| Title: | Iterative Hard Thresholding Extensions to Cyclops |
|---|---|
| Description: | Fits large-scale regression models with a penalty that restricts the maximum number of non-zero regression coefficients to a prespecified value. While Chu et al (2020) <doi:10.1093/gigascience/giaa044> describe the basic algorithm, this package uses Cyclops for an efficient implementation. |
| Authors: | Marc A. Suchard [aut, cre], Patrick Ryan [aut], Observational Health Data Sciences and Informatics [cph] |
| Maintainer: | Marc A. Suchard <[email protected]> |
| License: | Apache License 2.0 |
| Version: | 1.0.2 |
| Built: | 2024-11-23 04:03:15 UTC |
| Source: | https://github.com/cran/IterativeHardThresholding |
createFastIhtPrior creates a fastIHT Cyclops prior object for use with fitCyclopsModel.
createFastIhtPrior( K, penalty = 0, exclude = c(), forceIntercept = FALSE, fitBestSubset = FALSE, initialRidgeVariance = 10000, tolerance = 1e-08, maxIterations = 10000, threshold = 1e-06 )createFastIhtPrior( K, penalty = 0, exclude = c(), forceIntercept = FALSE, fitBestSubset = FALSE, initialRidgeVariance = 10000, tolerance = 1e-08, maxIterations = 10000, threshold = 1e-06 )
K |
Maximum # of non-zero covariates |
penalty |
Specifies the IHT penalty |
exclude |
A vector of numbers or covariateId names to exclude from prior |
forceIntercept |
Logical: Force intercept coefficient into regularization |
fitBestSubset |
Logical: Fit final subset with no regularization |
initialRidgeVariance |
Numeric: variance used for algorithm initiation |
tolerance |
Numeric: maximum abs change in coefficient estimates from successive iterations to achieve convergence |
maxIterations |
Numeric: maximum iterations to achieve convergence |
threshold |
Numeric: absolute threshold at which to force coefficient to 0 |
An IHT Cyclops prior object of class inheriting from
"cyclopsPrior" for use with fitCyclopsModel.
nobs = 500; ncovs = 100 prior <- createFastIhtPrior(K = 3, penalty = log(ncovs), initialRidgeVariance = 1 / log(ncovs))nobs = 500; ncovs = 100 prior <- createFastIhtPrior(K = 3, penalty = log(ncovs), initialRidgeVariance = 1 / log(ncovs))
createIhtPrior creates an IHT Cyclops prior object for use with fitCyclopsModel.
createIhtPrior( K, penalty = "bic", exclude = c(), forceIntercept = FALSE, fitBestSubset = FALSE, initialRidgeVariance = 0.1, tolerance = 1e-08, maxIterations = 10000, threshold = 1e-06, delta = 0 )createIhtPrior( K, penalty = "bic", exclude = c(), forceIntercept = FALSE, fitBestSubset = FALSE, initialRidgeVariance = 0.1, tolerance = 1e-08, maxIterations = 10000, threshold = 1e-06, delta = 0 )
K |
Maximum # of non-zero covariates |
penalty |
Specifies the IHT penalty; possible values are 'BIC' or 'AIC' or a numeric value |
exclude |
A vector of numbers or covariateId names to exclude from prior |
forceIntercept |
Logical: Force intercept coefficient into regularization |
fitBestSubset |
Logical: Fit final subset with no regularization |
initialRidgeVariance |
Numeric: variance used for algorithm initiation |
tolerance |
Numeric: maximum abs change in coefficient estimates from successive iterations to achieve convergence |
maxIterations |
Numeric: maximum iterations to achieve convergence |
threshold |
Numeric: absolute threshold at which to force coefficient to 0 |
delta |
Numeric: change from 2 in ridge norm dimension |
An IHT Cyclops prior object of class inheriting from
"cyclopsPrior" for use with fitCyclopsModel.
prior <- createIhtPrior(K = 10)prior <- createIhtPrior(K = 10)