Smoothing/averaging and outlier rejection for GPS elevation data

[privong]

I've noticed data from some GPS systems (phones, on my experience) can be relatively noisy, particularly the elevation values. This results in imported GPX files giving erroniously large elevation gain/loss values. Smoothing over several adacent values would give a more realistic elevation profile and total gain/loss values. Perhaps this could be added as a checkbox option (with a smoothing window width) to the new record entry window?

[dgranda]

Hi privong, thanks for feedback. I think it is a good idea to add it as an option when importing new files.

Currently this feature is not working well (see #17), so it is a good idea to have your suggestion in mind when fixing it

[privong]

Sounds good. What is the estimate time scale for fixing issue #17? I'm not sure I can delve into the pytrainer code to add this type of capability, but I may code up a pre-processing script with python/numpy to do this before I import the gpx files into pytrainer. If I can get that done, would you find that useful?

[dgranda]

What is the estimate time scale for fixing issue #17?

I don't have a fixed answer on this, but if it prevents someone to do something that improves pytrainer I could rearrange some entries in my to-do list ;)

If I can get that done, would you find that useful?

Well, if the idea is to have the option of smoothing altitude figures when importing files (which is in place and in use in a later stage), I don't see the benefit of the mentioned pre-processing script.

Actually I think current feature request would be easier than #17 to get done because logic to change altitude data works fine and changes are written into xml file (dealing with graphs is what fails) so we just miss adding the graphical check on gui, implement the request and write some testcases, not a bad option to get familiar with pytrainer ;)

[privong]

Well, if the idea is to have the option of smoothing altitude figures when importing files (which is in place and in use in a later stage), I don't see the benefit of the mentioned pre-processing script.

I was just thinking that having an algorithm written up already might help with implementing it into pytrainer.

Actually I think current feature request would be easier than #17 to get done because logic to change altitude data works fine and changes are written into xml file (dealing with graphs is what fails) so we just miss adding the graphical check on gui, implement the request and write some testcases, not a bad option to get familiar with pytrainer ;)

I'll go ahead and fork it and see if I can find some time to dig around and implement this.

[jstepien]

Two years ago I tried to tackle a similar problem: I tried to make pytrainer smooth graphs instead of imported data. You can find my patches in mailing list archives. HTH.