Kernel-weighted multinomial logit on (x - x0), last class as baseline; the
class probabilities at x0 are the softmax of the local intercepts (ESL eqs
6.18-6.20).
Usage
morie_esl_local_logistic(X, g, x0, lambda, kernel = "gaussian")
Arguments
- X
Predictors, n by p.
- g
Class labels.
- x0
Targets, one per row.
- lambda
Bandwidth.
- kernel
"gaussian", "epanechnikov" or "tri-cube".
Value
Named list: prob (targets by classes), intercepts, classes.
References
Hastie, Tibshirani & Friedman (2009), sec. 6.5.
Examples
i <- 1:40
morie_esl_local_logistic(cbind(sin(i), cos(3 * i)), (i * 7) %% 3, rbind(c(0.2, 0.3)), 1)$prob
#> [,1] [,2] [,3]
#> [1,] 0.3058034 0.3823314 0.3118652
