brentonk/polywog

R package for flexible functional form estimation via oracle-penalized basis regression

polywog's Issues

Rewrite documentation for predVals()

Since the functionality has changed, the documentation should reflect that (and cite Hanmer and Kalkan)

.parallel in cv.polywog

.parallel appears to be ignored in cv.polywog.r. perhaps related to this is the difficulty in using keyboard interrupt to stop execution of polywog (basically I have to kill R every time).

Implement logit transformation in C++ backend to predict()

Error when running cv.polywog

Hi,

I've tried running cv.polywog but I can't figure out what is causing this error.
My data are just a dataframe

I run:
cv.polywog(y_dat~x_dat, data=forpoly, degrees.cv = 1:10, nfolds = 5)

and the error I get is:

Error in { :
task 1 failed - "x should be a matrix with 2 or more columns"

I tried inputing my data as a matrix instead, but then got an error that my data should be a dataframe. Any assistance resolving this would be great! Thanks!

Julie

Implement NA handling in C++ backend to predict()

Implement margEff in Rcpp

Eliminate preplot.polywog

A "preplot.polywog" object is just a reclassed predVals() return, so eliminate it and have the plotting methods deploy right onto predVals() output.

Update namespace directives for external packages

To fix the build warning

These packages needs to imported from for the case
when this namespace is loaded but not attached.

Reimplement "unpenalized"

Since we're no longer explicitly constructing the model matrix, it isn't possible to exclude terms from the penalty according to their names in the model matrix. That leaves us with a few natural ways to implement unpenalized:

1. Name raw terms to leave unpenalized, then set the penalty weight of all polynomial terms including any of these to 0. The upside is that this is easy to implement and reasonably robust to changes in the data used for fitting. The downside is that it comes with considerable loss of generality except when used for linearly included terms.
2. Provide indices of the rows of polyTerms to leave unpenalized. Then we are fully general, but some problems arise:
   • The user has to inspect polyTerms to know which rows are which, meaning running polywog(..., method = "none") once before fitting
   • The identity of a row can change if terms are added or dropped from the model formula, or if a change in the dataset changes the set of linearly independent polynomial terms. This could be partly alleviated by having the print and summary methods tell which terms were unpenalized, but it still seems risky.
3. Perhaps the best way would be to reimplement the old way, by generating the "variable name" corresponding to each row of polyTerms after it is computed, then checking those against the supplied character vector unpenalized

Remove "bag" arguments from plotting functions

Can also remove ... argument from predVals() once this is done

Add model.frame and model.matrix methods