[Request] Max Error parameter for rounding

[supRy]

The examples I am providing are based on the following use case, but I believe this applies more generally across the tool. I am using naturaldelta to describe the duration of a route (for example Miami, FL to Atlanta, GA is 9 hours).

One concern is the rounding (2 hours 59 minutes becomes 2 hours), but #154 seems to resolve that concern.

My second concern is the precision in the range of about 1-10 hours. Currently, 2 hours 29 minutes becomes 2 hours. I feel like that misrepresents the actual duration given that it is off by 20%. I am wondering if "2 and a half hours" would be a better result. I'm not sure on the most natural language way to write this, "2 hours 30 minutes" is probably the most generalized form.

Initially, I was thinking this would be connected to the rounding, perhaps as an alternative option rounding to half values. So 45 minutes to 1 hour 15 minutes would round to 1 hour. 1 hour 15 minutes to 1 hour 45 minutes would round to 1.5 hour. However, thinking more broadly different applications will have different precision requirements. Allowing users to set a max_error and then using that to determine the best rounding would be a good solution.

So considering a max_error of 10%, 8,880 seconds (2 hours, 28 minutes) would become 2 hours and 30 minutes. This is because 2 hours is 19% off (greater than 10%). So we drop to the next smallest unit (30 minutes) and that results in an error of 1.4%, so we stop here. Now thinking about 9,660 seconds (2 hours 41 minutes) and a max error of 5%. This gets rounded to 3 hours, but has an error of 12%, so we cut to the next increment (30 minutes) and round to 2 hours 30 minutes. That has an error of 7%, so still too much. The next increment would be 15 minutes, so we round to 2 hours 45 minutes and stop there with an error of 2%.

Using a max_error parameter also resolves the statement I made earlier "My second concern is the precision in the range of about 1-10 hours". From this statement I can see that I care about the precision of 30 minutes up to 9 hours 30 minutes, so my desired precision is 5% (0.5 hour/9.5 hour). Beyond that, I would only get hours.

I'm thinking these increments (for timedelta) would be:

• millenium
• century
• decade
• years
• 6 months
• months
• days
• hours
• 30 minutes
• 15 minutes
• 5 minutes
• 1 minute
• 30 seconds
• 15 seconds
• seconds
• milliseconds
• microseconds

Other functions would need to have their own increments developed.

[hugovk]

🚀 Development has moved to https://github.com/python-humanize/humanize 🚀

Please open new issues at https://github.com/python-humanize/humanize/issues