subject:"\[Computer\-go\] CNN with 54% prediction on KGS 6d\+ data"

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2016-01-10 Thread Josef Moudrik

Hi,

Winrate of your pure CNN againts pachi retsugen is:

GAMES WINRATE S.D.PAIRING
224   0.558   0.033   19-7.5-1-pachi-=1-detlef_54
221   0.407   0.033   19-7.5-1-pachi-=2-detlef_54

I used the
https://github.com/jmoudrik/deep-go-wrap
for the player.

Regards,
Josef

On Tue, Dec 29, 2015 at 10:24 AM Detlef Schmicker <d...@physik.de> wrote:

> -BEGIN PGP SIGNED MESSAGE-
> Hash: SHA1
>
> Hi,
>
> I am fighting with the problem most seem to have with the strong move
> predictions at the moment, MCTS is not increasing the players a lot :)
>
> I wonder, if somebody measured the performance of the pure CNN54
> against pachi 10k (or 100k), to get a comparison with the darkforest CNN.
>
> It is not too much work, but you probably did it already.
>
> Thanks,
>
> Detlef
>
> Am 21.12.2015 um 12:42 schrieb Hiroshi Yamashita:
> > Hi Detlef,
> >
> > Thank you for publishing your data and latest oakform code! It was
> > very helpful for me.
> >
> > I tried your 54% data with Aya.
> >
> > Aya with Detlef54% vs Aya with Detlef44%, 1 playout/move Aya
> > with Detlef54%'s winrate is 0.569 (124wins / 218games).
> >
> > CGOS BayseElo rating Aya with Detlef44%  (aya786n_Detlef_10k) 3040
> > Aya with Detlef54%  (Aya786m_Det54_10k ) 3036
> > http://www.yss-aya.com/cgos/19x19/bayes.html
> >
> > Detlef54% is a bit stronger in selfplay, but they are similar on
> > CGOS. Maybe Detlef54%'s prediction is strong, and Aya's playout
> > strength is not enough.
> >
> > Speed for a position on GTS 450. Detlef54%   21ms Detlef44%   17ms
> >
> > Cumulative accuracy from 1000 pro games.
> >
> > move rank  AyaDetlef54%  Mixture 1  40.8  47.6
> > 48.0 2  53.5  62.4 62.7 3  60.2  70.7 71.0
> > 4  64.8  75.8 76.1 5  68.1  79.5 79.9 6
> > 71.0  82.3 82.6 7  73.2  84.5 84.8 8  75.2
> > 86.3 86.6 9  76.9  87.8 88.1 10  78.3  89.0
> > 89.3 11  79.6  90.2 90.6 12  80.8  91.2
> > 91.4 13  81.9  92.0 92.2 14  82.9  92.7
> > 92.9 15  83.8  93.3 93.5 16  84.6  93.9
> > 94.1 17  85.4  94.3 94.5 18  86.1  94.8
> > 95.0 19  86.8  95.2 95.4 20  87.4  95.5
> > 95.7
> >
> > Mixture is pretty same as Detlef54%. I changed learning method from
> > MM to LFR. Aya's own accuracy is from LFR rank, not MM gamma. So
> > comparison is difficult.
> >
> > Cumulative accuracy Detlef44%
> > http://computer-go.org/pipermail/computer-go/2015-October/008031.html
> >
> >  Regards, Hiroshi Yamashita
> >
> >
> > - Original Message - From: "Detlef Schmicker"
> > <d...@physik.de> To: <computer-go@computer-go.org> Sent: Wednesday,
> > December 09, 2015 12:13 AM Subject: [Computer-go] CNN with 54%
> > prediction on KGS 6d+ data
> >
> >
> >> -BEGIN PGP SIGNED MESSAGE- Hash: SHA1
> >>
> >> Hi,
> >>
> >> as somebody ask I will offer my actual CNN for testing.
> >>
> >> It has 54% prediction on KGS 6d+ data (which I thought would be
> >> state of the art when I started training, but it is not
> >> anymore:).
> >>
> >> it has: 1 2 3
> >>> 4 libs playing color
> >> 1 2 3
> >>> 4 libs opponent color
> >> Empty points last move second last move third last move forth
> >> last move
> >>
> >> input layers, and it is fully convolutional, so with just editing
> >> the golast19.prototxt file you can use it for 13x13 as well, as I
> >> did on last sunday. It was used in November tournament as well.
> >>
> >> You can find it http://physik.de/CNNlast.tar.gz
> >>
> >>
> >>
> >> If you try here some points I like to get discussion:
> >>
> >> - - it seems to me, that the playouts get much more important
> >> with such a strong move prediction. Often the move prediction
> >> seems better the playouts (I use 8000 at the moment against pachi
> >> 32000 with about 70% winrate on 19x19, but with an extremely
> >> focused progressive widening (a=400, a=20 was usual).
> >>
> >> - - live and death becomes worse. My interpretation is, that the
> >> strong CNN does not play moves, which obviously do not help to
> >> get a group life, but would help the playouts to recognize the
> >> group is dead. (http://physik.de/example.sgf top black group was
&g

