Closed kerschke closed 8 years ago
Currently I am working on implementing mgcv::gam in mlr. And then I'd carry on with implementation of regularized random forest, relaxed lasso and bagged CART as well. So, maybe some other contributor may start with other learners.
Thx. Please post here, if you look at something so effort is not duplicated
Working on sparseLDA::sparseLDA.
I think I will focus on adding neural network related learners:
I intended to add xgboost first, but soon figured out that it has been added already.
@hetong007 What is the status here? Have you added the learners? Do I need to look at the PRs?
@berndbischl I have hold it down because it has some confliction, and I want to put SwarmSVM
in the PR first. I am waiting for SwarmSVM
to appear on cran and open the PR. I could also add these codes together in the PR.
why not do it now for the learners that are already on cran? that is much easier
Then can I make a new fork and put the code for neuralnet
first?
sure
deepnet would also be nice
Actually I found out the deepnet
is not so usable. The model quality is really low. I tested the deepnet::sae.dnn.train
and deepnet::nn.train
and had a hard time to tune the parameters but still got really bad on the binaryclass
data. Even on the demo data set provided in its examples, the model predict every data point with almost the same result (probability).
Tong, please dont judge the learners at this level. To properly benchmark them, we need them in mlr. Please add the deepnet learner nonetheless
Now the situation is they can't pass my local check in test_learners_classiflabelswitch.R
: https://github.com/mlr-org/mlr/blob/master/tests/testthat/test_learners_classiflabelswitch.R#L63
I have them grammar correctly implemented in my fork of mlr
, but the performance seems just not strong enough to pass this test.
Then adjust the error rate threshold in the test.
Thanks, now it passes the check.
working on rknn::rknn
we should move this to the wiki and update it a bit
@florianfendt Please do this. Just have a smaller list of models in the wiki and give a few helpful pointers to people how to integrate them
then close here
ok, will do that after we talked about it on Monday!
we have that list in the wiki now. closing
I just had a look at caret and made a list of learners, which have not yet been integrated into mlr (at least as far as I know).
This list is very likely not complete and might also include some functions, which might not be useful for our purposes. But at least, we now have an overview of possible extensions :-)