chunlinli / glmtlp

Constrained likelihood estimation and inference with truncated lasso penalty for linear, generalized linear, and Gaussian graphical models.

glmtlp: An R Package For Truncated Lasso Penalty

Efficient procedures for constrained likelihood estimation and inference with truncated lasso penalty (Shen et al., 2010; Zhang 2010) for linear, generalized linear, and Gaussian graphical models.

Currently this package is in active development and will be released very soon.

New features [Unreleased]

This version supports regression from summary statistics and out-of-core model fitting using an ultrahigh-dimensional, multi-gigabyte datasets that cannot be loaded into memory. It's highly optimized and much more memory-efficient as compared to existing penalized regression packages like glmnet and ncvreg.

• Add regression with summary data input

• Add inference function

• Add OpenMP support

• Add external memory computation support

• Add sparse coefficient matrix output

• Add implementation of Poisson regression

Highlights

Constrained likelihood, inference

An improved algorithm

Any GLM (to do)

GGM (to do)

Summary data

Big data, memory management

Visualization

Citing information

If you find this project useful, please consider citing

@article{
    author = {Chunlin Li, Yu Yang, Chong Wu, Xiaotong Shen, Wei Pan},
    title = {{glmtlp: An R package for truncated Lasso penalty}},
    year = {2022}
}

References

Li, C., Shen, X., & Pan, W. (2021). Inference for a large directed graphical model with interventions. arXiv preprint arXiv:2110.03805. https://arxiv.org/abs/2110.03805.

Shen, X., Pan, W., & Zhu, Y. (2012). Likelihood-based selection and sharp parameter estimation. Journal of the American Statistical Association, 107(497), 223-232. https://doi.org/10.1080/01621459.2011.645783.

Shen, X., Pan, W., Zhu, Y., & Zhou, H. (2013). On constrained and regularized high-dimensional regression. Annals of the Institute of Statistical Mathematics, 65(5), 807-832. https://doi.org/10.1007/s10463-012-0396-3.

Tibshirani, R., Bien, J., Friedman, J., Hastie, T., Simon, N., Taylor, J., & Tibshirani, R. J. (2012). Strong rules for discarding predictors in lasso‐type problems. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 74(2), 245-266. https://doi.org/10.1111/j.1467-9868.2011.01004.x.

Yang, Y. & Zou, H. A coordinate majorization descent algorithm for l1 penalized learning. Journal of Statistical Computation and Simulation 84.1 (2014): 84-95. https://doi.org/10.1080/00949655.2012.695374.

Zhu, Y., Shen, X., & Pan, W. (2020). On high-dimensional constrained maximum likelihood inference. Journal of the American Statistical Association, 115(529), 217-230. https://doi.org/10.1080/01621459.2018.1540986.

Zhu, Y. (2017). An augmented ADMM algorithm with application to the generalized lasso problem. Journal of Computational and Graphical Statistics, 26(1), 195-204. https://doi.org/10.1080/10618600.2015.1114491.

Part of the code is adapted from glmnet, ncvreg, and biglasso.

Warm thanks to the authors of above open-sourced softwares.

glmtlp's Issues

Not a single variable selected when kappa is large

I encountered an issue when using glmtlp() for a single run. I got output named integer(0), but neither a warning nor an error was produced. To replicate this issue, run the following codes:

library(glmtlp)

set.seed(1234)
n <- 100
p <- 300
rho <- 0.3
kappa0 <- 10
# snr = 17 ensures that sigma is about 1 as absolute noise level
gau_data <- gen_gaussian_data(n, p, rho, kappa0, beta_type = 2, snr = 17)

fit_raw <- glmtlp(
    gau_data$X, gau_data$y, family = "gaussian",
    method = "tlp-constrained", kappa = 40
)

selection <- which(abs(fit_raw$beta[, 2]) > 0) 
selection # none selected