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-29 Thread Detlef Schmicker

-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1

Hi,

I am fighting with the problem most seem to have with the strong move
predictions at the moment, MCTS is not increasing the players a lot :)

I wonder, if somebody measured the performance of the pure CNN54
against pachi 10k (or 100k), to get a comparison with the darkforest CNN.

It is not too much work, but you probably did it already.

Thanks,

Detlef

Am 21.12.2015 um 12:42 schrieb Hiroshi Yamashita:
> Hi Detlef,
> 
> Thank you for publishing your data and latest oakform code! It was
> very helpful for me.
> 
> I tried your 54% data with Aya.
> 
> Aya with Detlef54% vs Aya with Detlef44%, 1 playout/move Aya
> with Detlef54%'s winrate is 0.569 (124wins / 218games).
> 
> CGOS BayseElo rating Aya with Detlef44%  (aya786n_Detlef_10k) 3040 
> Aya with Detlef54%  (Aya786m_Det54_10k ) 3036 
> http://www.yss-aya.com/cgos/19x19/bayes.html
> 
> Detlef54% is a bit stronger in selfplay, but they are similar on
> CGOS. Maybe Detlef54%'s prediction is strong, and Aya's playout
> strength is not enough.
> 
> Speed for a position on GTS 450. Detlef54%   21ms Detlef44%   17ms
> 
> Cumulative accuracy from 1000 pro games.
> 
> move rank  AyaDetlef54%  Mixture 1  40.8  47.6
> 48.0 2  53.5  62.4 62.7 3  60.2  70.7 71.0 
> 4  64.8  75.8 76.1 5  68.1  79.5 79.9 6
> 71.0  82.3 82.6 7  73.2  84.5 84.8 8  75.2
> 86.3 86.6 9  76.9  87.8 88.1 10  78.3  89.0
> 89.3 11  79.6  90.2 90.6 12  80.8  91.2
> 91.4 13  81.9  92.0 92.2 14  82.9  92.7
> 92.9 15  83.8  93.3 93.5 16  84.6  93.9
> 94.1 17  85.4  94.3 94.5 18  86.1  94.8
> 95.0 19  86.8  95.2 95.4 20  87.4  95.5
> 95.7
> 
> Mixture is pretty same as Detlef54%. I changed learning method from
> MM to LFR. Aya's own accuracy is from LFR rank, not MM gamma. So
> comparison is difficult.
> 
> Cumulative accuracy Detlef44% 
> http://computer-go.org/pipermail/computer-go/2015-October/008031.html
>
>  Regards, Hiroshi Yamashita
> 
> 
> - Original Message - From: "Detlef Schmicker"
> <d...@physik.de> To: <computer-go@computer-go.org> Sent: Wednesday,
> December 09, 2015 12:13 AM Subject: [Computer-go] CNN with 54%
> prediction on KGS 6d+ data
> 
> 
>> -BEGIN PGP SIGNED MESSAGE- Hash: SHA1
>> 
>> Hi,
>> 
>> as somebody ask I will offer my actual CNN for testing.
>> 
>> It has 54% prediction on KGS 6d+ data (which I thought would be
>> state of the art when I started training, but it is not
>> anymore:).
>> 
>> it has: 1 2 3
>>> 4 libs playing color
>> 1 2 3
>>> 4 libs opponent color
>> Empty points last move second last move third last move forth
>> last move
>> 
>> input layers, and it is fully convolutional, so with just editing
>> the golast19.prototxt file you can use it for 13x13 as well, as I
>> did on last sunday. It was used in November tournament as well.
>> 
>> You can find it http://physik.de/CNNlast.tar.gz
>> 
>> 
>> 
>> If you try here some points I like to get discussion:
>> 
>> - - it seems to me, that the playouts get much more important
>> with such a strong move prediction. Often the move prediction
>> seems better the playouts (I use 8000 at the moment against pachi
>> 32000 with about 70% winrate on 19x19, but with an extremely
>> focused progressive widening (a=400, a=20 was usual).
>> 
>> - - live and death becomes worse. My interpretation is, that the
>> strong CNN does not play moves, which obviously do not help to
>> get a group life, but would help the playouts to recognize the
>> group is dead. (http://physik.de/example.sgf top black group was
>> with weaker move prediction read very dead, with good CNN it was
>> 30% alive or so :(
>> 
>> 
>> OK, hope you try it, as you know our engine oakfoam is open
>> source :) We just merged all the CNN stuff into the main branch! 
>> https://bitbucket.org/francoisvn/oakfoam/wiki/Home 
>> http://oakfoam.com
>> 
>> 
>> Do the very best with the CNN
>> 
>> Detlef
> 
> ___ Computer-go mailing
> list Computer-go@computer-go.org 
> http://computer-go.org/mailman/listinfo/computer-go
> 
-BEGIN PGP SIGNATURE-
Version: GnuPG v2.0.22 (GNU/Linux)

