Open jacobbieker opened 1 year ago
Going to attempt a few more tests -> 1. Try MetNet that outputs all future timesteps at once 'MetNetSingleShot', 2. Run with just sun positions to see if it can learn the general trends of night time, 3. Run with smaller inputs (64x64) to fit all 96 timesteps in at once. For all of these, also try running with multiple GPUs on donatello, the new torchdata might have fixed some issues
Pull Request
Description
Adds training script for the new MetNetPV model version
Fixes #
How Has This Been Tested?
Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration
If your changes affect data processing, have you plotted any changes? i.e. have you done a quick sanity check?
Checklist: