pangoLEARN data release 2021-03-29

Release notes

• pangoLEARN trained with downsample. Sequences dowsampled from 382,806 designated seqs to 177,916 seqs, maintaining diversity representing every node in the tree.
• Downsampling took 2.5 hours: training time went from >20 hours to 3 hours.
• We're hosting both the designated metadata here as well as the final downsample that went into the training.
• 1,249 lineages, 1,169 lineages downsampled as they had greater than 10 sequences designated.