iQIcBAEBAgAGBQJWglFcAAoJEInWdHg+Znf42q4P/AnMdgqhps4RSJG3NoLiwEUq
QmT4mQd58WbuxnXRO4xiyIKGTQq13+FOpqVu7RgFPXxQaKS+8Hi1qpGVjg8aE8Zh
bnHb3D+p30hv9lCT8e4xNQ2B1J

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-09 Thread Michael Markefka

Thank you for the feedback, everyone.


Regarding the CPU-GPU roundtrips, I'm wondering whether it'd be
possible to recursively apply the output matrix to the prior input
matrix to update board positions within the GPU and  without any
actual (possibly CPU-based) evaluation until all branches come up with
game ending states. I assume illegal moves would mostly fall away when
sticking to the top ten or top five move considerations provided by
the CNN.

As for performance, I could imagine initialization being relatively
slow, but wouldn't be surprised if the GPU-based CNN performance could
offer a branch size, running through many parallel boards with
comparatively minor performance impact, where this outweighed the
initial overhead again.

Whether this would provide a better evaluation function than MCTS I
don't know, but just like Alvaro I would love to see this tried, even
if just to rule it out for the moment.


I've got a GTX 980 Ti on a 4790k with 16 GB at home. For a low key
test I could run Windows (CUDA installed and running, tested with
pylearn2) or Ubuntu from a live setup on USB and would be willing to
run test code, if somebody provided a package I could simply download
and execute.


