So I see that @rjnw and @samth are working on space reduction. I’m just about to join the fray. We should coordinate, so that we don’t trample on each other’s work.

If you can fix the figures that I’ve left in a bad state, that would be great

But really we need to cut multiple pages of content

Note: due to previous commitments, I will be offline essentially all day tomorrow. So whatever space reduction I will be doing, I will be doing today.

I can definitely work on figures. I’ve been able to hack that before.

But as I’ve said before, I have no clue how to cut multiple pages of content (and have the paper remain understandable).

One suggestion sam has is to remove LDA from evaluation section, but that is not going to cut a lot.

Whether we keep X in the evaluation section, for any value of X which is currently there, really depends on what story we’re telling, and thence what best illustrates that.

I think LDA doesn’t contribute to any particular story, it’s just another piece of evidence

I think that if we want to aggressively cut, we should start thinking about appendices (including online). We can have the results in the paper, but the details elsewhere.

That seems reasonable

I think we need to tell the story in section 3 with at least one fewer page

In which case, sure, we can cut LDA. Or even move most of it to online. As I don’t think PLDI has appendices-that-don’t-count-lengthwise rules.

And probably you’re the best positioned to do that

Sure, I will work on that.

It’s 6.6 pages now, cutting it to 5.6 ought to be feasible. And yeah, that section is indeed the one where I have the best chance of being able to cut without making it impossible to understand.

PLDI web site says Appendices should not be part of the paper, but should be submitted as supplementary material. Supplementary material should also be anonymized, as described below.

Supplementary Material Authors are free to provide supplementary material if that material supports the claims in the paper. Such material may include proofs, experimental results, and/or data sets. This material should be uploaded at the same time as the submission. Reviewers are not required to examine the supplementary material but may refer to it if they would like to find further evidence supporting the claims in the paper.

I think we should use that to our advantage — headline results of evaluation in paper, everything else in appendices.

Anyways, I will work on section 3 first. Then figures. I am sure someone else is in a much better position than I to deal with the evaluation section.

PLDI guidelines say to use numeric references, but we use \citet, \citeauthor, \citep — should all these be changed to plain \cite ?

I think we can configure that at the top

It is already configured like that.

right, we’re already doing it right

I think it might be \citeauthor that overrides that

citet is to put the author’s name in, citations are not a part of speech

Jacques, what are you working on?

Shrinking section 3.

Would you like some suggestions as to what to move to appendix?

Anything from section 3?

Yeah. Figure 2: Cauchy, StudentT, Gamma can become "…" Figure 4 can go away (and I wonder how much of Section 3.1 can be smooshed into Section 3.2) Section 3.3 can have its first half (the intuition part) removed, and Figure 6 can have most rules moved to appendix (except the top two rules, which are the base cases, and one of the remaining rules, to show how recursion works)

And I guess nobody is working on Figure 1 currently?

On Figure 1: I don’t think so. I certainly have a few ideas of how to squish it into 1 column. Might be finicky though.

If you’re in Section 3 then let me take Figure 1.

Ok.

And thanks for those suggestions - good ideas. Will implement.

I have implemented all of @ccshan’s suggestions (except for the smooshing, which I think should not be done), to good effect. Section 3 is now 5 pages, which is shorter than what I was hoping to achieve. There are still a couple of equations that go wonky to fix (so we’ll lose a bit of space there), but I think this works.

Section 3 is in decent shape now. I’m going to look for ‘long equations’ in other sections now.

Figure 1 fits!

Hurray!

I bet that was, er, challenging!

Wow, Figure 1 more than fits, that’s a work of spacing-saving art too. Nicely done.

I’ve just fixed the last of the really bad overhangs. The rest can be fixed later, potentially post-submit even.

13.5 pages, with the double introduction still in place. So there’s hope that after that is rewritten, there will be just 1 page left to shrink.

Unfortunately, I’m just not sure how much time I’ll have to help with that (sorry).

Oh, and I switched the title to the one that @samth liked and fits with the ‘introduction seed’.

so I am able to recreate the gmm gibbs plots, @samth any request for dimensions

2:1 ratio

And can you make it look like what I produced?

the second plot in figure 10

I will change the llvm-backend to hakaru

Can you remove the extra lines as well?

I added more to x-axis, let me try again

how about this?

Can you remove the top and right hand axes?

I also made one 4:3

I pushed new smaller plots, can someone take a look and give feedback on how they look in the paper

the plots look good

i only wish we could add some error bars in the legends for figure 10

I couldn’t find any option is the plot library for that

@rjnw yeah, that’s what I figured

I just pushed a change to Fig 9

can you check that I didn’t break it

and also, it would be nice to move some of the let bindings to single lines

@ccshan do you plan to do the intro revision?

@samth Feel free to take a first stab… I’m just exhausted right now.

new intro stabbed

@rjnw can you shrink fig 11 too?

we still need 1.25 fewer pages, but we’re doing better