Introduction

This package is the implementation example of tree-structured parzen estimator (TPE). TPE is an hyperparameter optimization (HPO) method invented in Algorithms for Hyper-Parameter Optimization.

NOTE: The sampling strategy is based on the BOHB implementation.

Setup

This package requires python 3.7 or later version and you can install

pip install -r requirements.txt

If you would like to play around using a tabular benchmark, you can download the tabular data for HPO on 4 different datasets:

$ wget http://ml4aad.org/wp-content/uploads/2019/01/fcnet_tabular_benchmarks.tar.gz
$ tar xf fcnet_tabular_benchmarks.tar.gz

# Run the optimization using TPE
$ optimize_hpolib.py

Note that you need to move the downloaded dataset accordingly or specify the path in optimize_hpolib.py.

Running example

# Optimize 10D sphere function
$ python optimize_sphere.py

# Optimize the hyperparameters defined in `cnn/hyperparameters.py` and `cnn/params.json`
$ python optimize_cnn.py

# Optimize the hyperparameters defined in `hpolib/hyperparameters.py` and `hpolib/params.json`
$ python optimize_hpolib.py