I'm thinking that instead of using state_vector_weather or optimal_state_vector_weather. We can instead get the raw weather data (made numerical) for all weather stations between the two airports averaged on 30 mins (or some other time range as a hyperparameter) around the departure time.
So the input would be like:
| tmpf | dwpf | relh | drct | sknt | ... | skyc1 | skyc2 | ... | skyl1 | skyl2 | ... | severity | ... | month |
n_rows = number_of_stations_in_between_airports
If we have enough data we don't have to include the aircraft type, we can train one model per type, since we know that can cause issues. I'm thinking we also add the month of the year there, since we know that plays a factor.
So we'd have one model per combination of airports, and per type of aircraft, which doesn't seem unreasonable. We need to learn how to save models on sqlite files as well, I'm sure it's possible.
I'm thinking that instead of using state_vector_weather or optimal_state_vector_weather. We can instead get the raw weather data (made numerical) for all weather stations between the two airports averaged on 30 mins (or some other time range as a hyperparameter) around the departure time.
So the input would be like:
If we have enough data we don't have to include the aircraft type, we can train one model per type, since we know that can cause issues. I'm thinking we also add the month of the year there, since we know that plays a factor.
So we'd have one model per combination of airports, and per type of aircraft, which doesn't seem unreasonable. We need to learn how to save models on sqlite files as well, I'm sure it's possible.