All the best

Michael


On Tue, Dec 8, 2015 at 7:52 PM, Álvaro Begué  wrote:
> Of course whether these "neuro-playouts" are any better than the heavy
> playouts currently being used by strong programs is an empirical question.
> But I would love to see it answered...
>
>
>
> On Tue, Dec 8, 2015 at 1:31 PM, David Ongaro 
> wrote:
>>
>> Did everyone forget the fact that stronger playouts don't necessarily lead
>> to an better evaluation function? (Yes, that what playouts essential are, a
>> dynamic evaluation function.) This is even under the assumption that we can
>> reach the same number of playouts per move.
>>
>>
>> On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:
>>
>> I don't think the CPU-GPU communication is what's going to kill this idea.
>> The latency in actually computing the feed-forward pass of the CNN is going
>> to be in the order of 0.1 seconds (I am guessing here), which means
>> finishing the first playout will take many seconds.
>>
>> So perhaps it would be interesting to do something like this for
>> correspondence games, but not for regular games.
>>
>>
>> Álvaro.
>>
>>
>>
>> On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  wrote:
>>>
>>>   Hi!
>>>
>>>   Well, for this to be practical the entire playout would have to be
>>> executed on the GPU, with no round-trips to the CPU.  That's what my
>>> email was aimed at.
>>>
>>> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
>>> > Regarding full CNN playouts, I think that problem is that a playout is
>>> > a
>>> > long serial process, given 200-300 moves a game. You need to construct
>>> > planes and transfer them to GPU for each move and read result back (at
>>> > least with current CNN implementations afaik), so my guess would be
>>> > that
>>> > such playout would take time in order of seconds. So there seems to be
>>> > a
>>> > tradeoff, CNN playouts are (probably much) better (at "playing better
>>> > games") than e.g. distribution playouts, but whether this is worth the
>>> > implied (probably much) lower height of the MC tree is a question.
>>> >
>>> > Maybe if you had really a lot of GPUs and very high thinking time, this
>>> > could be the way.
>>> >
>>> > Josef
>>> >
>>> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
>>> >
>>> > >   Hi!
>>> > >
>>> > >   In case someone is looking for a starting point to actually
>>> > > implement
>>> > > Go rules etc. on GPU, you may find useful:
>>> > >
>>> > >
>>> > >
>>> > > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
>>> > >
>>> > >   I wonder if you can easily integrate caffe GPU kernels in another
>>> > > GPU
>>> > > kernel like this?  But without training, reimplementing the NN could
>>> > > be
>>> > > pretty straightforward.
>>> > >
>>> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
>>> > > > Hello Detlef,
>>> > > >
>>> > > > I've got a question regarding CNN-based Go engines I couldn't find
>>> > > > anything about on this list. As I've been following your posts
>>> > > > here, I
>>> > > > thought you might be the right person to ask.
>>> > > >
>>> > > > Have you ever tried using the CNN for complete playouts? I know
>>> > > > that
>>> > > > CNNs have been tried for move prediction, immediate scoring and
>>> > > > move
>>> > > > generation to be used in an MC evaluator, but couldn't find
>>> > > > anything
>>> > > > about CNN-based playouts.
>>> > > >
>>> > > > It might only be feasible to play out the CNN's first choice move
>>> > > > for
>>> > > > evaluation purposes, but considering how well the performance of
>>> > > > batch
>>> > > > sizes scales, especially on GPU-based CNN applications, it might be
>>> > > > possible to setup

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-09 Thread Igor Polyakov

I doubt that the illegal moves would fall away since every professional 
would retake the ko... if it was legal


On 2015-12-09 4:59, Michael Markefka wrote:

Thank you for the feedback, everyone.


Regarding the CPU-GPU roundtrips, I'm wondering whether it'd be
possible to recursively apply the output matrix to the prior input
matrix to update board positions within the GPU and  without any
actual (possibly CPU-based) evaluation until all branches come up with
game ending states. I assume illegal moves would mostly fall away when
sticking to the top ten or top five move considerations provided by
the CNN.

As for performance, I could imagine initialization being relatively
slow, but wouldn't be surprised if the GPU-based CNN performance could
offer a branch size, running through many parallel boards with
comparatively minor performance impact, where this outweighed the
initial overhead again.

Whether this would provide a better evaluation function than MCTS I
don't know, but just like Alvaro I would love to see this tried, even
if just to rule it out for the moment.


I've got a GTX 980 Ti on a 4790k with 16 GB at home. For a low key
test I could run Windows (CUDA installed and running, tested with
pylearn2) or Ubuntu from a live setup on USB and would be willing to
run test code, if somebody provided a package I could simply download
and execute.


