Using null in train config leads to errors

[keraJLi]

null is standard yaml syntax, the equivalent of python's None. Therefore using it to specifiy missing values in a training config should work, but it leads to an error. I'm not sure why exactly, but its related to the way saving yamls is handled in mle-hyperopt, more specifically in save_yaml in mle_hyperopt/helpers/utils.py. At the point where the file is evaluated as

data = json.dumps(convert(obj), indent=1)
data_dump = ast.literal_eval(data)

the data variable contains null without quotes, which cannot be parsed (its also not a variable).