What is hpopt?

hpopt is a Python library for automatic hyper-parameter optimization, which finds optimized hyper-parameters work best for given tasks. hpopt aims fast hyper-parameter optimization with some convenient features like auto-config.
Hyper-parameter optimization is time-consuming process even with state-of-the-art off-the-shelf libraries. hpopt adopts proxy task, early stopping, and multi-band optimization in a single framework, resulting in several times faster than conventional hyper-parameter optimization. Key features include:
Easy of usability : hpopt takes only time as control parameter, which we believe the most intuitive to most users. auto-config sets internal control parameters automatically with the given time constraint, and guarantees HPO will finish in the constraint.
Free from interrupt : hpopt communicates via persistent storage. Using this, hpopt supports pause/resume/reuse. hpopt always checks that there are previous successful HPO runs and ask you about reusing optimized hyper parameters.
Scalability : hpopt provides both sequential and parallel methods. You can select one depending on your training environment. If you have multiple GPUs, you can accelerate HPO utilizing all of GPU resources.