All the best

Michael


On Tue, Dec 8, 2015 at 7:52 PM, Álvaro Begué  wrote:

Of course whether these "neuro-playouts" are any better than the heavy
playouts currently being used by strong programs is an empirical question.
But I would love to see it answered...



On Tue, Dec 8, 2015 at 1:31 PM, David Ongaro 
wrote:

Did everyone forget the fact that stronger playouts don't necessarily lead
to an better evaluation function? (Yes, that what playouts essential are, a
dynamic evaluation function.) This is even under the assumption that we can
reach the same number of playouts per move.


On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:

I don't think the CPU-GPU communication is what's going to kill this idea.
The latency in actually computing the feed-forward pass of the CNN is going
to be in the order of 0.1 seconds (I am guessing here), which means
finishing the first playout will take many seconds.

So perhaps it would be interesting to do something like this for
correspondence games, but not for regular games.


Álvaro.



On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  wrote:

   Hi!

   Well, for this to be practical the entire playout would have to be
executed on the GPU, with no round-trips to the CPU.  That's what my
email was aimed at.

On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:

Regarding full CNN playouts, I think that problem is that a playout is
a
long serial process, given 200-300 moves a game. You need to construct
planes and transfer them to GPU for each move and read result back (at
least with current CNN implementations afaik), so my guess would be
that
such playout would take time in order of seconds. So there seems to be
a
tradeoff, CNN playouts are (probably much) better (at "playing better
games") than e.g. distribution playouts, but whether this is worth the
implied (probably much) lower height of the MC tree is a question.

Maybe if you had really a lot of GPUs and very high thinking time, this
could be the way.

Josef

On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:


   Hi!

   In case someone is looking for a starting point to actually
implement
Go rules etc. on GPU, you may find useful:



https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html

   I wonder if you can easily integrate caffe GPU kernels in another
GPU
kernel like this?  But without training, reimplementing the NN could
be
pretty straightforward.

On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:

Hello Detlef,

I've got a question regarding CNN-based Go engines I couldn't find
anything about on this list. As I've been following your posts
here, I
thought you might be the right person to ask.

Have you ever tried using the CNN for complete playouts? I know
that
CNNs have been tried for move prediction, immediate scoring and
move
generation to be used in an MC evaluator, but couldn't find
anything
about CNN-based playouts.

It might only be feasible to play out the CNN's first choice move
for
evaluation purposes, but considering how well the performance of
batch
sizes scales, especially on GPU-based CNN applications, it might be
possible to setup something like 10 candidate moves, 10 reply
candidate moves and then have the CNN play out the first choice
move
for those 100 board positions until the end and then sum up scores
again for move evaluation (and/or possibly apply some other tried
and
tested methods like minimax). Given that the number of 10 moves is
supposed to be illustrative rather than representative, other

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-09 Thread Michael Markefka

I think ko moves are taken into account on one of in the input planes
for most configurations. At least I hope remember that correctly.
Could it be achieved to create such a plane from the prior input
matrix and following output matrix by difference?

On Wed, Dec 9, 2015 at 2:08 PM, Igor Polyakov
 wrote:
> I doubt that the illegal moves would fall away since every professional
> would retake the ko... if it was legal
>
>
> On 2015-12-09 4:59, Michael Markefka wrote:
>>
>> Thank you for the feedback, everyone.
>>
>>
>> Regarding the CPU-GPU roundtrips, I'm wondering whether it'd be
>> possible to recursively apply the output matrix to the prior input
>> matrix to update board positions within the GPU and  without any
>> actual (possibly CPU-based) evaluation until all branches come up with
>> game ending states. I assume illegal moves would mostly fall away when
>> sticking to the top ten or top five move considerations provided by
>> the CNN.
>>
>> As for performance, I could imagine initialization being relatively
>> slow, but wouldn't be surprised if the GPU-based CNN performance could
>> offer a branch size, running through many parallel boards with
>> comparatively minor performance impact, where this outweighed the
>> initial overhead again.
>>
>> Whether this would provide a better evaluation function than MCTS I
>> don't know, but just like Alvaro I would love to see this tried, even
>> if just to rule it out for the moment.
>>
>>
>> I've got a GTX 980 Ti on a 4790k with 16 GB at home. For a low key
>> test I could run Windows (CUDA installed and running, tested with
>> pylearn2) or Ubuntu from a live setup on USB and would be willing to
>> run test code, if somebody provided a package I could simply download
>> and execute.
>>
>>
>> All the best
>>
>> Michael
>>
>>
>> On Tue, Dec 8, 2015 at 7:52 PM, Álvaro Begué 
>> wrote:
>>>
>>> Of course whether these "neuro-playouts" are any better than the heavy
>>> playouts currently being used by strong programs is an empirical
>>> question.
>>> But I would love to see it answered...
>>>
>>>
>>>
>>> On Tue, Dec 8, 2015 at 1:31 PM, David Ongaro 
>>> wrote:

