Currently we don't have duplicate names in experiments in a namespace. This is not going to work when a training run is run more than once by changing the hyperparemeters. We should
- Hash the metadata attached to the experiment.
- Create an auto sequence number and append to the experiment name if the user doesn't set a unique name.
- The other idea would be to figure out how to create unique names from the user provided names and the unique hyper parameters.
Currently we don't have duplicate names in experiments in a namespace. This is not going to work when a training run is run more than once by changing the hyperparemeters. We should