Large User Study (ProtoECGNet over HEEDB)

[StevenSong]

• tune multi-arm ProtoECGNet
  • check loss term weights
    • potentially zero out contrastive loss and use protopnet defaults (clst 0.8, sep 0.08, div 100)
    • if current and defaults are bad, need to optimize
  • check protoecgnet cls init
  • replace learnable l1 and l2, optimize
  • optimize LR schedule and learning rate
• get sahil things to send to attending (need to confirm details)