-
Notifications
You must be signed in to change notification settings - Fork 1.2k
Description
What did you find confusing? Please describe.
The documentation is incorrect. It defined max_runtime_in_seconds for HyperparameterTuner to be
The maximum time in seconds that a training job launched by a hyperparameter tuning job can run.
Describe how documentation can be improved
Change it to the correct one as documented in here.
The maximum time in seconds that a hyperparameter tuning job can run.
Additional context
I took a job that finished in 2 days and modified the ResourceLimits as follows:
tuning_job_config['ResourceLimits'] = {
'MaxRuntimeInSeconds': 12000,
'MaxParallelTrainingJobs': 1,
'MaxNumberOfTrainingJobs': 20,
}
and it stopped on the 3rd job after 19 minutes for the 3rd job.
Furthermore, the training job itself had a much longer run time limit, and the previous job I copied the config from had all training jobs succeeding and none stopped.
This confirms it's not the maximum time for training jobs, but for the tuning job.