Use sweeper result in consecutive training run

[faroit]

We are using a sweeper plugin to find a learning rate for a lightning system.

What is the best workflow to use the returned optimized hyperparameters from the sweeper in a new run?

Example

the sweeper is configures with

hydra:
  sweeper:
    _target_: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper
    params:
      optimizer.lr: interval(0.00001, 0.01)

python train.py hydra/sweeper=optuna
python train.py optimizer.lr=$optimized_value$

I guess there are a couple of options:

• read the sweeper result file and extract the parameters from there manually e.g. python train.py sweeper_result_path=... Likely related to Specify the schema for returning the result of optimization. #468
• grab the optimized values from the logger when the sweeper has finished and convert them to variables using bash to set the optimized value
• modify the sweeper or add a callback to set environment variables with the optimized values
• ...

Did I miss something? Which one would you recommend?