Showing progress of a run in qanat experiment statusΒΆ

You may hae noticed in the output of the qanat experiment status command that a column progress is present. This is a nice option that allows to know where your experiment is at.

Obviously, the notion of progress differs from experiments to experiments. For example, in a machine learning experiment, it can be the training progress. In an optimisation problem, it can be the number of iterations.

To be able to have generalisation, Qanat parse a file progress.txt in the storage_path results directory that you have to create in your experiment executable.

Two options:

  • TQDM progress: Tqdm is a nice python progress bar library. It gives the option to redirect its output to a file that can be parsed by Qanat. To tell Qanat that this is a TQDM format, you need to write at the first line of the progress.txt file: tqdm and then redirects the output to the file. For example, by writing:

f_tqdm = open(os.path.join(args.storage_path, 'progress.txt'), 'w')
f_tqdm.write('tqdm\n')
for epoch in tqdm(range(args.epochs), file=f_tqdm):
    ...

  • count_total progress. This is a more general approach where you write at the top of the file a string of the format: count_total=NUMBER where NUMBER is the total number of operations to do. Then at each subsequent line, you can write the number of operations done (1 or more). This is is useful for process that are in parallel and append at the end of the file. Finally, if you write finished at the last line, Qanat will skip the computation of the percentage and yield 100%.

What happens when there are several groups?

Qanat will do an average of all the percentages over all the groups.