Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.theory > #35914 > unrolled thread

revised relativization theorem: P=NP for small N, P!=NP for large N

Back to article view | Back to comp.theory

Contents

  revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-07 18:19 -0700
    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-08 02:44 -0700
      Re: revised relativization theorem: P=NP for small N, P!=NP for large N DV <xlt.pjw@gmail.com> - 2021-07-08 05:37 -0700
        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Jeff Barnett <jbb@notatt.com> - 2021-07-08 14:21 -0600
          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-14 07:28 -0700
            Re: revised relativization theorem: P=NP for small N, P!=NP for large N DV <xlt.pjw@gmail.com> - 2021-07-14 14:28 -0700
              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-15 12:51 -0700
                Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-18 03:45 -0700
                  Re: revised relativization theorem: P=NP for small N, P!=NP for large N DV <xlt.pjw@gmail.com> - 2021-07-18 06:48 -0700
                    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-18 14:28 -0700
                      Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-26 04:45 -0700
                        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-26 14:35 +0100
                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 00:35 -0700
                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 02:40 -0700
                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-28 17:27 +0100
                              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 11:50 -0700
                                Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-28 20:52 +0100
                                  Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 15:53 -0700
                                    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 16:58 -0700
                                      Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 17:41 -0700
                                        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 01:45 +0100
                                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 20:07 -0700
                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-28 20:27 -0700
                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 14:59 +0100
                                        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Malcolm McLean <malcolm.arthur.mclean@gmail.com> - 2021-07-29 02:27 -0700
                                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 06:24 -0700
                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Malcolm McLean <malcolm.arthur.mclean@gmail.com> - 2021-07-29 07:50 -0700
                                              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Jeff Barnett <jbb@notatt.com> - 2021-07-29 11:44 -0600
                                                Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 11:59 -0700
                                                  Re: revised relativization theorem: P=NP for small N, P!=NP for large N Jeff Barnett <jbb@notatt.com> - 2021-07-29 13:23 -0600
                                              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 20:04 +0100
                                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 15:56 +0100
                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 11:53 -0700
                                              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 20:12 +0100
                                                Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 13:12 -0700
                                                  Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-29 23:18 +0100
                                                    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 15:36 -0700
                                                      Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 16:38 -0700
                                                        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 00:59 +0100
                                                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 17:05 -0700
                                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 01:19 +0100
                                                              Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 17:27 -0700
                                                                Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 02:22 +0100
                                                                  Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-29 19:16 -0700
                                                                    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Jeff Barnett <jbb@notatt.com> - 2021-07-29 22:25 -0600
                                                                      Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-30 00:35 -0700
                                                                    Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 10:53 +0100
                                                                      Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-30 05:19 -0700
                                                                        Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 14:22 +0100
                                                                          Re: revised relativization theorem: P=NP for small N, P!=NP for large N Daniel Pehoushek <pehoushek1@gmail.com> - 2021-07-30 07:16 -0700
                                                                            Re: revised relativization theorem: P=NP for small N, P!=NP for large N Ben Bacarisse <ben.usenet@bsb.me.uk> - 2021-07-30 15:22 +0100

Page 2 of 3 — ← Prev page 1 [2] 3  Next page →

#37235

Daniel Pehoushek <pehoushek1@gmail.com> writes:

> ben, is this clear?
>
> in track 1 the correctness track
> approximation is senseless is what i am saying
>
> in track 2 the incorrectness track 
> approximation is fine

These statements are clear, yes.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37239

how about this version? is holy grail over the top?
i can well define what i mean by "all universal truths" of p...

in track 1 the correctness track
the correct number of models is called #P of the formula
the are N bits of truth for every single model 
the correct program guarantees a 
truth structure with size N*#P bits 
from which universal truths of p 
may be logically derived 
programs in track 1 
never return any 
wrong answers

bob is called an Oracle for the number of models
correct track 1 programs are oracles 
guessing an answer is outside of their power
using all models bob computes and prints 
all and only universal truths of small forms
so in addition to being an oracle of model counting 
bob is an oracle of universal truths of small forms
bob is an oracle of universal truth
bob is the holy grail

approximation of #P 
is senseless is 
what i am 
saying

in track 2 the incorrectness track 
approximation is fine

On Wednesday, July 28, 2021 at 8:45:08 PM UTC-4, Ben Bacarisse wrote:
> Daniel Pehoushek <pehou...@gmail.com> writes: 
> 
> > ben, is this clear? 
> > 
> > in track 1 the correctness track 
> > approximation is senseless is what i am saying 
> > 
> > in track 2 the incorrectness track 
> > approximation is fine
> These statements are clear, yes. 
> 
> -- 
> Ben.

