This Pull Request gives the option of running the training script on AzureML. To launch a training run, modify the config file in the azureml folder with your subscription ID, workspace name, etc. You can launch multiple runs with different hyperparameter settings by using the sampling, similarly to the default config file which launches 3 runs on different datasets. The PR made a new argument parser to be compatible with both the existing code and the AzureML argument format. There is also a new file, train_azureml.py which is identical to train.py except for the argument parser it uses.
This PR should enable more automated experimentation with the software, as well as enabling faster training through the use of cloud computing.
This Pull Request gives the option of running the training script on AzureML. To launch a training run, modify the config file in the azureml folder with your subscription ID, workspace name, etc. You can launch multiple runs with different hyperparameter settings by using the sampling, similarly to the default config file which launches 3 runs on different datasets. The PR made a new argument parser to be compatible with both the existing code and the AzureML argument format. There is also a new file, train_azureml.py which is identical to train.py except for the argument parser it uses.
This PR should enable more automated experimentation with the software, as well as enabling faster training through the use of cloud computing.