Possible to use a training job that has not been launched yet to launch a prediction job

[chlmarchal]

Hi! I noticed that when creating a new prediction job, it is possible to choose a training job that exists but has not been launched yet. In this case, the job fails and the error is the following:

Traceback (most recent call last):
  File "/app/run.py", line 378, in <module>
    main(sys.argv[1:])
  File "/app/run.py", line 288, in main
    binary = str2bool(properties["binary"].value)
KeyError: 'binary'

Since the training job has not been launched yet, its PropertyCollection containing the keys "binary" and "classes" is empty when the prediction job needs it during its initialization and the shown error is raised.