[toc] | [prev] | [next] | [standalone]

#37241

universal truths
here is a start on good definition

given p bob returns q 
all and only the propositional universal truths of p
universal truths have some very pleasant properties 
the universal truths of a set of universal truths is 
the same set of universal truths 
every valid quantification of p satisfies every universal truth
every invalid qbf falsifies at least one universal truth 
the time to evaluate a quantification is linear in q 
one clause of q is a single universal truth 
naming a set of variables that must 
have at least one existentially quantified variable 
bob minimizes the list size through subsumption 
as p gets larger so does q 

On Wednesday, July 28, 2021 at 11:07:08 PM UTC-4, Daniel Pehoushek wrote:
> how about this version? is holy grail over the top? 
> i can well define what i mean by "all universal truths" of p...
> in track 1 the correctness track
> the correct number of models is called #P of the formula 
> the are N bits of truth for every single model 
> the correct program guarantees a 
> truth structure with size N*#P bits 
> from which universal truths of p 
> may be logically derived 
> programs in track 1 
> never return any 
> wrong answers 
> 
> bob is called an Oracle for the number of models 
> correct track 1 programs are oracles 
> guessing an answer is outside of their power 
> using all models bob computes and prints 
> all and only universal truths of small forms 
> so in addition to being an oracle of model counting 
> bob is an oracle of universal truths of small forms 
> bob is an oracle of universal truth 
> bob is the holy grail 
> 
> approximation of #P
> is senseless is 
> what i am 
> saying 
> 
> in track 2 the incorrectness track 
> approximation is fine
> On Wednesday, July 28, 2021 at 8:45:08 PM UTC-4, Ben Bacarisse wrote: 
> > Daniel Pehoushek <pehou...@gmail.com> writes: 
> > 
> > > ben, is this clear? 
> > > 
> > > in track 1 the correctness track 
> > > approximation is senseless is what i am saying 
> > > 
> > > in track 2 the incorrectness track 
> > > approximation is fine 
> > These statements are clear, yes. 
> > 
> > -- 
> > Ben.

[toc] | [prev] | [next] | [standalone]

#37259

Daniel Pehoushek <pehoushek1@gmail.com> writes:

> how about this version?

No, this version is not clear (to me).  By the way, the exchange would
be clearer if you did not top post.

> is holy grail over the top?
> i can well define what i mean by "all universal truths" of p...
>
> in track 1 the correctness track
> the correct number of models is called #P of the formula
> the are N bits of truth for every single model 
> the correct program guarantees a 
> truth structure with size N*#P bits 
> from which universal truths of p 
> may be logically derived 
> programs in track 1 
> never return any 
> wrong answers
>
> bob is called an Oracle for the number of models
> correct track 1 programs are oracles 
> guessing an answer is outside of their power
> using all models bob computes and prints 
> all and only universal truths of small forms
> so in addition to being an oracle of model counting 
> bob is an oracle of universal truths of small forms
> bob is an oracle of universal truth
> bob is the holy grail
>
> approximation of #P 
> is senseless is 
> what i am 
> saying
>
> in track 2 the incorrectness track 
> approximation is fine
>
> On Wednesday, July 28, 2021 at 8:45:08 PM UTC-4, Ben Bacarisse wrote:
>> Daniel Pehoushek <pehou...@gmail.com> writes: 
>> 
>> > ben, is this clear? 
>> > 
>> > in track 1 the correctness track 
>> > approximation is senseless is what i am saying 
>> > 
>> > in track 2 the incorrectness track 
>> > approximation is fine
>> These statements are clear, yes. 

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37257

On Thursday, 29 July 2021 at 01:41:29 UTC+1, pehou...@gmail.com wrote:
> ben, is this clear? 
> 
> in track 1 the correctness track 
> approximation is senseless is what i am saying 
> 
> in track 2 the incorrectness track 
> approximation is fine
>
From what I understand, bob creates an exact solution to the problem
whilst the competition specifies an order of magnitude estimate.

