search

optuna / optuna

A hyperparameter optimization framework
https://optuna.org
MIT License
10.84k stars 1.03k forks source link

[feature request] More samplers. #346

Closed Rishav1 closed 5 years ago

Rishav1 commented 5 years ago

It would be amazing if support for more samplers and pruners are added. Some Meta-heuristic search algorithms like simulated annealing or hill climbing can be easily implemented. However, more complex methods like population based techniques (evolutionary algorithms, particle swarm optimization) could be challenging to implement given your current schema.

It would be nice to have more samplers in my opinion.

sile commented 5 years ago

Thank you for your suggestion.

Optuna dev team is also planning to make more sampling/pruning algorithms available. As part of that, we are planning to expose the interface of sampler and pruner so that users can implement their own samplers and pruners. Now, we are working on interface design, and it will be delivered in the near future.

About complex algorithms, as you say, it is not easy to develop them within the current Optuna framework. We are considering how to integrate those algorithms with Optuna, but it may take some time before we can provide them.

g-votte commented 5 years ago

Closing, since we're going to have more samplers such as #399