Jacob Prince-Bieker

Yeah, so here is the code we use for loading the netCDF files for training: https://github.com/openclimatefix/ocf_datapipes/blob/a687eee51545411a0e80ff13ef49d1c67b1dca5b/ocf_datapipes/load/pv/pv.py#L26-L65

And yeah, each file is read individually, we pass in the metadata and the netcdf file at the same time though

Yes

First of all, thanks for submitting this! And releasing some of the data! Would it be okay if I rehost it on HuggingFace Datasets the sample you released? Even not...

@JackKelly @peterdudfield Some more support for satellite being quite important for short term forecasting at least. Not solar, but they did use EUMETSAT data

https://arxiv.org/pdf/2203.10114.pdf has more details, and is the lightning-only original work

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...

So, this is the actual implementation being used for axial attention: https://github.com/lucidrains/axial-attention/blob/eff2c10c2e76c735a70a6b995b571213adffbbb7/axial_attention/axial_attention.py#L153-L178 which seems like ti is doing both vertical and horizontal passes. But, I just realized that we don't...

Yeah, set to 1 for the debugging to get it to overfit first. And yeah, the position embedding is saying which pixel we are in, and the location information of...

The paper just says the 16 channels of Goes, so unfortunately, I don't know, I would assume whichever of those options are the raw data, and not post-processed images.