glasp_regression: parsnip interface for GLASP - Regression Models

In jlaria/glasp: Group Linear Algorithm with Sparse Principal decomposition

parsnip interface for GLASP - Regression Models

Description

parsnip interface for GLASP - Regression Models

Usage

glasp_regression(
  mode = "regression",
  l1 = NULL,
  l2 = NULL,
  frob = NULL,
  num_comp = NULL
)

Arguments

mode	A single character string for the type of model.
l1	Regularization in norm l1 (lasso)
l2	Regularization in norm l2 (ridge)
frob	Regularization in Frobenius norm. It is unique to glasp and controls the importance of the clustering on the variables. If frob=0, we want to fit a model equivalent to elastic-net, without clustering. If frob > l1+l2 means that we are more interested on the feature clustering than the variable selection. If frob > 1 we are more interested on finding clusters than on getting a linear model with good predictive power.
num_comp	Maximum number of clusters to search.

Value

A glasp_regression parsnip model

Examples

library(parsnip)
library(yardstick)

set.seed(0)
data <- simulate_dummy_linear_data()

model <- glasp_regression(l1 = 0.05, l2 = 0.01, frob = 0.001, num_comp = 3) %>%
  set_engine("glasp") %>%
  fit(y~., data)

pred = predict(model$fit, data, type="numeric")
rmse = rmse_vec(data$y, pred)

#--- with tune ----
## Not run: 
library(parsnip)
library(tune)
library(yardstick)
library(rsample)

set.seed(0)
data <- simulate_dummy_linear_data()

model <- glasp_regression(l1 = tune(), l2 = tune(), frob = tune(), num_comp = tune()) %>%
set_engine("glasp")

data_rs <- vfold_cv(data, v = 4)
hist <- tune_grid(model, y~.,
        resamples = data_rs,
        metrics = metric_set(rmse),
        grid = 100,
        control = control_grid(verbose = T, save_pred = T))
show_best(hist, 'rmse')

## End(Not run)

jlaria/glasp documentation built on Dec. 5, 2022, 6:42 a.m.