Closed JasonCEC closed 8 years ago
This may have been caused by an update to caret: https://github.com/topepo/caret/blob/c4a76780810875d9b957ee2cc0f17db6f0c1786c/pkg/caret/R/train.default.R#L164
The check in caretEnsable is here: https://github.com/zachmayer/caretEnsemble/blob/308fa6e672f6bf2c3654d92f30ec409f035ae889/R/helper_functions.R#L170
I'd be happy to issue the pull request if you could outline what the appropriate fix for this is?
Another check also occurs here. https://github.com/zachmayer/caretEnsemble/blob/308fa6e672f6bf2c3654d92f30ec409f035ae889/R/helper_functions.R#L53
In general, caretEnsemble
needs some work, but I haven't had time recently. I'll try to do an update soon to fix this and other issues.
I also think this change to caret was at my request— I'd asked to be able to save just the predictions for the best model, which really cuts down on the size of models like glmnet and gbm that can have hundreds of sub models.
Would you like me to issue a pull request just removing those lines, or checking that it's not "none" (the new alternative value instead of FALSE)?
If you want to submit a pull request checking for "not none", I would appreciate it!
Sent from my iPhone
On Dec 19, 2015, at 6:31 PM, Jason Cohen notifications@github.com wrote:
Would you like me to issue a pull request just removing those lines, or checking that it's not "none" (the new alternative value instead of FALSE)?
— Reply to this email directly or view it on GitHub.
Working on the pull request - I don't understand what needs to be changed in the location you linked (R/helper_functions.R#53);
all of the observations are still there, I believe?
Can I issue the pull request fixing only line 170 in helper _functions?
If I don't need to change anything on line 53, the above pull request should close this issue.
@JasonCEC You might want to submit a PR in zachmayer/caretEnsemble instead of in your forked copy?
@JasonCEC: Is there also another possiblity for Ensemble model training in R apart from caretEnsemble ?
Thank you.
@Mosquito00 You could build the ensemble yourself.... but caretEnsemble does much of the hard work for you, even if its a bit behind caret at the moment.
I believe this has been closed on the current branch with commit "fix for savePredictions".
The most recent development version of caretEnsemble has incorrect assumptions for the return value of
x$control$savePredictions
causing ensemble training to fail.A reproducible example is below, and includes a workaround: