Apache® Spark™ News

How (Not) to Tune Your Model With Hyperopt

So, you want to build a model. You’ve solved the harder problems of accessing data, cleaning it and selecting features. Now, you just need to fit a model, and the good news is that there are many open source tools available: xgboost, scikit-learn, Keras, and so on. The bad news is also that there are so many of them, and that they each have so many knobs to turn. How much regularization do you need? What learning rate? And what is “gamma” anyway?