1.4.7 A binomial fit🔗ℹ

With #:family 'binomial, the response column holds 0/1 labels, or two classes as strings, symbols or booleans (see Categorical responses). Here the formula takes every column except the other responses:

> (define relapse-cv     (formula-cv (~ relapse (- all bp time status)) patients                 #:family 'binomial))

> relapse-cv

#<glmnet-cv:binomial (~ relapse (- all bp time status)) Measure: Binomial Deviance      Lambda Index Measure      SE Nonzero min 0.00980    38  0.3976 0.11556       2 1se 0.03604    24  0.5025 0.08812       2>

> (coef relapse-cv)

'(("(Intercept)" . 0.8782313960189869)   ("age" . 0.0)   ("dose" . -0.5364786058863007)   ("marker" . 4.33160821549671))

> (predict relapse-cv new-patients #:type 'response)

'(0.8777268852046514 0.003760628144205349)

> (predict relapse-cv new-patients #:type 'class)

'(1 0)

At lambda-1se, the model keeps the dose and the marker, the two predictors that the relapse depends on.