@ccshan Could you please delete tag clich�-lathered-stiffener-muckiest ? It causes me all sorts of headaches (git pull fails, etc).

We’re at 27 pages minus 5 lines… hopefully the rest of the fixes won’t need any more space!

I don’t know how

https://gist.github.com/mobilemind/7883996

I tried to do it myself, but since I don’t have that tag nor can I type it properly, I can’t.

cliché seems to be the problem. And no, typing it like I just did, doesn’t work.

I’ve read all the changes, and am happy with them. The only thing that I don’t like (but am willing to live with, as I don’t know how to fix it myself!) are Figures 8 and 9. I find the excessive indentation because of all the lambdas quite ugly.

Now that I look deeper - Figure 11 now looks tiny when compared to Figures 10 and 12. How much information is really in Figure 12? Yes, the Log Likelihood is different, and does change, but the vertical axis is misleading — even on a log scale, those two graphs would be indistiguishable if the scale was absolute instead of a massive zoom. In other words, Figure 12 could be made smaller to allow Figure 11 to be larger. The lhs of Fig. 11 seems informative — the rhs has the same issue as Fig. 12.

Should we also thank the referees in the Acknowledgements? And how far back do we go with our thanks?

Did I just do it?

I see we’re at 27 pages minus a dozen lines. I see Section 7 start at the top of page 27. You might need to upgrade your TeX installation to get newtxmath version 1.5 or later.

Do you mean that in each of those two figures you want to add one (indented) line break between the initial lambdas and the remainder body?

The meaning of a log-likelihood is invariant up to adding a constant. (In other words, the meaning of a likelihood is invariant up to multiplying by a constant.) (I’m not sure what you mean by “two graphs”.) To change the graphs’ aspect ratios, please feel free to dig in hakaru-benchmarks/output/*.rkt

I was going to ask if we should say “We thank our anonymous reviewers at PLDI 2018, ICFP 2018, POPL 2019, PLDI 2019, and ICFP 2019.” Could send a nice inspirational message. :slightly_smiling_face:

You did - successfully too! thanks.

I didn’t know that (invariance)! Thanks for the additional information, that helps. I’ll try to twiddle a bit to see if I can come up with something satisfactory.

I’m game to say that.

ha

Where are we for artifact registration?

Hard to interpret that ‘ha’.

do we have a paper draft/artifact description that we can submit?

What’s wrong with submitting the current HEAD?

I don’t know of anything, so I’m asking

(currently the acknowledgements flows onto page 28)

note that for that actual final copy we will have less space

copyright block etc

I just tried uncommenting TeX source line 4 and didn’t see any change on where the first page ends.

Saving a few lines should be quite easy, so I certainly wouldn’t be too concerned.

@ccshan you have to make a few more changes to get the actual effect

The question is do we have numbers from the latest experiments (re-running with the same version of Hakaru) that substantiate our evaluation conclusions

“relationships [Carette and Shan 2016]” ends up on p 28

I’ll do the space hacking later. For now the HEAD (https://github.iu.edu/ccshan/ppaml/commit/70243badcb381c52f1decf53e3f80c39b63b8435) should be ok to submit for artifact evaluation — except what about re-running the experiments with the same version of Hakaru?

I reran all the experiments and checked them in hakaru-benchmarks

I can add them to the paper when I find wifi

I submitted the paper and a short description. We can update until 8 AM tomorrow

So is there something I should ‘test’ to help with reproducibility? [I know @samth has done that himself already, but that’s the kind of thing that is good to have multiple testers]. In particular, what platforms have already been tested?

@samth are we doing a docker image or a vm image? If vm image I will upload it today and share it for testing. I have already tested gmm and partial nb it just needs time to run the rest.

I think we should do the VM