ENH: Canonic SI frequency

[JoElfner]

Since pandas is widely used for data science, including working with time series, a canonic representation of the frequency, as defined by SI standards, physics etc. would be a nice improvement.

To summarize differences between the general definition of frequency and the pandas definition of frequency:

• General definition: Frequency is in the unit of 1/time_period, for SI unis more specifically 1/second=1Hz.
• pandas definition: Frequency is the time period, so it is exactly the opposite of the physical/general definition.

Since many calculations require the actual frequency, for example summing up velocities to get the length, this would facilitate many calculations.

Proposed solution:

Additionally pd.DatetimeIndex etc. should get a new attribute, f.i. named freq_canonic, that represents the actual frequency of the index in Hertz.

To avoid breaking the API, I'd implement the frequency as a data attribute, which can be accessed by the user, but has no effect on time series calculations.

I could try making a pull request for this feature, if you are interested.

[jbrockmendel]

Is the DatetimeIndex attribute the main sticking point? Or do you need something like a Hertz dtype (which would be similar to #10349)?

[JoElfner]

Good question... :)
I guess if the Hertz dtype was added automatically during DatetimeIndex creation, it would be a really nice solution. But would it be possible to directly make some calculations with the Hertz dtype without the need for any conversions? Like f.i. velocity/Hertz_dtype?
If not, imho an attribute is quite important. F.i. a period/freq of '20s' would yield DatetimeIndex.freq_canonic as a float, in this case 0.05 (unit Hz: 1/s).

[jbrockmendel]

closing as no action