cxhernandez
commented
5 years ago Thanks @RobertArbon! This looks great. Sorry for reviewing these changes so late— has been hard to find the time to get the master branch back into a state where PRs can be merged.
If you don't mind, I'll remove the pyemma and SobolSearch additions (feel free to add them back in a future PR) so as to better organize the focus of this PR a bit.
I've refactored the code so that there are now four strategies:
The GP strategy is now a type of surrogate function. There is also now a general Acquisition function class.
I've not added any more tests at the moment as I'd like some guidance on that if possible.
I'm starting a PL but it doesn't seem to pass my own Travis CI - it fails on three tests which require the fs_peptide data set. See here. I keep getting
Import Error msmbuilder.io.sampling No module named 'sklearn.grid_search'The three tests that are failing are:
test_cli_worker_and_dump.test_msmbuilder_skeleton()test_cli_worker_and_dump.test_msmb_feat_select_skeleton()test_dataset_loader.test_MDTrajDatasetLoader_1()test_fit_estimator.test_1()All the tests are passed when running the same test suite on my local computer however. The last one I think is because you need to download the data files properly.
UPDATE:
The problem for 1 - 2 is an import in msmbuilder. I've made a pull request there to fix the problem. In the mean time I've fixed that by making wrapping the imports in a try/except block. This means that these tests don't get loaded at the moment.
The fix for 3 is to actually download the data in the test in the same way as the other fs_peptide tests are downloaded. The import will be in a try/except block.
The fix for 4 was to change to the sklean 0.20 api for the GridSearchCV object.
There is a mystery as to why
test_dataset_loader.test_MSMBuilderDatasetLoader_1()passes the test though - apparently there is no problem in importingmsmbuilder.dataset