>>13973820>be me, work in cheminformatics>do a lot of QSAR modeling>see some sperg on sci yelling about symbolic regression>Never used it, decide to try it out>gplearn>play with some HIV QSAR datasets; got a few big ones, ROC of ~0.92 on test data from our standard xgboost models>try out the symbolic classfier>utter shit, ROC of 0.53, terrible precision and recall>Okay, maybe regressors are better, try it out against our LSTM-Convolutional model>more shit results>wrap it in a bayes opt hyperparameter search, 70/15/15 split, holdout same test set>shit r2 and pearsonsinto the trash it goes
at least I added another shitty algorithm to my arsenal, I guess
back in the trash it goes
