Implementing alternative weather API (VisualCrossing.com)

[jat255]

As described in #51, DarkSky's API will be retired in a few months and isn't open for new signups, meaning new users of FitTrackee cannot get weather data.

After analyzing a few options, I've chosen to implement VisualCrossing.com's timeline API.

Tasks:

• Add environment variable and backend python code to get weather data from new API
• Add icons for VC's particular status codes
• Update client to show wind speed in miles per hour in "imperial" unit mode (at least in the US, mph is used for wind rather than ft/s)
• Update client to show temperature in Fahrenheit in "imperial" unit mode
• Tests

[jat255]

@SamR1 aside from tests (are there existing tests for weather functionality? I don't see them, if so), this should be ready for review.

It implements VisualCrossing as a new weather data source, and makes some small tweaks to how the weather is displayed when using Imperial units. It looks as follows:

Imperial:

Metric:

I'm unclear why the sun is displaying as blue... The icon I uploaded has it as yellow.

EDIT: I see the inversion is happening because of the --workout-img-color style. I guess this is because the DarkSky icons are monochrome

[SamR1]

Hi

Thanks for the PR.

Just for information, I haven't yet reviewed the alternative weather service providers (sorry, I don't have a lot of free time at the moment to work on FitTrackee).

[jat255]

👍 I went ahead with VisualCrossing.com due to limitations I felt would cause issues with the other providers mentioned in that issue. Namely, Openmeteo's history API is delayed 5-7 days, so is not useful in real-time, and meteostat has a relatively low free tier of 500 calls/month. VisualCrossing seemed like it checked all the boxes, so to speak.

I implemented it with a new environment variable VC_WEATHER_API_KEY and a set of icons that match what their API might return. It could be more general by instead using something like two env variables such as, and then selecting the right code to use in weather.py, but I wasn't sure if supporting multiple APIs was really worth it (since DarkSky is going away in a month or two):

WEATHER_API=visualcrossing   # or darksky, meteostat, openmeteo, etc.
WEATHER_API_KEY=123456abcdef

[SamR1]

Hi,

So far, I haven't found the perfect replacement for DarkSky.
I reviewed some weather APIs, I updated the issue. It's not exhaustive, I may review others later.
VisualCrossing historical data is only per hour (DarkSky is more accurate).
Another alternative is OpenWeather One Call API 3.0 but credit card is required if I'm not wrong.

I worked on a base for multiple APIs support to keep DarkSky in addition to VisualCrossing and in case other APIs are supported in the future: 64d770a...alternate_weather_api
I refactored some code, in particular to add tests on API and Client sides.
I prefer to keep the generic icons (since VisualCrossing icon labels match existing ones with default value (iconSet=icons1)).

If it's OK for you, I will push the changes on this PR.

---

Edit:
Comparison between the 2 APIs

• workout duration: ~48minutes

DarkSky	VisualCrossing

• workout duration: ~35minutes

DarkSky	VisualCrossing

[jat255]

Thanks for pushing this through! I just got around to updating after a while, and it appears to be working well for me. Different but related question... do you know if there's a simple-ish way to reprocess old workouts to add weather info? I see it's stored with each workout in the DB as a JSON in the weather_start and weather_end. I could certainly write a script to do this, but I figured I should check if there's a tool for this already

[SamR1]

Hi,

Unfortunately it's not possible to reprocess workouts for now (maybe in the version with workout processing rewrite).

[jat255]

Gotcha, I see that in edit_workout, weather information is not processed at all, so even via a PATCH command the weather info cannot be edited.

To get around this I wrote a little script that uses the FT API to get workouts and their GPX files, then uses the FT code to call the weather API, then finally directly updates the weather_start and weather_end columns in the backend database directly. Hacky, but it works!

That's available here in case that helps anyone. It makes use of/requires the FitTrackeeConnector class from s2f.py in https://github.com/jat255/strava-to-fittrackee