-10^seq(-3, 3, length.out= 500)

lasso.fit<-glmnet(X, Y, alpha = 1, lambda = lambdas)

Error:

Error in if (nulldev == 0) stop("y is constant; gaussian glmnet fails at standardization step") :

missing value where TRUE/FALSE needed

In addition: Warning message:

In storage.mode(y) <- "double" : NAs introduced by coercion

any suggestions?