cogsci19 todo's

[judithfan]

• Notebook cleanup: construct clean versions of all jupyter/R notebooks in a "clean" directory
• Behavioral measures, 1: distill the myriad things we've been plotting/fitting (e.g., num strokes, draw
  duration, prop correct) into sensible composite measures, e.g., that bundle accuracy/cost
• Behavioral measures: BIS plot (for background: https://www.ncbi.nlm.nih.gov/pubmed/30022459)
  • include control condition (pool with repeated condition)
  • write out error bars
  • lmer/brms it
• between-sketch similarity analyses:
  • within-pair converge + across-pair divergence
  • compile t-stats for slope on permutations, make insert (and report in text)
  • compare different visual features
  • within-pair baseline: use permutation test as baseline (for each round/target, scramble across pairs... then scramble across pairs and round)
  • between-pair baseline: permute across rounds, preserve target, within-pair structure
• overhaul the svg rendering pipeline to avoid matplotlib artifact
• within-sketch analyses
  • temporal structure of strokes
    • self-similarity baseline sanity check -- 1 deleted, 2 deleted, 3, ... how similar is the same sketch to itself with accumulated stroke deletion?
    • stroke deletion analysis: from beginning vs. from end, what impacts between-rep similarity the most?
• Human experiment: test recognizability of sketches from diff rounds
• cogsci paper:
  • write communication task methods
  • write recognition task methods
  • write communication task results
  • write sketch recognition results
  • generate task/stimuli figure (Figure 1)
  • generate behavioral results figure (Figure 2)
  • generate sketch feature drift schematic (Figure X)
  • outline model results
  • write within-pair sketch-similarity results
  • write between-pair sketch-similarity results
  • write temporal structure sketch-similarity results
  • generate model results figure(s)
  • draft introduction
  • draft discussion
• Future human experiments: design context-manipulation & generalization studies, apply tweaks (interface familiarization trials, etc.)
• Future fine-grained properties of individual sketches
  • sensitivity analyses (i.e. heat map tracing)
  • similarity of part-level features across reps (i.e. finer-grained)