It would be astonishing if bob also ran in similar time or was capable of
tackling the same size of problem, whilst producing an exact answer whilst
the competitors produced only approximations. So either this is the wrong
competition for bob, or there should be a separate track for exact solvers.
Or, this being computer science and not the Olympics, you can say that
you entered bob without any expectation that it would do well, to demonstrate
an exact solver. I'd suggest maybe the third option is the most dignified one.

[toc] | [prev] | [next] | [standalone]

#37258

they had a track 1 for correct solvers but 
they accepted boneheaded approzimations 
as better than "no answer in allotted time".
now they refuse to respond to email 
without any details of how bob did 
for three weeks of my trying

+ whilst the competition specifies an order of magnitude estimate.

little better than gibberish when compared to correct answers

+It would be astonishing if bob also ran in similar time or was capable of
+tackling the same size of problem, whilst producing an exact answer whilst

bob is very efficient on moderate sizes 
i have never ever considered studying approzimation algorithms for #P
the value of the number #P is in the accessibility of all models 
guessing the value of #P with approzimation is like saying 
about the number of leafs on all trees there are perhaps ten

+competition for bob, or there should be a separate track for exact solvers.

track 1 is the correctness track
i call track 2 the incorrectness track

+Or, this being computer science and not the Olympics, you can say that
+you entered bob without any expectation that it would do well, to demonstrate
+an exact solver. I'd suggest maybe the third option is the most dignified one.

well bob and i have moved beyond models to universal propositional truths


On Thursday, July 29, 2021 at 5:27:13 AM UTC-4, malcolm.ar...@gmail.com wrote:
> On Thursday, 29 July 2021 at 01:41:29 UTC+1, pehou...@gmail.com wrote: 
> > ben, is this clear? 
> > 
> > in track 1 the correctness track 
> > approximation is senseless is what i am saying 
> > 
> > in track 2 the incorrectness track 
> > approximation is fine 
> >
> From what I understand, bob creates an exact solution to the problem 
> whilst the competition specifies an order of magnitude estimate. 
> 
> It would be astonishing if bob also ran in similar time or was capable of 
> tackling the same size of problem, whilst producing an exact answer whilst 
> the competitors produced only approximations. So either this is the wrong 
> competition for bob, or there should be a separate track for exact solvers. 
> Or, this being computer science and not the Olympics, you can say that 
> you entered bob without any expectation that it would do well, to demonstrate 
> an exact solver. I'd suggest maybe the third option is the most dignified one.

[toc] | [prev] | [next] | [standalone]

#37260

On Thursday, 29 July 2021 at 14:24:10 UTC+1, pehou...@gmail.com wrote:
> they had a track 1 for correct solvers but 
> they accepted boneheaded approzimations 
> as better than "no answer in allotted time". 
> now they refuse to respond to email 
> without any details of how bob did 
> for three weeks of my trying 
> 
It seems you've got a legitimate complaint. If track 1 is for exact solvers,
but if the program runs out of time it is expected to submit an approximate
result, which bob doesn't do, it would seem to be rather unfair. However
the organisers may have their reasons.
>
> + whilst the competition specifies an order of magnitude estimate. 
> 
> little better than gibberish when compared to correct answers 
> 
That sort of comment tends to serve to rub people up the wrong way, and
won't do you any favours if you want to enter the next competition.
No organiser of a competition wants an entrant who refuses to accept 
the result. 

[toc] | [prev] | [next] | [standalone]

#37266

On 7/29/2021 8:50 AM, Malcolm McLean wrote:
> On Thursday, 29 July 2021 at 14:24:10 UTC+1, pehou...@gmail.com wrote:
>> they had a track 1 for correct solvers but
>> they accepted boneheaded approzimations
>> as better than "no answer in allotted time".
>> now they refuse to respond to email
>> without any details of how bob did
>> for three weeks of my trying
>>
> It seems you've got a legitimate complaint. If track 1 is for exact solvers,
> but if the program runs out of time it is expected to submit an approximate
> result, which bob doesn't do, it would seem to be rather unfair. However
> the organisers may have their reasons.

In real life problems, there is often a flavor of "give me your best 
answer by such and such time". In fact there are many military 
optimization problems that effect routing, safest path and timing to 
accomplish multiple tasks, for examples. There are many algorithms with 
names such as iterative deepening. These are not counting solutions (a 
la #p), its actually finding one or more solutions. If in battle or 
emergency, you have finite time to calculate something and hope it's 
better than nothing. Another area where real time pressures indicate a 
need for good-quick rather than optimal solutions is stock trading: 
given what I know, should I buy this stock now* at this price? Where 
"now" means in the next few seconds.

