Tweets about hyperparameter optimization

Find best hyperparameter for your model

<aside> 💡

Finding the right parameters only works through experiments. How to conduct and evaluate these experiments can be seen in the XGBoost book, pages 92-96. But keep in mind: First tune the architecture, don’t tune the parameters! If you want to train multiple architectures check this.

</aside>

When looking for the best hyperparameters you can spend a lot of compute. So much so, that you can also spend too much. It is a sutble thing, but if you're not careful you can become a victim to something that's known as "the optimisers curse". One advice would be to plot the effect of tuning one parameter agains your desired metric. This way you get a feeling if this parameter only adds noise or if it actually helps. → very general explanation incl. code in video.

Libraries for hyperparameter optimization

<aside> 💡

Check out: AutoML & Experiment tracking

</aside>

Santiago

You should systematically and automatically search for the optimal hyperparameters for your model.

Scikit-learn for Hyperparameter Tuning

Optuna

Hyperopt

Scikit-optimize

Ray-tune

For neural networks:

scikeras