Open sophie-xhonneux opened 1 year ago
Because it is a plugin for hydra you can simply follow the hydra examples. It tries to follow the hydra way and minimize the amount a new concepts it could introduce. In fact the plugin use the exact same example as the builtin nevergrad plugin.
So there ends up not being a lot of orion specific things to document. The plugin readme discusses how relative/absolute paths are handled which is important if you desire to resume your HPO,
There is also a section showing you how to launch your hydra app to do an HPO but as noted above this is just standard hydra and nothing new is introduced.
Well I find there a couple orion specific things with regards to the database config, because orion requires a config file to access the database and visualise the results and I am unable to do this hydra orion sweep config, so an example how to make that work would be really nice because it is non-trivial.
Similarly, hydra now makes orion create a database per multirun and various questions to how that should be set up also exist. I really don't feel like it's simple an example in the documentation would be really appreciated.
Would it be possible to expand the documentation to explain how to use orion with hydra together for experiment management and hyper-parameter search?