>> + whilst the competition specifies an order of magnitude estimate.
>>
>> little better than gibberish when compared to correct answers
>>
> That sort of comment tends to serve to rub people up the wrong way, and
> won't do you any favours if you want to enter the next competition.
> No organiser of a competition wants an entrant who refuses to accept
> the result.

Or the conditions of contest. He might also contact his "colleagues" in 
this contest and find out what problems they are actually working on 
where they got their grant money and for what.
-- 
Jeff Barnett

[toc] | [prev] | [next] | [standalone]

#37273

+Or the conditions of contest. He might also contact his "colleagues" in
+this contest and find out what problems they are actually working on

i have tried for years to engage with their community.
i have almost entirely the "null response".
i am unable to get a response.
i have told them my gre number is 2380, 
but they have probably inverted the validity of 2380 
and then concluded that i am mad instead of reasonable.

On Thursday, July 29, 2021 at 1:45:01 PM UTC-4, Jeff Barnett wrote:
> On 7/29/2021 8:50 AM, Malcolm McLean wrote: 
> > On Thursday, 29 July 2021 at 14:24:10 UTC+1, pehou...@gmail.com wrote: 
> >> they had a track 1 for correct solvers but 
> >> they accepted boneheaded approzimations 
> >> as better than "no answer in allotted time". 
> >> now they refuse to respond to email 
> >> without any details of how bob did 
> >> for three weeks of my trying 
> >> 
> > It seems you've got a legitimate complaint. If track 1 is for exact solvers, 
> > but if the program runs out of time it is expected to submit an approximate 
> > result, which bob doesn't do, it would seem to be rather unfair. However 
> > the organisers may have their reasons.
> In real life problems, there is often a flavor of "give me your best 
> answer by such and such time". In fact there are many military 
> optimization problems that effect routing, safest path and timing to 
> accomplish multiple tasks, for examples. There are many algorithms with 
> names such as iterative deepening. These are not counting solutions (a 
> la #p), its actually finding one or more solutions. If in battle or 
> emergency, you have finite time to calculate something and hope it's 
> better than nothing. Another area where real time pressures indicate a 
> need for good-quick rather than optimal solutions is stock trading: 
> given what I know, should I buy this stock now* at this price? Where 
> "now" means in the next few seconds.
> >> + whilst the competition specifies an order of magnitude estimate. 
> >> 
> >> little better than gibberish when compared to correct answers 
> >> 
> > That sort of comment tends to serve to rub people up the wrong way, and 
> > won't do you any favours if you want to enter the next competition. 
> > No organiser of a competition wants an entrant who refuses to accept 
> > the result.
> Or the conditions of contest. He might also contact his "colleagues" in 
> this contest and find out what problems they are actually working on 
> where they got their grant money and for what. 
> -- 
> Jeff Barnett

[toc] | [prev] | [next] | [standalone]

#37277

On 7/29/2021 12:59 PM, Daniel Pehoushek wrote:
> +Or the conditions of contest. He might also contact his "colleagues" in
> +this contest and find out what problems they are actually working on
> 
> i have tried for years to engage with their community.
> i have almost entirely the "null response".
> i am unable to get a response.
> i have told them my gre number is 2380,
> but they have probably inverted the validity of 2380
> and then concluded that i am mad instead of reasonable.

I don't think that is what's going on. First I can't imagine anyone 
caring about your GRE other than to decide whether to take you on as a 
student. They might also object to your in-your-face, hard to read or 
scan email structure; it just takes a bunch of extra time and is about 
as reader friendly as chicken scratches. The trade off is in part up to 
you: maintain your iconoclastic approach to communications, etc., or 
seek collegial contacts in a way that doesn't cause so much stress. If 
you always start off telling folks how much better you are then they 
are, you have created a rivalry and they have no incentive to play. Go 
slow. You'll probably do better.
-- 
Jeff Barnett

[toc] | [prev] | [next] | [standalone]

#37274

Malcolm McLean <malcolm.arthur.mclean@gmail.com> writes:

> On Thursday, 29 July 2021 at 14:24:10 UTC+1, pehou...@gmail.com wrote:
>> they had a track 1 for correct solvers but 
>> they accepted boneheaded approzimations 
>> as better than "no answer in allotted time". 
>> now they refuse to respond to email 
>> without any details of how bob did 
>> for three weeks of my trying 
>> 
> It seems you've got a legitimate complaint. If track 1 is for exact solvers,
> but if the program runs out of time it is expected to submit an approximate
> result, which bob doesn't do, it would seem to be rather unfair. However
> the organisers may have their reasons.

