Add support for NNJA-AI Data

[jacobbieker]

BrightBand has released the NNJA-AI dataset now here: https://www.brightband.com/data that is cloud-optimized data of observations from NOAA's data assimilation process.

Detailed Description

Context

This could be a really nice way of training on pure observations or just including observations in a consistent format.

Possible Implementation

They have a package that supports querying and getting the data. This would be wrapping that in a dataset to be able to load it on the fly, or generate examples on disk.

[yuvraajnarula]

@jacobbieker can I work on this issue?

[jacobbieker]

Yeah, go right ahead!

[yuvraajnarula]

hey @jacobbieker, I was trying to work on this issue and I had a doubt related to this do you want more of a nnja-ai wrapper for the dataloader.py and show observation in the form of graph?

[jacobbieker]

Hey, yeah, what we want is a data loader to returns the observations and their Metadata. Visualizing it would be helpful but not needed. But we do want it to be an output of shape (time, latitude, longitude, metadata-channels) for each observation.

[yuvraajnarula]

@jacobbieker

In the first output, I tried a batch size of 16; in the second one, I tried without it and ran a loop to show you the output of the first iteration.

[jacobbieker]

Hey, yeah, exactly like that!