nh_run_scheduler

neuralhydrology.nh_run_scheduler.schedule_runs(mode: str, directory: Path, gpu_ids: List[int], runs_per_gpu: int)

Schedule multiple runs across one or multiple GPUs.

Parameters:

• mode ({'train', 'evaluate', 'finetune'}) – Use ‘train’ if you want to schedule training of multiple models, ‘evaluate’ if you want to schedule evaluation of multiple trained models and ‘finetune’ if you want to schedule finetuning with multiple configs.

• directory (Path) – If mode is one of {‘train’, ‘finetune’}, this path should point to a folder containing the config files (.yml) to use for model training/finetuning. For each config file, one run is started. If mode is ‘evaluate’, this path should point to the folder containing the different model run directories.

• gpu_ids (List[int]) – List of GPU ids to use for training/evaluating.

• runs_per_gpu (int) – Number of runs to start on a single GPU.