It's a bit odd.  According to the explanatory documents, the score is
simply the number of acceptable solutions over all the competition
instances, but an acceptable solution can be approximate.  Thus it would
seem that a solver that gets 10 problems exactly right won't score
better than one that gets 10 problems approximately right, and will
actually loose to a solver that get not exactly right but gets 11
approximations in the right range.

Unless this is not the scoring used for all tracks, I'm not sure how
track 1 can be described as "for exact solvers only".  Maybe the
permission for a solution to be within some error bound is not applied
for track 1 solvers.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37261

Malcolm McLean <malcolm.arthur.mclean@gmail.com> writes:

> On Thursday, 29 July 2021 at 01:41:29 UTC+1, pehou...@gmail.com wrote:
>> ben, is this clear? 
>> 
>> in track 1 the correctness track 
>> approximation is senseless is what i am saying 
>> 
>> in track 2 the incorrectness track 
>> approximation is fine
>>
> From what I understand, bob creates an exact solution to the problem
> whilst the competition specifies an order of magnitude estimate.

I've gone to have a look...  The competition has three tracks.  Track
one is for exact solvers and the other two allow either exact or
approximate solvers but exact solvers receive what they call a
"performance bias".

All solvers must output the base 10 log of the result, but the format
allows for an exact count as well.

> It would be astonishing if bob also ran in similar time or was capable of
> tackling the same size of problem, whilst producing an exact answer whilst
> the competitors produced only approximations. So either this is the wrong
> competition for bob, or there should be a separate track for exact solvers.
> Or, this being computer science and not the Olympics, you can say that
> you entered bob without any expectation that it would do well, to demonstrate
> an exact solver. I'd suggest maybe the third option is the most dignified one.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37272

+All solvers must output the base 10 log of the result, but the format
+allows for an exact count as well.

the log10 format saves a few bytes on a two day computation 
where accuracy of the answer is paramount.  they allow printing out 
the precise answer but the competition fails to use it.

they did that because they have a clique of researchers 
that do not give precise answers to the precise 
question of model counts.  they have the impression 
that making a wild guess at the value of a deeply 
meaningful number may also be deeply meaningful. 
i do not share their impression.  the deep meaning 
of the correct model count is related to the models,
not so much the precise value of the model count. 

i am trying to argue with them, and trying to 
get information on how well bob did and 
who won track 1.  of course i care little about 
track 2 and track 3.

On Thursday, July 29, 2021 at 10:56:57 AM UTC-4, Ben Bacarisse wrote:
> Malcolm McLean <malcolm.ar...@gmail.com> writes: 
> 
> > On Thursday, 29 July 2021 at 01:41:29 UTC+1, pehou...@gmail.com wrote: 
> >> ben, is this clear? 
> >> 
> >> in track 1 the correctness track 
> >> approximation is senseless is what i am saying 
> >> 
> >> in track 2 the incorrectness track 
> >> approximation is fine 
> >> 
> > From what I understand, bob creates an exact solution to the problem 
> > whilst the competition specifies an order of magnitude estimate.
> I've gone to have a look... The competition has three tracks. Track 
> one is for exact solvers and the other two allow either exact or 
> approximate solvers but exact solvers receive what they call a 
> "performance bias". 
> 
> All solvers must output the base 10 log of the result, but the format 
> allows for an exact count as well.
> > It would be astonishing if bob also ran in similar time or was capable of 
> > tackling the same size of problem, whilst producing an exact answer whilst 
> > the competitors produced only approximations. So either this is the wrong 
> > competition for bob, or there should be a separate track for exact solvers. 
> > Or, this being computer science and not the Olympics, you can say that 
> > you entered bob without any expectation that it would do well, to demonstrate 
> > an exact solver. I'd suggest maybe the third option is the most dignified one.
> -- 
> Ben.

[toc] | [prev] | [next] | [standalone]

#37275

Daniel Pehoushek <pehoushek1@gmail.com> writes:

> +All solvers must output the base 10 log of the result, but the format
> +allows for an exact count as well.
>
> the log10 format saves a few bytes on a two day computation 
> where accuracy of the answer is paramount.  they allow printing out 
> the precise answer but the competition fails to use it.

