What does “Sham language constructs build syntax objects instead of direct language expressions” mean? Can someone give me an example of a syntax object that’s not a direct language expression and/or a direct language expression that’s not a syntax object?

I meant to say sham constructs build syntax objects of expressions rather than them being language expressions

Is there a difference between “syntax objects of expressions” and “language expressions”? Or is the difference between “build” and “be”?

build and be

@rjnw You may already have the per-run timings? I have the code for computing std err

for synth?

I used racket time to compute them so it’s hard to get std err, I am going to try changing the code right now to get more precision

I also commited compile times for naive bayes

https://github.com/rjnw/papers/commit/29a83e95363dfa89786448ca339982c7ec185f5c pre run timings for synth over 100 trials

Section 5.4 still talks as if the Synth performance evaluation measures 3 systems, but it only shows 2 (Pure Racket (“improved”?) and Sham)

I see that Sham run-time code generation takes 400–500ms per Naive Bayes benchmark run. (Interestingly, it takes a bit longer when runtime specialization is ablated.) Given that the generated code is then run at least 2000 times, it would be misleading to just add these timings to the bar chart. So I’m just going to add the numbers to the text instead.

I cited the UAI paper for Hakaru, does that seem sufficient?

Is it just to cite Hakaru in general? The FLOPS paper is probably more appropriate. Just push now and let me fix it?

I’ll fix

Just let me know if you need the bibtex or anything

i get all my bibtex from dblp

done

i also rearranged a figure a bit

I resubmitted

I think I’m genuinely not conflicted with Ron so don’t worry about that warning

ok, sorry I might have changed some the conflict checkboxes when I submitted earlier

no problem

I think I’m happy with the current status

well done everyone