In this project, we explore several techniques that can be used for hyperparameter optimization. More specifically, we explore the following techniques:
- Grid Search (with and without the usage of pipelines)
- Random Search (with and without the usage of pipelines)
- Bayesian Optimization with Gaussian Processes
- Hyperopt (https://github.com/hyperopt/hyperopt)
- Optuna (https://github.com/optuna/optuna)
We use a cleaned and balanced dataset for this project so that we can solely focus on the hyperparameter optimization. We use the Mobile Price Classification dataset, which contains information about mobile devices. (See the Kaggle link for details.) Furthermore, we use a random forest model to predict the price_range column. This target variable can take four different values, namely 0 (low cost), 1 (medium cost), 2 (high cost), and 3 (very high cost).