[BioC] lower.limits and upper.limits in glmnet

Yue Li yueli at cs.toronto.edu
Mon Oct 21 20:31:24 CEST 2013


Hi Steve,

Thanks for the reply. Yes, I tried it and it does seem to do that. I was just uncertain about how is achieved. Thanks anyways.

Input:
mycoef1 <- as.matrix(coef(glmnet(x,y,alpha=1,intercept=F),s=0.01))

mycoef2 <- as.matrix(coef(glmnet(x,y,alpha=1,intercept=F,upper.limits=c(rep(Inf,5), rep(0,15))),s=0.01))

print(cbind(mycoef1, mycoef2))

Output:
                       1            1
(Intercept)  0.000000000  0.000000000
V1           0.073877294  0.066218767
V2           0.034539506  0.004229772
V3           0.056682635  0.061373751
V4          -0.048408058 -0.067313603
V5           0.000000000  0.001866883
V6          -0.027033061 -0.024216745
V7          -0.053746653 -0.048835781
V8           0.151853999  0.000000000
V9           0.006023245  0.000000000
V10          0.007404765  0.000000000
V11         -0.087274939 -0.129084080
V12          0.019070975  0.000000000
V13         -0.031480104 -0.024312149
V14          0.075601445  0.000000000
V15         -0.021138293 -0.015392988
V16          0.013038938  0.000000000
V17         -0.046345120 -0.043179775
V18         -0.166853254 -0.158282024
V19         -0.051904691 -0.025384641
V20         -0.008883316  0.000000000



On 2013-10-21, at 2:13 PM, Steve Lianoglou <lianoglou.steve at gene.com> wrote:

> Hi,
> 
> On Mon, Oct 21, 2013 at 10:51 AM, Yue Li <yueli at cs.toronto.edu> wrote:
>> Dear List,
>> 
>> A quick question about using the glmnet package (which I know is not a BioC package ... apologies):
>> 
>> Would the options "upper.limits" and "lower.limits" in glmnet be equivalent to an additionally constrained optimization on the range of the coefficients?
>> 
>> For instance, would setting upper.limits to 0 be equivalent to a non-positive least squared linear regression (regularized)?
> 
> Reading the documentation for those parameters in the `?glmnet` docs
> certainly suggests so, no?
> 
> This is simple enough for you to try yourself, though, so why not just
> give it a shot and report back with your results?
> 
> -steve
> 
> -- 
> Steve Lianoglou
> Computational Biologist
> Bioinformatics and Computational Biology
> Genentech



More information about the Bioconductor mailing list