CLI?

[erikhuck]

Right now, the Tracker can only be used within python scripts. There may be users who would like to get gpu (and other usage) information about their job without needing to write a python script (e.g. a python script that they don't want to edit or a job that doesn't run python at all). It may be helpful to introduce a commandline interface where the tracker can run in the background while some other terminal command runs. The challenge would be getting the process information of the command to profile. It could be the the subprocess library provides that information.

[erikhuck]

Other things to add:

• An option to have --output
• By default, the output should print to the console
• If a file path is provided for the --output option, the output can be saved to that file
• Option for output --format as well
• Choices can be text, json (use json library), or csv (use pandas)
• Default to text
• The text format outputs the same as Tracker.__str__()
• Should add to_frame() (returns pandas dataframe) and to_json() (returns dictionary) methods to Tracker.
• to_frame can reuse to_json and convert the dictionary to a dataframe.
• These methods can optionally save the measurements to disc with am optional file path parameter. Either way, returns the corresponding object.
• We need to make sure __str__ is included in the documentation.