So it would appear.  I've just read the scoring system used, add all
"accepted" solutions count.  A solution is acceptable if it is within
some relative error bound of the expected result.

> they did that because they have a clique of researchers 
> that do not give precise answers to the precise 
> question of model counts.  they have the impression 
> that making a wild guess at the value of a deeply 
> meaningful number may also be deeply meaningful.

Wild guesses won't win.  The error bound is not wide enough for that.

I wonder how these approximate solvers work.  Maybe they have a way to
count all the "low hanging fruit" first, and then expect the other
solutions not yet found will not put the result out of bounds.

> i am trying to argue with them, and trying to 
> get information on how well bob did and 
> who won track 1.

I think you need to give them time.  It can't be easy in these troubled
times to collate all of the data.

But arguing for different criteria when the scoring was publish before
the competition was run just won't work.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37285

On Thursday, July 29, 2021 at 3:12:36 PM UTC-4, Ben Bacarisse wrote:
> Daniel Pehoushek <pehou...@gmail.com> writes: 
> 
> > +All solvers must output the base 10 log of the result, but the format 
> > +allows for an exact count as well. 
> > 
> > the log10 format saves a few bytes on a two day computation 
> > where accuracy of the answer is paramount. they allow printing out 
> > the precise answer but the competition fails to use it.
> So it would appear. I've just read the scoring system used, add all 
> "accepted" solutions count. A solution is acceptable if it is within 
> some relative error bound of the expected result.
> > they did that because they have a clique of researchers 
> > that do not give precise answers to the precise 
> > question of model counts. they have the impression 
> > that making a wild guess at the value of a deeply 
> > meaningful number may also be deeply meaningful.
> Wild guesses won't win. The error bound is not wide enough for that. 
> 
> I wonder how these approximate solvers work. Maybe they have a way to 
> count all the "low hanging fruit" first, and then expect the other 
> solutions not yet found will not put the result out of bounds.
> > i am trying to argue with them, and trying to 
> > get information on how well bob did and 
> > who won track 1.
> I think you need to give them time. It can't be easy in these troubled 
> times to collate all of the data. 
> 
> But arguing for different criteria when the scoring was publish before 
> the competition was run just won't work. 
> 
> -- 
> Ben.

if the problem has a validateable answer bob gets it.
bob does one trillion inferences per hour.

the "no answer yet" after alotted time on track 1 could 
"legally" be considered better than an approximate answer.
but i do not think they do that, else bob would be winning 
and they would tell me so.

they have told me nothing after three weeks of trying 
since the end of the conference.  no reports. nothing.
null responses only. like a blacklist.
daniel

[toc] | [prev] | [next] | [standalone]

#37291

Daniel Pehoushek <pehoushek1@gmail.com> writes:

> On Thursday, July 29, 2021 at 3:12:36 PM UTC-4, Ben Bacarisse wrote:
>> Daniel Pehoushek <pehou...@gmail.com> writes: 
>> 
>> > +All solvers must output the base 10 log of the result, but the format 
>> > +allows for an exact count as well. 
>> > 
>> > the log10 format saves a few bytes on a two day computation 
>> > where accuracy of the answer is paramount. they allow printing out 
>> > the precise answer but the competition fails to use it.
>> So it would appear. I've just read the scoring system used, add all 
>> "accepted" solutions count. A solution is acceptable if it is within 
>> some relative error bound of the expected result.
>> > they did that because they have a clique of researchers 
>> > that do not give precise answers to the precise 
>> > question of model counts. they have the impression 
>> > that making a wild guess at the value of a deeply 
>> > meaningful number may also be deeply meaningful.
>> Wild guesses won't win. The error bound is not wide enough for that. 
>> 
>> I wonder how these approximate solvers work. Maybe they have a way to 
>> count all the "low hanging fruit" first, and then expect the other 
>> solutions not yet found will not put the result out of bounds.
>> > i am trying to argue with them, and trying to 
>> > get information on how well bob did and 
>> > who won track 1.
>> I think you need to give them time. It can't be easy in these troubled 
>> times to collate all of the data. 
>> 
>> But arguing for different criteria when the scoring was publish before 
>> the competition was run just won't work. 
>> 
>
> if the problem has a validateable answer bob gets it.
> bob does one trillion inferences per hour.

That's not a helpful number because an inference might lead to large
or a small change in the count.

