Yesterday

• Worked a little on Lyapunov but mostly cleaning up HTR code and notebooks all day
• Messaged T

TODO

• NSG group meeting
• Meeting with A and possibly S/J
• HTR
  • Properly figure out what’s going on with outliers and whether you can fix them with simple rules
  • Repo big merge:
    • preprocessing per-dataset,
    • a single experiment running notebook, and
    • separate postprocessing repos per table/results analysis
      • Have ‘big table’ ‘drop page table’ ready to go EoD
  • Try WER and ANLS*; see if they’re reasonable
  • Craft prompts R3 and run them overnight (use batch mode)