 Did everyone forget the fact that stronger playouts don't necessarily
 lead
 to an better evaluation function? (Yes, that what playouts essential
 are, a
 dynamic evaluation function.) This is even under the assumption that we
 can
 reach the same number of playouts per move.


 On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:

 I don't think the CPU-GPU communication is what's going to kill this
 idea.
 The latency in actually computing the feed-forward pass of the CNN is
 going
 to be in the order of 0.1 seconds (I am guessing here), which means
 finishing the first playout will take many seconds.

 So perhaps it would be interesting to do something like this for
 correspondence games, but not for regular games.


 Álvaro.



 On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  wrote:
>
>Hi!
>
>Well, for this to be practical the entire playout would have to be
> executed on the GPU, with no round-trips to the CPU.  That's what my
> email was aimed at.
>
> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
>>
>> Regarding full CNN playouts, I think that problem is that a playout is
>> a
>> long serial process, given 200-300 moves a game. You need to construct
>> planes and transfer them to GPU for each move and read result back (at
>> least with current CNN implementations afaik), so my guess would be
>> that
>> such playout would take time in order of seconds. So there seems to be
>> a
>> tradeoff, CNN playouts are (probably much) better (at "playing better
>> games") than e.g. distribution playouts, but whether this is worth the
>> implied (probably much) lower height of the MC tree is a question.
>>
>> Maybe if you had really a lot of GPUs and very high thinking time,
>> this
>> could be the way.
>>
>> Josef
>>
>> On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
>>
>>>Hi!
>>>
>>>In case someone is looking for a starting point to actually
>>> implement
>>> Go rules etc. on GPU, you may find useful:
>>>
>>>
>>>
>>>
>>> https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
>>>
>>>I wonder if you can easily integrate caffe GPU kernels in another
>>> GPU
>>> kernel like this?  But without training, reimplementing the NN could
>>> be
>>> pretty straightforward.
>>>
>>> On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:

 Hello Detlef,

 I've got a question regarding CNN-based Go engines I couldn't

[Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Detlef Schmicker

-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1

Hi,

as somebody ask I will offer my actual CNN for testing.

It has 54% prediction on KGS 6d+ data (which I thought would be state
of the art when I started training, but it is not anymore:).

it has:
1
2
3
> 4 libs playing color
1
2
3
> 4 libs opponent color
Empty points
last move
second last move
third last move
forth last move

input layers, and it is fully convolutional, so with just editing the
golast19.prototxt file you can use it for 13x13 as well, as I did on
last sunday. It was used in November tournament as well.

You can find it
http://physik.de/CNNlast.tar.gz



If you try here some points I like to get discussion:

- - it seems to me, that the playouts get much more important with such
a strong move prediction. Often the move prediction seems better the
playouts (I use 8000 at the moment against pachi 32000 with about 70%
winrate on 19x19, but with an extremely focused progressive widening
(a=400, a=20 was usual).

