Before I view the serious response draft, I wrote one to let me say what I really wanted to say. Now that I’ve done that, it will be easier to be polite and restrained. However, there are some ideas in my response that I do hope can be used.

@ccshan can you please give access to to edit the response google doc?

@samth Do you have any means, beyond slack, to ping @ccshan? I’m unable to access the google doc, so I can’t participate in drafting the response.

you should now have access

:+1:

I’ve edited a couple of places to be more forceful in our language (without being rude/impolite/snarky).

The response, up to the first indented paragraph, is excellent. If we’re tight for space, it seems that the sections ‘startup speed’ and ‘Backend optimization benefits’ are overly wordy and could be cut down and still say the same thing.

But I’m not quite sure how to word-count only part of a document in Google Docs.

if people are happy with my unindented versions we can cut the indented ones

I’ve edited the “Startup speed” one so that I am now happy with the unindented versions.

Select it

I tried, didn’t see a word count. But I’ll try again.

Under the Tools menu

When do we expect to submit the response? It’s due 10am tomorrow, right?

Yes

I’m just final-editing

offline? [I’m not seeing live changes to the doc]

Ah, and now I see changes!

I’m done with final editing. I’m fine with the text above the line without the indented parts. Please review the shaded bits (1 yellow comment + 3 purple highlighting) to make sure they don’t need to be dealt with in the response.

Current word count is 726 - probably fine?

So we should just delete the indented parts? [or move them to below the table]

That can be done on the way to Markdown

adding a (such as XXX) after the yellow comment would really help. What’s our favourite?

I don’t know what @samth meant when he wrote that

That’s about the inability of augur to handle two at a time

Oh ok, let me revise

I kind of like the first purple comment; but I don’t have a strong feeling about include/exclude from the response.

2nd purple (Review B minor), I don’t think we really need to address that, given the length we already have.

“or the robust algebraic rewriting we perform”?

:+1:

I’ve read the last purple 3 times, and it is too cryptic/indirect for me to understand what it entails.

I mean I wouldn’t call our algebraic rewriting robust, but compared to AugurV2…

The last purple is just my asking y’all, should we give a pointer to the hakaru-benchmark repository?

It’s a response, not something meant for the paper :sunglasses:

Definitely didn’t get that! I certainly like that idea. But we should definitely hear @rjnw’s opinion on that.

Also, I always have trouble writing the “we thank all” sentence… The sentence “we thank all the reviewers for their helpful comments” kind of suggests there are un-helpful comments

I’m completely ok with omitting it.

Given the reviews and this response, I’m actually willing to be hopeful this time.

Reviewer C should champion this “very nice paper” with only one “small concern” already.

I agree, I don’t know why Reviewer C is so reluctant.

So are we waiting on @rjnw on that last purple thing before we submit?

I’d much rather get it done this evening.

I am fine giving a pointer to the repository

what example are we thinking? hakaru code for all our tests: https://github.com/rjnw/hakaru-benchmarks/tree/master/testcode/hksrc simplified code: https://github.com/rjnw/hakaru-benchmarks/blob/master/testcode/hksimp/

Point to the repo and let them find it?

We can choose which exact example we want to give. I gave link to the folder containing all of our benchmarks.

note that we are still blinded

Will a gist work?

yes, use an anonymous one

So what example should we give?

can someone remind me what they asked for?

Review A: “Minors: It is not clear how to use the proposed methods. It would be nice to include a code example of how to invoke the proposed methods, similar to examples in https://danehuang.github.io/papers/augurv2.pdf”

Review :anguished: “The ideas of the paper are quite straightforward. First, the idea of unproduct only applies to pointwise independent formula. In such cases, it seems that one can directly apply one-dimensional optimizations pointwise in the program syntax level. I do not clearly see the need for such a straightforward detour.”

Ok, then something like the classic sprinkler model seems like a good example, or something we actually mention in the paper

On a different note, I just re-read Reviewer D saying “the performance gain by 2000 times is mostly due to the combination of the histogram optimization and LICM”, as if the simplification pass has no effect. But actually the simplification pass is what makes it possible to execute the specified algorithms at all, inefficiently or efficiently. I’m going to edit the response to point this out, but first let me confirm with @rjnw, does the first row in Table 2 “No optimizations” mean that simplification was run but nothing else was?

Yes that is correct.

Thanks!

As for which code example to provide, if we don’t need to show off histogram or Gibbs, then maybe LinearRegression?

Please take a look at my new edit, in orange.

I was going to say “looks good”, but then you started re-editing… and now I can still say “looks good” !

I like the orange text and am happy with submitting now

No gist

?

Not sure if @rjnw is listening in if he’s not explicitly pinged…

https://github.blog/2018-02-18-deprecation-notice-removing-anonymous-gist-creation/

is there any other good service for making anonymous code snippet

It seems that we can just put verbatim code at the bottom of our response?

oh annoying

I would do what @ccshan suggests

There’s no enforced limit to the response, just guidelines?

I think it’s just guidelines painted in pink. So should I just write “Here’s our code for linear regression” above https://github.com/rjnw/hakaru-benchmarks/blob/master/testcode/hksrc/LinearRegression.hk and maybe also include https://github.com/rjnw/hakaru-benchmarks/blob/master/testcode/hksimp/LinearRegressionSimp.hk ?

I think just the former

No opinion from me.

so who’s submitting?

I don’t think it matters…

since http://hotcrp.com\|hotcrp.com is down?

That’s not where I was going!

Well, I need to go to bed, I’ve got an early morning tomorrow. But if it doesn’t get submitted tonight, I can do it around 7:30am.

I think the deadline is 10am

It is. But I’ll be offline from 8–11. So if it’s me, that’s when it will be!

I converted the Google Doc to Markdown: https://github.iu.edu/ccshan/ppaml/blob/master/writing/pipeline/pldi2019-response-346.mdwn