cuda.ml (version 0.3.2)

cuda_ml_logistic_reg: Train a logistic regression model.

Description

Train a logistic regression model using Quasi-Newton (QN) algorithms (i.e., Orthant-Wise Limited Memory Quasi-Newton (OWL-QN) if there is L1 regularization, Limited Memory BFGS (L-BFGS) otherwise).

Usage

cuda_ml_logistic_reg(x, ...)
# S3 method for default
cuda_ml_logistic_reg(x, ...)
# S3 method for data.frame
cuda_ml_logistic_reg(
  x,
  y,
  fit_intercept = TRUE,
  penalty = c("l2", "l1", "elasticnet", "none"),
  tol = 1e-04,
  C = 1,
  class_weight = NULL,
  sample_weight = NULL,
  max_iters = 1000L,
  linesearch_max_iters = 50L,
  l1_ratio = NULL,
  ...
)
# S3 method for matrix
cuda_ml_logistic_reg(
  x,
  y,
  fit_intercept = TRUE,
  penalty = c("l2", "l1", "elasticnet", "none"),
  tol = 1e-04,
  C = 1,
  class_weight = NULL,
  sample_weight = NULL,
  max_iters = 1000L,
  linesearch_max_iters = 50L,
  l1_ratio = NULL,
  ...
)
# S3 method for formula
cuda_ml_logistic_reg(
  formula,
  data,
  fit_intercept = TRUE,
  penalty = c("l2", "l1", "elasticnet", "none"),
  tol = 1e-04,
  C = 1,
  class_weight = NULL,
  sample_weight = NULL,
  max_iters = 1000L,
  linesearch_max_iters = 50L,
  l1_ratio = NULL,
  ...
)
# S3 method for recipe
cuda_ml_logistic_reg(
  x,
  data,
  fit_intercept = TRUE,
  penalty = c("l2", "l1", "elasticnet", "none"),
  tol = 1e-04,
  C = 1,
  class_weight = NULL,
  sample_weight = NULL,
  max_iters = 1000L,
  linesearch_max_iters = 50L,
  l1_ratio = NULL,
  ...
)

Arguments

x

Depending on the context:

* A __data frame__ of predictors. * A __matrix__ of predictors. * A __recipe__ specifying a set of preprocessing steps * created from [recipes::recipe()]. * A __formula__ specifying the predictors and the outcome.

...

Optional arguments; currently unused.

y

A numeric vector (for regression) or factor (for classification) of desired responses.

fit_intercept

If TRUE, then the model tries to correct for the global mean of the response variable. If FALSE, then the model expects data to be centered. Default: TRUE.

penalty

The penalty type, must be one of "none", "l1", "l2", "elasticnet". If "none" or "l2" is selected, then L-BFGS solver will be used. If "l1" is selected, solver OWL-QN will be used. If "elasticnet" is selected, OWL-QN will be used if l1_ratio > 0, otherwise L-BFGS will be used. Default: "l2".

tol

Tolerance for stopping criteria. Default: 1e-4.

C

Inverse of regularization strength; must be a positive float. Default: 1.0.

class_weight

If NULL, then each class has equal weight of 1. If class_weight is set to "balanced", then weights will be inversely proportional to class frequencies in the input data. If otherwise, then class_weight must be a named numeric vector of weight values, with names being class labels. If class_weight is not NULL, then each entry in sample_weight will be adjusted by multiplying its original value with the class weight of the corresponding sample's class. Default: NULL.

sample_weight

Array of weights assigned to individual samples. If NULL, then each sample has an equal weight of 1. Default: NULL.

max_iters

Maximum number of solver iterations. Default: 1000L.

linesearch_max_iters

Max number of linesearch iterations per outer iteration used in the LBFGS- and OWL- QN solvers. Default: 50L.

l1_ratio

The Elastic-Net mixing parameter, must NULL or be within the range of [0, 1]. Default: NULL.

formula

A formula specifying the outcome terms on the left-hand side, and the predictor terms on the right-hand side.

data

When a __recipe__ or __formula__ is used, data is specified as a __data frame__ containing the predictors and (if applicable) the outcome.

Examples

# NOT RUN {
library(cuda.ml)

X <- scale(as.matrix(iris[names(iris) != "Species"]))
y <- iris$Species

model <- cuda_ml_logistic_reg(X, y, max_iters = 100)
predictions <- predict(model, X)

# NOTE: if we were only performing binary classifications (e.g., by having
# `iris_data <- iris %>% mutate(Species = (Species == "setosa"))`), then the
# above would be conceptually equivalent to the following:
#
# iris_data <- iris %>% mutate(Species = (Species == "setosa"))
# model <- glm(
#   Species ~ ., data = iris_data, family = binomial(link = "logit"),
#   control = glm.control(epsilon = 1e-8, maxit = 100)
# )
#
# predict(model, iris_data, type = "response")
# }