I’ll be there. I just pushed some questions that Rajan should answer before the new text gets revised for organization and grammar.
@ccshan There were a bunch of comments from the PLDI reviewers on ‘our’ sections. We should coordinate on how to address them. [I have a little bit of time today, lots tomorrow and Friday]. In particular, although the principal evaluation part of the paper covers efficiency/accuracy, the reviewers did ask about applicability of our simplifications, which could indeed be ‘evaluated’ — if we wished. We do have a test suite for all of this code, we might be able to leverage that?
two more things to address:
- how to measure per-model startup time on JAGS (maybe print timings at various points in the script?)
- how to measure simplification time on the same machine as everything else (needs maple license?)
Let me annotate the PLDI reviews with suggestions for what you can do. I’m on that now.
Done. Please let me know if you’d like more coordination.
Meetings until 2:30, then I’ll look.
@rjnw I’m converting the listings to the math-ish notation, so please leave them alone for the next 30 minutes
oh okay, I just moved the text around it while writing.
Yeah that is all good
for the listings there is a full gmmGibbs program before after ANF step in the pipeline folder.
Are you saying you want to change the listings in the paper?
no if you need more information about those listings
Please digest that information for me so as to answer the question: “Where’s this running example from? …”
That is a small part in the full gmmGibbs program
and the full gmmGibbs is also in the repo along with the tex file
I find the use of (index z docUpdate) in a sum very suspicious. Is that really part of the full gmmGibbs program?
I kept the summate product and histogram and removed the internals
that I might have changed after
I started with gmmGibbs but trying to make it smaller I made some changes
What is the information that the reader needs to know to follow your argument in support of the punch line?
The location of histogram, prod, sum and the free variables inside them is closely related to the actual gmmGibbs program
and the transformations we show are also what happen when we are compiling it
Please put this information in the manuscript. Please disambiguate pronouns such as “them” and “it”. What close relation do you have in mind?
are you keeping the content in the listing same and just changing them to math-ish notation?
I will add that information based on the current example
I am keeping the content in the listing same, except I’m renaming variables, and also in Listing 4 I’m swapping 9 and 1000 to make it more realistic (I think you switched them)
I might have, may be a good idea to use 50 10000 as we use that in our evaluation
Please do that consistently throughout the section after I commit the listings change.
I also find (index t (index z i)) suspicious. Is that really what you want? Does it originate correctly from (\i1 -> \i2 -> (index t i2))? Is the latter correct?
yes but I probably used t and z in place of something else
What’s the something else?
okay no this one is exactly from gmmGibbs
There it seems that the “same” i66 is used to index into both z46 and t47. That’s what I expect. But how does that become (index t (index z i))?
yeah they both use the same index i I miss placed the i as (index z i)
sorry
ok
ok, I pushed the changed listing notation
Nice! Now it’s time to work on something else :slightly_smiling_face:
(For example, your revision of Section 5.0 is a definite improvement.)
(It’s better if you remove remarks as you address them)
@rjnw I just found the Maple Linux installer on a work account that I misplaced root access to, and it’s a 1.2GB file, so maybe the easiest way for you to get Maple is for me to give you a USB drive tomorrow?
What time can I come and pick it up?
rkt mean over 10 trials, jags only one trial, point is after every sweep. jags sweep is 500 seconds
startup removed
which is 25000 seconds for jags