Elastic nets in R, with some improvements.
Originally developed by Jerome Friedman, Trevor Hastie, Noah Simon, Rob Tibshirani & others.
CRAN glmnet changed cross validation so that the lambda vector is recomputed for each fold. I am not convinced that this is a good idea, and can lead to errors on small samples.
CRAN glmnet can't score on a single row:
library("glmnet")
x <- matrix(rnorm(100 * 20), 100, 20)
y <- rnorm(100)
fit1 <- glmnet(x, y)
predict(fit1, newx = x[1, ], s = 0.005)
# Actually, this version can't do this either
predict(fit1, newx = as.matrix(x[1, ]), s = 0.005)
# You have to do this, which is weird
predict(fit1, newx = t(as.matrix(x[1, ])), s = 0.005)Barebones, experimental and completely undocumented formula & data.frame interface à la glm, with optional enforcement of one-column-per-factor-level without having to hack contrasts.arg. Also subsettable coefficient profiles.
# install.packages("devtools")
devtools::install_github("DexGroves/glmnet")
library("glmnet")
generate_data <- function(N) {
# Make some fake, noisy data with a few factors
df <- data.frame(factor_a = as.factor(sample(letters[1:4], N, TRUE)),
factor_b = as.factor(sample(LETTERS[1:4], N, TRUE)),
num_a = runif(N),
num_b = runif(N),
noise_a = rnorm(N),
noise_b = rnorm(N),
noise_factor = as.factor(sample(letters[20:26], N, TRUE)))
df$response <- as.numeric(df$factor_a) +
2 * as.numeric(df$factor_b) +
0.25 * df$num_a +
0.5 * df$num_b +
rnorm(N) * 7
data.frame(df)
}
# Some data.frames
train <- generate_data(2000)
hold <- generate_data(2000)
m_formula <- response ~ factor_a + factor_b + num_a + num_b +
noise_a + noise_b + noise_factor
# glmnet ex data.frame, where design matrix has one column per factor level
g <- df.glmnet(m_formula, train, cv = TRUE, nfolds = 10, auto_contrast = TRUE)
# predict on a data.frame.
pred <- predict(g, hold, s = "lambda.min")
# Coefficient profile for specific coefficients only (labels are WIP)
coef.plot(g, coefs = c("factor_aa","factor_ab", "factor_ac", "factor_ad"))