new example src/logisticRegression.cpp

[MHenderson]

update header, also include generated RcppExports.R

[MHenderson]

This includes two new examples logisticRegressionand logisticRegressionParameters. The first function does binary classification. The second one just does the training and returns the parameters of the trained model.

[eddelbuettel]

Hm. We could make the test set 'Nullable' (I can help with an example, we have that in a few places now) and could then test for it, and just have one function. Better?

[MHenderson]

Yes, one function instead of two does sound better. Can you point me towards an example of the 'Nullable' test set approach?

I suppose that the return value ought then to be a list? So we can return both the parameter vector and the results.

[eddelbuettel]

Regarding Nullable<>: It is used in mvabund three times and also once here in Rblpapi. See how you need .isNotNull() to test before instantiating. Should work cleanly here too.

As for the return, I am confused. Aren't both just vectors? So we return a vector either way? Or too little coffee here?

[MHenderson]

Thanks for the examples!

As for the return, I was thinking that the user might like to have both the test set classification and the model parameters without having to call the function twice.

[eddelbuettel]

Yes, then we'd use a list. Minor downside is that you then need to 'unpack'. But eventually we'll need that anyway, possibly also for existing classes.

Nullable<> is still somewhat new but it can help here. Strikes a fine balance between the run-time payload of NULL in a SEXP and the fact that we need types in C++.

[MHenderson]

So I think I need to convert from Rcpp::Nullable<arma::mat> to arma::mat because Classify expects the latter.

if (test.isNotNull()) {
      arma::mat test2(test);
      lrc.Classify(test2, resultsur);
      arma::vec results = arma::conv_to<arma::irowvec>::from(resultsur);
    } else {
      arma::vec results = R_NilValue;
    }

But this doesn't work because I get the following error:

logisticRegression.cpp:26:27: error: no matching function for call to ‘arma::Mat<double>::Mat(const Rcpp::Nullable<arma::Mat<double> >&)’

[eddelbuettel]

Ah, yes -- Nullable<> can (I think) only wrap around SEXP types so that needs to be a Rcpp::Nullable<Rcpp::NumericMatrix> , and in the non-null case you can instantiate an Arma matrix from the NumericMatrix via as<>():

Rcpp::cppFunction("void foo(NumericMatrix m) { arma::mat am = Rcpp::as<arma::mat>(m); 
                                               am.print(); }",
                               depends="RcppArmadillo")

Doable?

[MHenderson]

Yes, that works. My code probably leaves a lot to be desired but it does (seem to) work, at least.

[eddelbuettel]

That looks good -- thanks for making those changes!