Ordering of `forecast_time` for training
The training data spans multiple years. How should the forecast_time
be ordered?
Currently, climetlab
returns non-monotonically forecast_time
. (Jan 2nd 2000, Jan 2nd 2000, ..., Jan 9nd 2000, Jan 9th 2001, ...)
In the renku datasets I sort forecast_time
monotonically by .sortby('forecast_time')
. (Jan 2nd 2000, Jan 9th 2000, ..., Jan 2nd 2001, Jan 9th 2001, ...)
I could change the renku datasets to non-monotonically if that is more appealing.
Which of the two ordering styles do you find more useful for your training pipeline? That involves how climetlab should return the data and how renku datasets stored via git lfs
are stored at your convenience.
Please answer by comments and/or upvote/like vs. downvote/dislike my three comments below.