nowcasting_dataset icon indicating copy to clipboard operation
nowcasting_dataset copied to clipboard

Consider that the Sun's elevation is dependent on time and location when generating list of locations for each example

Open JackKelly opened this issue 3 years ago • 0 comments

Detailed Description

At present, Manager first gets a list of all available t0 datetimes (independent of location), and then samples available locations from GSPDataSource for a random sample of those available t0 datetimes.

The problem is that this makes it hard to ensure that examples are only generated when, say, the Sun's elevation is > 10 degrees above the horizon (because the Sun's elevation is depending on time and location).

Context

See issue #373 for more context.

Possible Implementation

Maybe one way to do this might be something like this (this should work but there's probably a more elegant solution):

  1. Sun.datetime_index() returns datetimes when the Sun's elevation is > 10 degrees for at least one location in Britain.
  2. For each t0 datetime, GSPDataSource.get_locations() first selects GSPs where the Sun's elevation is > 10 degrees at that t0 datetime, and then randomly picks one those GSPs for each example.

Some problems with this plan:

  1. It'll be slow!
  2. How will GSPDataSource get the Sun's elevation?

JackKelly avatar Nov 17 '21 11:11 JackKelly