You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Computing is bottleneck! Full hyperparam search with good nested-cross-validation will eat tons of compute. In world of chemistry and biology 2% can be translated into +2year of life of a patient, and hundreds of $.
You don't need extensive hyperparameter search. If you do it anyway, you'll simply overfit to your validation procedure. A better way to gain the +2% is to use a diverse ensemble of models (never seen in papers, because that's besides the point of the publication game).
For chemistry we gained a lot with different DL architectures. But I agree playing 'academic' game with extensive optimization is not beneficial. PS: building good multi-modal DL model on heterogenous incomplete data is hard (at least yet in our hands).
Re ensembles (but not diverse ones) there is some very recent work: Fast Geometric Ensembling (FGE) https://arxiv.org/abs/1802.10026 by @andrewgwils & team and Snapshot Ensembles @iclr2017https://arxiv.org/abs/1704.00109
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
