This PR implements a new WandBLogger class that can be added to and Exploration to track the evolution of all parameters, objectives and other analyzed parameters to Weights & Biases (https://wandb.ai).
Changes
Implemented new base Logger class, and WandBLogger.
Add logger argument to Exploration.
Add a new data property to Trial, which returns a dict with all parameters and values.
When running with threading, do not force the generator to be updated with the one from persis_info. This is only needed when running with multiprocessing, as a workaround to not being able to pass the gen to libE by reference.
When not running with threading, remove logger from generator before passing it back to the manager. This prevents a RuntimeError. Since this only needs to be done when sending back, a new _prepare_to_send_back method has been added to the gen.
Add the new class to the API reference.
Create new user guide entry about how to log to W&B.
Added test that will log to W&B (currently, to my user account).
This PR implements a new
WandBLogger
class that can be added to andExploration
to track the evolution of all parameters, objectives and other analyzed parameters to Weights & Biases (https://wandb.ai).Changes
Logger
class, andWandBLogger
.logger
argument toExploration
.data
property toTrial
, which returns a dict with all parameters and values.persis_info
. This is only needed when running with multiprocessing, as a workaround to not being able to pass the gen to libE by reference.RuntimeError
. Since this only needs to be done when sending back, a new_prepare_to_send_back
method has been added to the gen.See example: