Selection of a threshold parameter \(\lambda\) for GLM-lasso and Square-root lasso. The method consists in considering a null model, finding the theoretical distribution of the threshold parameter under the null, and setting \(\lambda\) to an upper quantile of that distribution. Although this strategy does not use the data to select \(\lambda\) but simply considers the behavior under the null model, it provides a theoretically and computationally sound selection.
| Package: | QUT |
| Type: | Package |
| License: | GPL-2 |
C. Giacobino, J. Diaz Rodriguez, S. Sardy, N. Hengartner. Quantile universal threshold for model selection. 2016