Explore hyperparameter tuning

A hyperparameter is a model parameter that is chosen by you before training begins. (This is in contrast to parameters, which are determined by the model training.) The hyperparameters available to set differ between types of model.

Here you see the results of a gradient boosting models (GBMs) that tries to predict whether or not people will vote in an election. GBMs are a type of ensemble model that create lots of regression trees. Hyperparameters for GBMs include the number of trees to generate, the complexity of each tree, and the learning rate (how much weight is given to each tree).

It's usually impossible to know which combination of hyperparameters will result in the best performing model, so you have to try lots of combinations of them.

Use the dashboard controls to change hyperparameters and find the combination that gives the highest accuracy.