lambdaSelection: Regularization parameter selection based on network...

Description Usage Arguments Details Value Note Author(s) References See Also Examples

View source: R/lambdaSelection.r

Description

lambdaSelection is a function designed to select the sparsity regularization parameter in graphical models.

There are available seven different criterion to select lambda with risk functions based on network characteristics: Path connectivity (PC), AGlommerative NESted (AGNES), Augmented-MSE (A-MSE), Vulnerability (VUL), AIC/BIC and StARS (from huge package).

Usage

1
2
lambdaSelection(obj, criterion = c("PC","AGNES",
  "A-MSE","VUL","STARS", "AIC", "BIC", "eBIC"), ...)

Arguments

obj

an object of class huge, camel.tiger or wfgl.

criterion

regularization parameter selection approach: name that uniquely identifies "PC" (Path connectivity), "AGNES" (AGlommerative NESted Algorithm), "A-MSE" (Augmented mean square error), "VUL" (Maximum VULnerability), "AIC"/"BIC"/"eBIC" (minimum AIC/BIC/eBIC) or "StARS" (Stability approach).

...

arguments passed to or from other methods to the low level. See pcLambdaSelection, agnesLambdaSelection, amseLambdaSelection, vulLambdaSelection,
aicAndbicLambdaSelection and huge for details.

Details

This function considers seven ways of selecting the regularization parameter in graphical models by minimizing a certain risk function based only on network characteristics of the underlying structure of Ω

\hatλ = \arg \min_{λ} R(λ, \hat{G}_λ),

where \hat{G}_λ is the estimated graph structure of \hat{Ω}. For instance see pcLambdaSelection,
agnesLambdaSelection, amseLambdaSelection, vulLambdaSelection,
aicAndbicLambdaSelection and huge for the implemented criterions to select λ.

For wfgl objects, only criterion = c("PC","AGNES","VUL") are available.

Value

An object of class lambdaSelection containing the following components:

opt.lambda

optimal lambda.

crit.coef

coefficients for each lambda given the criterion.

criterion

criterion used to select lambda.

Note

For large dimensions, "A-MSE", "VUL" (the most) and "StARS" can be computationally intensive.

Author(s)

Caballe, Adria <[email protected]>, Natalia Bochkina and Claus Mayer.

References

Caballe, A., N. Bochkina, and C. Mayer (2016). Selection of the Regularization Parameter in Graphical Models using network charactaristics. eprint arXiv:1509.05326, 1-25.

See Also

pcLambdaSelection, agnesLambdaSelection, amseLambdaSelection
vulLambdaSelection, aicAndbicLambdaSelection and huge.

Examples

1
2
3
4
5
6
7
8
9
# example to use agnes function
EX1         <- pcorSimulator(nobs = 50, nclusters = 3, nnodesxcluster = c(40,30,30), 
                             pattern="powerLaw")
y           <- EX1$y
Lambda.SEQ  <- seq(.35,0.70, length.out = 40)
out3        <- huge(y, method = "mb", lambda = Lambda.SEQ)
PC.COEF     <- lambdaSelection(out3, criterion = "PC")
#AG.COEF     <- lambdaSelection(out3, criterion = "AGNES")
 

ldstatsHD documentation built on Aug. 14, 2017, 5:06 p.m.