This fixes prediction when a learner fails for methods: NNLS, NNloglik, CC_nloglik, and AUC, per #94. NNLS2 and CC_LS still have this bug as I couldn't identify how to resolve the issue and I also don't know if anyone uses those methods.
This fix required that an additional optional argument "errorsInLibrary" be passed to methods. This argument is a vector set to TRUE for learners that failed during model fitting. I think it's reasonable in general that the metalearner should be given direct information about which algorithms failed so that it can adjust accordingly, since Z will only contain 0s for those predictions.
I updated the package version to -9000 so that we can more easily tell whether an in-development github version or a CRAN version is installed, as recommended at http://r-pkgs.had.co.nz/description.html#version
While I was at it I also converted the manual page for predict.SuperLearner to roxygen using https://yihui.name/Rd2roxygen
jlstiles
commented
7 years ago
This fixes prediction when a learner fails for methods: NNLS, NNloglik, CC_nloglik, and AUC, per #94. NNLS2 and CC_LS still have this bug as I couldn't identify how to resolve the issue and I also don't know if anyone uses those methods.
This fix required that an additional optional argument "errorsInLibrary" be passed to methods. This argument is a vector set to TRUE for learners that failed during model fitting. I think it's reasonable in general that the metalearner should be given direct information about which algorithms failed so that it can adjust accordingly, since Z will only contain 0s for those predictions.
I updated the package version to -9000 so that we can more easily tell whether an in-development github version or a CRAN version is installed, as recommended at http://r-pkgs.had.co.nz/description.html#version
While I was at it I also converted the manual page for predict.SuperLearner to roxygen using https://yihui.name/Rd2roxygen