Good morning. My default task is to make Section 3 shorter and more readable.

[Sorry for being offline, travel and conference have interfered]

It seems that we’re currently almost 2 pages over length?

yes

As suggested (quite strong) by Klaus Ostermann, we could get rid of Figure 1. I don’t think it really communicates much. It would be good to resurrect it in a talk.

I’ll start doing some small space-saving edits. Not in Section 3.

@ccshan I need you to write something about LDA in section 5, where there’s currently a fixme

I also took out several of your plots to save space

The plots really help with the message that we are much faster. We should try hard to keep them, even if we had to reduce their size. And be aggressive about cutting in other places.

the plots I cut were the update times for NB, and the plots with only Psi on them

I think the message about both of those benchmarks is still very clear

Ok!

If we (selectively) had ‘let … and … in’ in some of the figures in section 5, we could gain quite a bit of space without sacrificing readability. Opinions?

Only if it doesn’t affect understanding what the optimizations do

Also, do we have any control at all over the space after the caption of a figure? Some pages have ridiculous amounts of space wasted there.

@ccshan I agree.

[I have to go off for dinner now, but I’ll be back in a couple of hours]

“Wasted” space is due to widow/orphan avoidance, which has its purpose. Space saving calls for directed efforts.

As usual with latex, saving space in earlier pages often has a dramatic effect. Section 3 is fairly ripe for that…

after removing the radar chart, we are exactly 1 page over

@rjnw I’d like to re-generate some of the plots, but the data doesn’t seem to be checked in. I guess it’s no longer possible for you to do that? Is the 50–10000 Gmm data somewhere?

No, the data files were pretty big so I didn’t check them in.

ah, ok

i’m just editing the pdf directly then

Altogether we seem to have ½ page to go.

9 lines in my local working copy

Ah, I thought your latest check-in meant you were ‘done’ with section 3 and had moved on to something else.

Ok, then I won’t be too aggressive with changes to section 1 and 2. But more frequent check-ins might be good.

I just noticed a FIXME in section 6!

I’ve now commented it out (the FIXME), so that it has no danger of slipping out in the final version.

@carette I believe @ccshan needs to fix that FIXME

Indeed. I just didn’t want to run the risk of it being forgotten.

I think the paper could go out without the missing introduction paragraph, and it would not be disastrous. With a clear FIXME, it would be.

@rjnw, would you please confirm or deny that in all plots, all Hakaru (LLVM) and JAGS curve markers represent 1 sweep, and all AugurV2 curve markers represent 100 sweeps?

In gmmgibbs both hakaru and jags are every ten, and 1 sweep for nauve bayes and lda

For augur I think it’s every 100, I will verify it when I get back to the hotel.

Praveen’s name had been misspelled on hotcrp, so I fixed it. And entered my conflicts while I was at it.

lda is every 10, nb every 100

Which system are you talking about?

(Hmm, we can almost fit the PSI plots back in…)