Feature Request - Constraining Covariates

[michaelwhammer]

It would be useful if the functions in this package supported or provided arguments similar to glmnet::glmnet()'s lower.limits and upper.limits. I frequently run adaptive elastic nets using glmnet and regularly have to constrain covariates to be non-negative or non-positive.

[nanxstats]

@michaelwhammer -- that's a very good suggestion -- as I'm going to make more important vanilla glmnet/ncvreg arguments available for msaenet, this one could have a high priority.

[michaelwhammer]

Additionally, if it's possible to enable a ... argument to allow for any argument to passed on to glmnet/ncvreg. I don't know if this would cause confusion as many arguments in msaenet seem to be wrappers around arguments in glmnet/ncvreg.

[nanxstats]

-- that is a very good point. I guess there is a certain balance between the flexibility and clarity for the API design:

• ... was supported in the first versions, then it was removed on purpose to avoid possible confusions. If you look into the code, the internal parameter tuning functions support ..., while the current implementation does not expose this API to the user. This ensures clarity but sacrifices flexibility.
• another design consideration would be keeping the high-level interface of different wrappers for glmnet and ncvreg as similar/consistent as possible -- so only a minimal but critical set of arguments was selected in the present implementation. This also increases clarity while decreases flexibility.

After these, we can consider adding more flexibility -- lower.limits/upper.limits can be a good start.

[nanxstats]

@michaelwhammer -- this has been fixed by 5e10777. The new version (v2.6) has been submitted to CRAN. In the same time, you can try the GitHub version using

devtools::install_github("road2stat/msaenet")

Here is some code for testing:

library("msaenet")

dat = msaenet.sim.gaussian(
  n = 150, p = 500, rho = 0.6,
  coef = rep(2, 5), snr = 5, p.train = 0.7,
  seed = 1001)

fit0 = msaenet(
  dat$x.tr, dat$y.tr,
  alphas = seq(0.05, 0.95, 0.05),
  nsteps = 3L, seed = 1003)

print(fit0)
msaenet.nzv(fit0)
msaenet.fp(fit0, 1:5)
msaenet.tp(fit0, 1:5)
msaenet.pred = predict(fit0, dat$x.te)
msaenet.rmse(dat$y.te, msaenet.pred)
plot(fit0)

fit1 = msaenet(
  dat$x.tr, dat$y.tr,
  alphas = seq(0.05, 0.95, 0.05),
  nsteps = 3L, lower.limits = 0L,
  seed = 1003)

print(fit1)
msaenet.nzv(fit1)
msaenet.fp(fit1, 1:5)
msaenet.tp(fit1, 1:5)
msaenet.pred = predict(fit1, dat$x.te)
msaenet.rmse(dat$y.te, msaenet.pred)
plot(fit1)

Please feel free to let me know if it does not work as expected. Thanks!

-Nan