Summary: weekly mtg 20170201 (Byron, Steve, Matt, me) - mobeets/nullSpaceControl GitHub Wiki
Fig. 1
- A,B: make neuron boxes bigger to fit size of funnel
- C,D: emphasize that this is a conceptual figure, and not data
- C: v_t is misleading
- remove the t
- relabel "output-potent" as cursor velocity?
- but how to talk about it in the text?
- in methods, distinguish between what we're calling "velocity" in the text
- note that o-p is the "push," which has units of velocity
- just don't worry about the distinction in the main text
- label different velocity values
- C: remove boxed v_t
- C: go back to the trace version
- but deemphasize the trace so it doesn't pop out
- make it gray, no shading, etc.
- D: try to make it an inset or something?
- label 'v=3' or whatever it happens to be in C
Fig. 2
- B,C: highlight in gray in both panels, e.g., so they're connected without inset lines
- A,B,C: make columns more distinct
- B: add pie wedge to the left
- put it under the text so the degrees are a title for the wedge
- then in C we can remove the degree text labels
- output-null and cursor dir labels on separate lines
- also still need to fix placement of text labels
- put it under the text so the degrees are a title for the wedge
Fig. 3
- Unc-uni: the purple better have SOME support where the data does
- e.g., see panel B, where the data is non-zero but the hist is not
- same for Unc-emp: as you go down the rows (for dim. 2, say), the distributions change slightly
- confirm that removing constraints on max-firing fixes this!
- also make sure this is explained in the text (not captions)
- Unc-emp: makes Byron think of what Constant-cloud is doing
- need to emphasize in text that this distribution doesn't change with the output-potent value
- modulo firing rate constraints
- but then this isn't exactly true because the dists won't be the same after the firing rate constraints
- maybe emphasize that this is the marginal empirical distribution
- and cloud is the conditional empirical distribution
- "the marginal distribution along the dotted axis subject to limits on minimum and maximum firing rate"
Fig. 4
- Hab-cor green squares are outside of where repertoire would have been
- add black label of "v=3" to black line, gray label of "v=3"
- maybe add repertoire back in
- make sure some corrected points fall outside of where cloud is
- Hab-cor cartoon: take out single points (looks like discrete predictions), replace with unfilled/filled rectangles
Fig. 5
- previously, solid green rect means uniform distribution
- but in cloud it's not uniform!
TEXT:
- could explain MinFiring first all the way
- then just add one sentence saying we considered an alternate version with baseline firing rate
- then label A,B,C,D,E,F
- every time you introduce a new hypothesis, anticipate confusion with previous hyps and explain the difference
- e.g., distinguish hab from unc-emp
ANALYSES:
- add in pruning with new metrics and Nelson data, see how well it does
- what if cloud predicted distribution directly (e.g., not timepoint-by-timepoint)? would SSS figs be better? would other results be better?
- check distribution of number of points in each bin