This package is mainly for personal use. It is extracted from one of my private projects so that it can be maintained independently.
It almost certainly does not work well for now.
This is a package that provides an OO style Training Job.
It aims to provide a common interface for different ML training job.
To use this package, you need to define a job class of your own which inherits TrainingJob.
Usually you would need a config file
(only yaml is supported for now).
from torch_training_job import TrainingJob
class MyTrainingJob(TrainingJob):
def get_model(self, model_config):
return mymodel(config)
def load_data(self, data_config):
train_loader = ...
test_loader = ...
return train_loader, test_loader
def run(self, *args, **kwargs):
"""
Main logic of your job
"""
...
...Your entry point should be like this:
job = MyTrainingJob(config)
job.run()- Make it a typical python package that can be installed by pip etc.
- Write documentation.
- Add a cookbook.
- improve default
runmethod