mlf: Conformalized Forecasting using Machine Leaning models

In Techtonique/ahead: Time Series Forecasting with uncertainty quantification

Conformalized Forecasting using Machine Leaning models

Description

Conformalized Forecasting using Machine Leaning models

Usage

mlf(
  y,
  h = 5,
  level = 95,
  lags = 15L,
  fit_func = ahead::ridge,
  predict_func = predict,
  coeffs = NULL,
  type_pi = c("kde", "surrogate", "bootstrap"),
  B = 250L,
  agg = c("mean", "median"),
  seed = 123,
  ...
)

Arguments

y	A numeric vector or time series of class ts
h	Forecasting horizon
level	Confidence level for prediction intervals
lags	Number of lags of the input time series considered in the regression
fit_func	Fitting function (Statistical/ML model). Default is Ridge regression.
predict_func	Prediction function (Statistical/ML model)
coeffs	Coefficients of the fitted model. If provided, a linear combination with the coefficients is used to compute the prediction.
type_pi	Type of prediction interval
B	Number of bootstrap replications or number of simulations
agg	"mean" or "median" (aggregation method)
...	additional parameters passed to the fitting function fit_func

Examples

res <- ahead::mlf(USAccDeaths, h=10L, lags=15L, type_pi="surrogate", B=250L)
plot(res)

res <- ahead::mlf(USAccDeaths, fit_func = glmnet::cv.glmnet, h=15L, lags=15L, 
type_pi="kde", B=250L) 
plot(res)

(res <- ahead::mlf(USAccDeaths, fit_func = e1071::svm, h=15L, lags=15L, 
type_pi="kde", B=250L)) 
plot(res)

Techtonique/ahead documentation built on April 14, 2025, 12:51 p.m.