Feature request: Access to results

[markdall]

• [X ] I have marked all applicable categories:
  • new feature request

tqdm gives nice human readable information about iterations per second. It would be great if we could access that information in order to log it, for times when we want to check the performance of a thing over time.

From this example: https://stackoverflow.com/questions/56677267/tqdm-extract-time-passed-time-remaining

0%| | 0/200 [00:00<?, ?it/s]
4%|▎ | 7/200 [00:00<00:02, 68.64it/s]
8%|▊ | 16/200 [00:00<00:02, 72.87it/s]
12%|█▎ | 25/200 [00:00<00:02, 77.15it/s]
17%|█▋ | 34/200 [00:00<00:02, 79.79it/s]
22%|██▏ | 43/200 [00:00<00:01, 79.91it/s]
26%|██▌ | 52/200 [00:00<00:01, 80.23it/s]
30%|███ | 61/200 [00:00<00:01, 82.13it/s]
....
100%|██████████| 200/200 [00:02<00:00, 81.22it/s]

I'd like to be able to retrieve 81.22, which is the last value for iterations per second. I have another script that uses tqdm and shows me k/s for a download, and I'd love to be able to log that over time. The goal is very much to have the values match what is shown (by tqdm) exactly, vs. calculating it separately and maybe coming close.

[casperdcl]

Use the tqdm object's .format_dict property

[casperdcl]

see #562 (#482 #494 #660 #172 #656 #587) and https://github.com/tqdm/tqdm#description-and-additional-stats

[markdall]

Thanks so much! This was tough to know the right search terms for.

[casperdcl]

yes but I agree that there's quite a bit of info that format_meter calculates internally from format_dict which may be nice to expose in some way. As of now it still seems that calling str() around a tqdm object and parsing the output manually might be the best way to go which seems weird.

[markdall]

I'm just not great at it and for some reason keep ending up with a number that's different than the last displayed number. But it's much better than nothing, and I really appreciate the help. Thanks!

[graue70]

yes but I agree that there's quite a bit of info that format_meter calculates internally from format_dict which may be nice to expose in some way.

This would be very useful.

I have something like the following at the moment:

logger.info(f"Reading things from '{file}'")
with file.open("rb") as f:
    progress = tqdm(
        f, desc="Reading things", unit="lines", unit_scale=True
    )
    for line_number, line in enumerate(progress):
        # do something with line

logger.info(
    f"Read {progress.format_dict['n']} things in"
    f" {progress.format_interval(progress.format_dict['elapsed'])}"
    f" ({(progress.format_dict['n'] / progress.format_dict['elapsed']):.2f}lines/s)"
)

Besides this being unreadable code, the unit is not scaled in the logging statement.

[casperdcl]

the unit is not scaled in the logging statement.

You can use progress.format_sizeof() for numbers the same way you use progress.format_interval() for seconds.

[graue70]

You can use progress.format_sizeof() for numbers the same way you use progress.format_interval() for seconds.

Wow thanks. That's awesome. Who would have thought there is some non-documented functionality hidden away in the package? :)

In that case it's only about unreadable code or the ease of extracting the relevant information.

[casperdcl]

Wow thanks. That's awesome. Who would have thought there is some non-documented functionality hidden away in the package? :)

well there's https://tqdm.github.io/docs/tqdm/#format_sizeof

[graue70]

well there's https://tqdm.github.io/docs/tqdm/#format_sizeof

You're absolutely right, but I only ever looked at https://github.com/tqdm/tqdm#documentation where it doesn't appear. I find it confusing to have multiple sources of documentation with varying content or levels of detail.