Moving R plots to julia

[ajinkya-k]

split off from #80

[palday]

thanks for this! I had a few suggestions for efficiency and making the plots better match the important details of the original.

It's far from trivial to match the lattice aspect = "xy" behavior, which goes back to some advice from Bill Cleveland about making the 'standard' line 45 degrees. (Logic is: the human eye is very good at detecting deviations from 45 degrees, so we quickly see deviations from standard. ) It's easy to have a 1:1 aspect ratio and a y=x line for the trivial case, but for the data here, we would have to do something like have the 45 degree line match the mean slope across participants.

[palday]

why disable the legend?

[ajinkya-k]

I disabled it because the exact subject ID's dont matter and the legend was way too big to fit the plot. This is the current version deployed on embraceuncertainty.com:

[palday]

Suggested change

	draw(
	data(pastes) *
	mapping(
	:cask,
	:strength,
	col = :batch => sorter(sort!(combine(groupby(pastes, :batch), :strength => mean => "strength" ), :strength).batch)
	) * visual(Scatter; marker = '∘', markersize = 15, color = :blue) +
	data(combine(groupby(pastes, [:cask, :batch]), :strength => mean => "strength" )) *
	mapping(
	:cask,
	:strength,
	col = :batch => sorter(sort!(combine(groupby(pastes, :batch), :strength => mean => "strength" ), :strength).batch)
	) * visual(Lines; color = :blue);
	axis = (; xticklabelsize = 10),
	figure = (; size=(700, 350))
	);
	f = current_figure();
	colgap!(f.layout, 4)
	f
	batch_sort = sort!(combine(groupby(pastes, :batch),
	:strength => mean => :strength),
	:strength; rev=true).batch
	paste_means = combine(groupby(pastes, [:cask, :batch]),
	:sample,
	:strength => mean => "strength" )
	draw(
	data(pastes) *
	mapping(
	:strength,
	:cask => "Cask within batch",
	row = :batch => sorter(batch_sort)
	) * visual(Scatter; marker = '∘', markersize = 15, color = :blue) +
	data(paste_means) *
	mapping(
	:strength,
	:cask => "Cask within batch",
	row = :batch => sorter(batch_sort)
	) * visual(Lines; color = :blue);
	# axis = (; xticklabelsize = 10),
	facet = (; linkyaxes=:minimal),
	figure = (; size=(350, 800))
	);
	f = current_figure()
	rowgap!(f.layout, 4)
	f

closer to the original and avoids computing some things twice

[ajinkya-k]

the call f = current_figure() is needed because draw actually refurns a FigureGrid not a figure. Also this was the suggested way of using row and colgap with aog figure by julius

[palday]

I left current_figure in this one, but if you look at the other suggestion, you'll see that I grab the figure out of the FigureGrid. 😉 I'm fine with leaving it as current_figure() though.

[ajinkya-k]

oh idk if i incorrectly applied the suggestion on my local but i got errors with that. can you paste a one line here to extract the figure from figure grid?

[palday]

figgrid = draw(...)
fig = figgrid.figure

[ajinkya-k]

I think the version with cask on the x-axis looks better. If we want to keep the original orientation we should atleast make the figiure flatter or more square

[ajinkya-k]

thanks for this! I had a few suggestions for efficiency and making the plots better match the important details of the original.

It's far from trivial to match the lattice aspect = "xy" behavior, which goes back to some advice from Bill Cleveland about making the 'standard' line 45 degrees. (Logic is: the human eye is very good at detecting deviations from 45 degrees, so we quickly see deviations from standard. ) It's easy to have a 1:1 aspect ratio and a y=x line for the trivial case, but for the data here, we would have to do something like have the 45 degree line match the mean slope across participants.

this is in reference to Figure 3.2 right?

[ajinkya-k]

I think i can recreate banking to 45-degrees. Do we want to do median banking or average banking?

[palday]

I think i can recreate banking to 45-degrees. Do we want to do median banking or average banking?

average, please, that would be swell!

[ajinkya-k]

okay! I have the median version ready so i will push because it has other computation in there, but once I have it working i will switch to average banking

(EDIT: I implemented median because it has a closed form solution)

[palday]

okay! I have the median version ready so i will push because it has other computation in there, but once I have it working i will switch to average banking

(EDIT: I implemented median because it has a closed form solution)

@ajinkya-k oh in that case, median is fine.

[ajinkya-k]

@palday I checked the lattice package documentation and they use median banking. So should we stick with that?

see page 142: https://cran.r-project.org/web/packages/lattice/lattice.pdf

[ajinkya-k]

@palday

For multiple.qmd figure we have two options (my preference is option A):

Option A:

Option B: this incorporates your suggestion but i resized the figure to 600 x 600

[ajinkya-k]

Once we decide on the orientation for the cask-strength I think this is good to merge

[palday]

@palday

For multiple.qmd figure we have two options (my preference is option A):

Option A:

Option B: this incorporates your suggestion but i resized the figure to 600 x 600

Vertical matches the plot for crossed random effects, so let's do that.

Can you label the axis "cask within batch" though to make it very clear that we're looking at nested effects?

[ajinkya-k]

Will do. Just to clarify by vertical you mean option A right?

[palday]

Will do. Just to clarify by vertical you mean option A right?

Yep, your preferred orientation 😄

Done!