- - live and death becomes worse. My interpretation is, that the strong
CNN does not play moves, which obviously do not help to get a group
life, but would help the playouts to recognize the group is dead.
(http://physik.de/example.sgf top black group was with weaker move
prediction read very dead, with good CNN it was 30% alive or so :(


OK, hope you try it, as you know our engine oakfoam is open source :)
We just merged all the CNN stuff into the main branch!
https://bitbucket.org/francoisvn/oakfoam/wiki/Home
http://oakfoam.com


Do the very best with the CNN

Detlef




code:
if (col==Go::BLACK) {
  for (int j=0;jinGroup(pos))
libs=board->getGroup(pos)->numRealLibs()-1;
if (libs>3) libs=3;
if (board->getColor(pos)==Go::BLACK)
  {
  data[(0+libs)*size*size + size*j + k]=1.0;
  //data[size*size+size*j+k]=0.0;
  }
  else if (board->getColor(pos)==Go::WHITE)
  {
  //data[j*size+k]=0.0;
  data[(4+libs)*size*size + size*j + k]=1.0;
  }
  else if
(board->getColor(Go::Position::xy2pos(j,k,size))==Go::EMPTY)
  {
data[8*size*size + size*j + k]=1.0;
  }
}
}
if (col==Go::WHITE) {
  for (int j=0;jinGroup(pos))
libs=board->getGroup(pos)->numRealLibs()-1;
if (libs>3) libs=3;
if (board->getColor(pos)==Go::BLACK)
  {
  data[(4+libs)*size*size + size*j + k]=1.0;
  //data[size*size+size*j+k]=0.0;
  }
  else if (board->getColor(pos)==Go::WHITE)
  {
  //data[j*size+k]=0.0;
  data[(0+libs)*size*size + size*j + k]=1.0;
  }
  else if (board->getColor(pos)==Go::EMPTY)
  {
data[8*size*size + size*j + k]=1.0;
  }
}
}
if (caffe_test_net_input_dim > 9) {
  if (board->getLastMove().isNormal()) {
int j=Go::Position::pos2x(board->getLastMove().getPosition(),size);
int k=Go::Position::pos2y(board->getLastMove().getPosition(),size);
data[9*size*size+size*j+k]=1.0;
  }
  if (board->getSecondLastMove().isNormal()) {
int
j=Go::Position::pos2x(board->getSecondLastMove().getPosition(),size);
int
k=Go::Position::pos2y(board->getSecondLastMove().getPosition(),size);
data[10*size*size+size*j+k]=1.0;
  }
  if (board->getThirdLastMove().isNormal()) {
int
j=Go::Position::pos2x(board->getThirdLastMove().getPosition(),size);
int
k=Go::Position::pos2y(board->getThirdLastMove().getPosition(),size);
data[11*size*size+size*j+k]=1.0;
  }
  if (board->getForthLastMove().isNormal()) {
int
j=Go::Position::pos2x(board->getForthLastMove().getPosition(),size);
int
k=Go::Position::pos2y(board->getForthLastMove().getPosition(),size);
data[12*size*size+size*j+k]=1.0;
  }
}

-BEGIN PGP SIGNATURE-
Version: GnuPG v2.0.22 (GNU/Linux)

iQIcBAEBAgAGBQJWZvOlAAoJEInWdHg+Znf4t8cP/2a9fE7rVb3Hz9wvdMkvVkFS
4Y3AomVx8i56jexVyXuzKihfizVRM7x6lBiwjYBhj4Rm9UFWjj2ZvDzBGCm3Sy4I
SpG8D01VnzVR6iC1YTu3ecv9Wo4pTjc7NL5pAxiZDB0V7OTRklfZAYsX4mWyHygn

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Petr Baudis

  Hi!

  In case someone is looking for a starting point to actually implement
Go rules etc. on GPU, you may find useful:

https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html

  I wonder if you can easily integrate caffe GPU kernels in another GPU
kernel like this?  But without training, reimplementing the NN could be
pretty straightforward.

