DSN AI Bootcamp Qualification Hackathon by Data Science Nigeria
Predict customers who will default on a loan
876 data scientists enrolled, 506 on the leaderboard
Financial ServicesPredictionStructured
9 September—3 October
Ends in 9 days
SVM,KNN,RandomForest, Grid Search Takes too much time on Colab, any help
published 14 Sep 2020, 14:48

i have been trying to tune the parameters for svc, knn and random forest on google colab, it like am just wasting my time

Change the knn to Decision Tree. KNN takes longer period to train on such dataset.

Did you on your GPU on colab? And forget about knn bcos it will take a while day to run.

yes GPU enabled on colab, can you suggest some models for me

Use Lightgbm..... It's more faster.

yeah, i have tried that, it worked well, i want to improve the score, any tip

Try ensemble techniques, crossvalidation techniques, stacking etc.

focus on models from different families of algorithm to build the ensembles. add a little bit randomness to the models and use a mode or mean probability score

Thanks, i will try that out

You don't need stacking, ensembling yet ,just focus on features engineering, feature selecting.. after all those you will see great change when you use stacking and the likes

use random search to get good parameters. then, fine tune the values using a smaller set of values . Alternatively, you can use bayesian search, its very fast and accurate!