add_effects: Add effects to a Xy simulation

In andrebleier/Xy: Simulating Supervised Learning Data

Add effects to a Xy simulation

Description

With the helping function below you can alter a simulation by simply adding the desired effects to the simulation object from Xy().

Usage

add_linear(object, p, family = xy_normal())

add_nonlinear(object, p, nlfun = function(x) x^2, family = xy_normal())

add_discrete(object, p, levels = 2)

add_uninformative(object, p, collinearity = FALSE, family = xy_normal())

add_intercept(object)

add_noise(object, collinearity = FALSE, family = xy_normal())

add_interactions(object)

Arguments

object	an object of class xy_recipe.
p	an integer specifying the number of effects to simulate
family	a distributional family (see families)
nlfun	a function which transforms the simulated variable
levels	an integer specifying the number of levels within the simulated discrete variable
collinearity	a boolean specifying whether there is collinearity between the features and uninformative variables

Details

A note on non-linear effects: Depending on the non-linear function it is highly recommended to adjust the parameters of the family, e.g. for a nonlinear quadratic function it is recommended to sample from a uniform distribution with minimum value of 0 and maximum value of 1000 such that the non-linearity is clearly visible in the data. ⁠# nonlinear simulation sim_nl <- Xy(task = "regression") %>% add_nonlinear(p = 5, nlfun = function(x) x^2, family = xy_uniform(min = 0, max = 1000))⁠

Value

an object of class xy_recipe

Examples

xy_recipe <- Xy(task = "regression") %>%
  # add linear features
  add_linear(p = 5)

# add cubic nonlinear features
xy_recipe <- xy_recipe %>%
  add_nonlinear(p = 3, nlfun = function(x) x^3, family = xy_uniform(min = 0, max = 1000))

# add discrete features with four unique factor levels
xy_recipe <- xy_recipe %>%
  add_discrete(p = 2, levels = 4)

# add uninformative features (they do not influence the target generating process)
xy_recipe <- xy_recipe %>%
  add_uninformative(p = 5)

# add features from the cauchy distribution
xy_recipe <- xy_recipe %>%
  add_linear(p = 2, family = xy_cauchy(location = 3, scale = 5))

# add random interactions between all informative features
xy_recipe <- xy_recipe %>%
  add_interactions()

# add a specific form of noise to your process, e.g. poisson distributed
# noise
xy_recipe <- xy_recipe %>%
  add_noise(family = xy_poisson(lambda = 3))

andrebleier/Xy documentation built on March 29, 2023, 12:41 p.m.