Analyze embeddings in Mozambique

[alterapars]

• move Moz. data to disk and connect vm with it
• generate embeddings (non-ts) there with one of the proven models
• train seco model on mozambique
• compare embeddings with county-level stats
• and coarse land use labels

this way, if it fails ... we at least have the code for embedding comparision on a grouped level than can be applied everywhere
if it works... we have smth to show in africa