> the "no answer yet" after alotted time on track 1 could 
> "legally" be considered better than an approximate answer.
> but i do not think they do that, else bob would be winning 
> and they would tell me so.

The scoring is explicitly defined in the competition's documents.  I
don't think "no answer yet" counts for anything.

> they have told me nothing after three weeks of trying 
> since the end of the conference.  no reports. nothing.
> null responses only. like a blacklist.

They don't do this for money.  They are all volunteers.  They all
probably have full-time jobs and families and might even hope to have
some free time for relaxation.

But if you got someone who you know has good writing skill to review you
communications with the competition you are more likely to get something
back.  Your posts here would stop me from replying at all if I were
involved with the conference.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37292

+The scoring is explicitly defined in the competition's documents. I
+don't think "no answer yet" counts for anything.

that is precisely the fundamental problem.

my totally correct program for model counts takes 
more than allotted time on all of the very large benchmarks
and thus returns no answer.
but when bob does answer the answer is right, 
implying deterministic access to a vast truth structure.

approximations get credit for an answer, 
even though, relative to correct answers,
the approximate answer is as good as gibberish.

bob is an oracle for #P.
all valid track 1 programs are oracles.
its a competition to detect oracles.
but they are failing to detect bob as totally correct.
daniel

On Thursday, July 29, 2021 at 6:18:25 PM UTC-4, Ben Bacarisse wrote:
> Daniel Pehoushek <pehou...@gmail.com> writes: 
> 
> > On Thursday, July 29, 2021 at 3:12:36 PM UTC-4, Ben Bacarisse wrote: 
> >> Daniel Pehoushek <pehou...@gmail.com> writes: 
> >> 
> >> > +All solvers must output the base 10 log of the result, but the format 
> >> > +allows for an exact count as well. 
> >> > 
> >> > the log10 format saves a few bytes on a two day computation 
> >> > where accuracy of the answer is paramount. they allow printing out 
> >> > the precise answer but the competition fails to use it. 
> >> So it would appear. I've just read the scoring system used, add all 
> >> "accepted" solutions count. A solution is acceptable if it is within 
> >> some relative error bound of the expected result. 
> >> > they did that because they have a clique of researchers 
> >> > that do not give precise answers to the precise 
> >> > question of model counts. they have the impression 
> >> > that making a wild guess at the value of a deeply 
> >> > meaningful number may also be deeply meaningful. 
> >> Wild guesses won't win. The error bound is not wide enough for that. 
> >> 
> >> I wonder how these approximate solvers work. Maybe they have a way to 
> >> count all the "low hanging fruit" first, and then expect the other 
> >> solutions not yet found will not put the result out of bounds. 
> >> > i am trying to argue with them, and trying to 
> >> > get information on how well bob did and 
> >> > who won track 1. 
> >> I think you need to give them time. It can't be easy in these troubled 
> >> times to collate all of the data. 
> >> 
> >> But arguing for different criteria when the scoring was publish before 
> >> the competition was run just won't work. 
> >> 
> >
> > if the problem has a validateable answer bob gets it. 
> > bob does one trillion inferences per hour.
> That's not a helpful number because an inference might lead to large 
> or a small change in the count.
> > the "no answer yet" after alotted time on track 1 could 
> > "legally" be considered better than an approximate answer. 
> > but i do not think they do that, else bob would be winning 
> > and they would tell me so.
> The scoring is explicitly defined in the competition's documents. I 
> don't think "no answer yet" counts for anything.
> > they have told me nothing after three weeks of trying 
> > since the end of the conference. no reports. nothing. 
> > null responses only. like a blacklist.
> They don't do this for money. They are all volunteers. They all 
> probably have full-time jobs and families and might even hope to have 
> some free time for relaxation. 
> 
> But if you got someone who you know has good writing skill to review you 
> communications with the competition you are more likely to get something 
> back. Your posts here would stop me from replying at all if I were 
> involved with the conference. 
> 
> -- 
> Ben.

