PROBLEM: the train_model task can ONLY WORK if the weather location are stored as NL in the DB

[sgatto]

We are testing OpenStef at RTE and we're currently trying to use the 'train_model' task.

Our weather locations have the country field set as "FR"

The function 'train_model' does not, for good reasons, receive the country as a parameter, BUT
it does call get_model_input from Openstef-dbc.
get_model_input has 'country' as an optional parameter with default value set to NL.
The parameter "country" is used, via multiple calls, to get the weather data using 'country' as a filter.
(get_wind_input, get_solar_input, get_weather_input get_predictors -> they all ends up calling Weather.get_weather_forecast_locations.

The fact that the task calls get_model_input with the default parameter, leaves no chance other that having NL as the only country for get_model_input.

Our easiest solution, besides changing the weather location of our database from "FR" to "NL" (bizarre), is to create a copy of the entire task and explicitly calls get_model_input using country="FR".

If you still wish to use that information, than it'll be necessary to add the information to the prediction job, it'd still be a strange concept and it would limit the scope of a prediction job.

My other question on the matter is : why would you want to limit the weather location to one country for a prediction job ? Either you only have one country in the whole database, than you can ignore the information or you have multiple countries but in that case why not use them ? One good reason might be to store a field country to the prediction job and use it to infer a specific market but not for to get the weather forecast locations.

Personally i don't see any benefit in using the country field for the weather locations and I would propose to ignore that information altogether (unless you think that two cities with same name can be found in 2 different countries).