Comments on chapter 3

[dicook]

• [x] abstract needs to be written to reflect this chapter/paper
• [x] hat should only be over the parameter letter (eg sigma), not both sigma and the power of subscript - seems ot be others in various places.
• [x] We use a computer vision model to estimate this distance measure with a residual plot (page 47): change with by for
• [x] Inconsistent heading capitalisation. Probably all sections should be sentence case. Most, but not all, are title case
• [x] there is potential for confusion between "distribution" of predictors and distribution of residuals. you have both in the training set, but only the discussion of the residual distribution is in section 3.7
• [x] 3.10.2 is empty
• [x] table 3.6 has a space after 5 rows, maybe this should be removed, or placed after four rows to be symmetric
• [ ] Fig 3.11 caption needs re-working: First sentence should be purpose not the type of plot, eg "Examining under- and over-prediction of the model relative to magnitude of distance ." Also, shouldn't the denominator be D not Dhat? If Dhat is bad, then our comparison is bad?
• [ ] Check your other plot captions for this structure too. First sentence needs to describe purpose.
• [x] Right-justify numbers in tables. Think table 3.7 might have left-justified numbers
• [ ] Fig 3.16 has horizontal striping. Why is this? Has the detection rate been computed for small sets of plots?
• [x] Be sure captions end in a "."
• [x] Spell check, "differnence", "gnealized"
• [ ] Fig 3.17 is an example of where the attention map shouldn't be used, because it's not significant. At least the summary of this plot is not helpful for understanding your set of plots. Would recommend using a strong result when you first show the sequence of summary plots, then later for this example maybe only the residual plot and plot C. Also in the design of plot C, can you make a shorthand for "95% quantile of the null distribution" so the legend isn't so cumbersome - the full text can be in the caption. Maybe have a letter to denote "95% quantile of the null distribution" generally in the paper for smoother reading, since you use it again and again.
• [x] include a "Supplementary materials" section that has links to the GitHub, etc