-
Notifications
You must be signed in to change notification settings - Fork 1
Requesting Hyperparameter Configurations and Reporting Loss
Using the next_configuration
option will request the next configuration from the server. You will then need to apply these hyperparameters to your model.
A single configuration can be requested, or a batch of configurations can be requested.
This will return a hyperparameter configuration stored as a (JSON) Dictionary for all variables in the experiment.
In a typical experiment setting, you will want to run a loop where you will continuously request configurations, apply the configuration to your model which should report some loss or error factor to be minimized, and then report_loss
to the wwu_tinker server.
The report_loss
function is critical to the process, as it will allow you to fetch the best configuration later on.
Below is a simple loop demonstrating this process
for i in range(1, 10):
config = test_experiment.next_configuration()
result = ml_model(config['x'], config['y'])
print (result)
config.report_loss(result)