[toc] | [prev] | [next] | [standalone]

#37299

i am pretty sure bob is the fastest totally correct program in the competition.
but they have for three weeks refused to tell me how bob did.
what can i do?
daniel

On Thursday, July 29, 2021 at 6:36:26 PM UTC-4, Daniel Pehoushek wrote:
> +The scoring is explicitly defined in the competition's documents. I 
> +don't think "no answer yet" counts for anything. 
> 
> that is precisely the fundamental problem. 
> 
> my totally correct program for model counts takes 
> more than allotted time on all of the very large benchmarks 
> and thus returns no answer. 
> but when bob does answer the answer is right, 
> implying deterministic access to a vast truth structure. 
> 
> approximations get credit for an answer, 
> even though, relative to correct answers, 
> the approximate answer is as good as gibberish. 
> 
> bob is an oracle for #P. 
> all valid track 1 programs are oracles. 
> its a competition to detect oracles. 
> but they are failing to detect bob as totally correct. 
> daniel
> On Thursday, July 29, 2021 at 6:18:25 PM UTC-4, Ben Bacarisse wrote: 
> > Daniel Pehoushek <pehou...@gmail.com> writes: 
> > 
> > > On Thursday, July 29, 2021 at 3:12:36 PM UTC-4, Ben Bacarisse wrote: 
> > >> Daniel Pehoushek <pehou...@gmail.com> writes: 
> > >> 
> > >> > +All solvers must output the base 10 log of the result, but the format 
> > >> > +allows for an exact count as well. 
> > >> > 
> > >> > the log10 format saves a few bytes on a two day computation 
> > >> > where accuracy of the answer is paramount. they allow printing out 
> > >> > the precise answer but the competition fails to use it. 
> > >> So it would appear. I've just read the scoring system used, add all 
> > >> "accepted" solutions count. A solution is acceptable if it is within 
> > >> some relative error bound of the expected result. 
> > >> > they did that because they have a clique of researchers 
> > >> > that do not give precise answers to the precise 
> > >> > question of model counts. they have the impression 
> > >> > that making a wild guess at the value of a deeply 
> > >> > meaningful number may also be deeply meaningful. 
> > >> Wild guesses won't win. The error bound is not wide enough for that. 
> > >> 
> > >> I wonder how these approximate solvers work. Maybe they have a way to 
> > >> count all the "low hanging fruit" first, and then expect the other 
> > >> solutions not yet found will not put the result out of bounds. 
> > >> > i am trying to argue with them, and trying to 
> > >> > get information on how well bob did and 
> > >> > who won track 1. 
> > >> I think you need to give them time. It can't be easy in these troubled 
> > >> times to collate all of the data. 
> > >> 
> > >> But arguing for different criteria when the scoring was publish before 
> > >> the competition was run just won't work. 
> > >> 
> > > 
> > > if the problem has a validateable answer bob gets it. 
> > > bob does one trillion inferences per hour. 
> > That's not a helpful number because an inference might lead to large 
> > or a small change in the count. 
> > > the "no answer yet" after alotted time on track 1 could 
> > > "legally" be considered better than an approximate answer. 
> > > but i do not think they do that, else bob would be winning 
> > > and they would tell me so. 
> > The scoring is explicitly defined in the competition's documents. I 
> > don't think "no answer yet" counts for anything. 
> > > they have told me nothing after three weeks of trying 
> > > since the end of the conference. no reports. nothing. 
> > > null responses only. like a blacklist. 
> > They don't do this for money. They are all volunteers. They all 
> > probably have full-time jobs and families and might even hope to have 
> > some free time for relaxation. 
> > 
> > But if you got someone who you know has good writing skill to review you 
> > communications with the competition you are more likely to get something 
> > back. Your posts here would stop me from replying at all if I were 
> > involved with the conference. 
> > 
> > -- 
> > Ben.

[toc] | [prev] | [next] | [standalone]

#37300

Daniel Pehoushek <pehoushek1@gmail.com> writes:

> i am pretty sure bob is the fastest totally correct program in the competition.
> but they have for three weeks refused to tell me how bob did.
> what can i do?

I would wait until the results come out.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#37301

On Thursday, July 29, 2021 at 7:59:34 PM UTC-4, Ben Bacarisse wrote:
> Daniel Pehoushek <pehou...@gmail.com> writes: 
> 
> > i am pretty sure bob is the fastest totally correct program in the competition. 
> > but they have for three weeks refused to tell me how bob did. 
> > what can i do?
> I would wait until the results come out. 
> 
> -- 
> Ben.

They were supposedly presented three weeks ago at the end of the conference.
that is all they have ever ever told me about how bob did.

[toc] | [prev] | [next] | [standalone]

Page 2 of 3 — ← Prev page 1 [2] 3  Next page →

Back to top | Article view | comp.theory

csiph-web