On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> Hello Detlef,
> 
> I've got a question regarding CNN-based Go engines I couldn't find
> anything about on this list. As I've been following your posts here, I
> thought you might be the right person to ask.
> 
> Have you ever tried using the CNN for complete playouts? I know that
> CNNs have been tried for move prediction, immediate scoring and move
> generation to be used in an MC evaluator, but couldn't find anything
> about CNN-based playouts.
> 
> It might only be feasible to play out the CNN's first choice move for
> evaluation purposes, but considering how well the performance of batch
> sizes scales, especially on GPU-based CNN applications, it might be
> possible to setup something like 10 candidate moves, 10 reply
> candidate moves and then have the CNN play out the first choice move
> for those 100 board positions until the end and then sum up scores
> again for move evaluation (and/or possibly apply some other tried and
> tested methods like minimax). Given that the number of 10 moves is
> supposed to be illustrative rather than representative, other
> configurations of depth and width in position generation and
> evaluation would be possible.
> 
> It feels like CNN can provide a very focused, high-quality width in
> move generation, but it might also be possible to apply that quality
> to depth of evaluation.
> 
> Any thoughts to share?
> 
> 
> All the best
> 
> Michael
> 
> On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker  wrote:
> > -BEGIN PGP SIGNED MESSAGE-
> > Hash: SHA1
> >
> > Hi,
> >
> > as somebody ask I will offer my actual CNN for testing.
> >
> > It has 54% prediction on KGS 6d+ data (which I thought would be state
> > of the art when I started training, but it is not anymore:).
> >
> > it has:
> > 1
> > 2
> > 3
> >> 4 libs playing color
> > 1
> > 2
> > 3
> >> 4 libs opponent color
> > Empty points
> > last move
> > second last move
> > third last move
> > forth last move
> >
> > input layers, and it is fully convolutional, so with just editing the
> > golast19.prototxt file you can use it for 13x13 as well, as I did on
> > last sunday. It was used in November tournament as well.
> >
> > You can find it
> > http://physik.de/CNNlast.tar.gz
> >
> >
> >
> > If you try here some points I like to get discussion:
> >
> > - - it seems to me, that the playouts get much more important with such
> > a strong move prediction. Often the move prediction seems better the
> > playouts (I use 8000 at the moment against pachi 32000 with about 70%
> > winrate on 19x19, but with an extremely focused progressive widening
> > (a=400, a=20 was usual).
> >
> > - - live and death becomes worse. My interpretation is, that the strong
> > CNN does not play moves, which obviously do not help to get a group
> > life, but would help the playouts to recognize the group is dead.
> > (http://physik.de/example.sgf top black group was with weaker move
> > prediction read very dead, with good CNN it was 30% alive or so :(
> >
> >
> > OK, hope you try it, as you know our engine oakfoam is open source :)
> > We just merged all the CNN stuff into the main branch!
> > https://bitbucket.org/francoisvn/oakfoam/wiki/Home
> > http://oakfoam.com
> >
> >
> > Do the very best with the CNN
> >
> > Detlef
> >
> >
> >
> >
> > code:
> > if (col==Go::BLACK) {
> >   for (int j=0;j > for (int k=0;k >   {
> > for (int l=0;l > data[l*size*size+size*j+k]=0;
> > //fprintf(stderr,"%d %d %d\n",i,j,k);
> > int pos=Go::Position::xy2pos(j,k,size);
> > int libs=0;
> > if (board->inGroup(pos))
> > libs=board->getGroup(pos)->numRealLibs()-1;
> > if (libs>3) libs=3;
> > if (board->getColor(pos)==Go::BLACK)
> >   {
> >   data[(0+libs)*size*size + size*j + k]=1.0;
> >   //data[size*size+size*j+k]=0.0;
> >   }
> >   else if (board->getColor(pos)==Go::WHITE)
> >   {
> >   //data[j*size+k]=0.0;
> >   data[(4+libs)*size*size + size*j + k]=1.0;
> >   }
> >   else if
> > (board->getColor(Go::Position::xy2pos(j,k,size))==Go::EMPTY)
> >   {
> > data[8*size*size + size*j + k]=1.0;
> >   }
> > }
> > }
> > if (col==Go::WHITE) {
> >   for (int j=0;j >

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Josef Moudrik

Regarding full CNN playouts, I think that problem is that a playout is a
long serial process, given 200-300 moves a game. You need to construct
planes and transfer them to GPU for each move and read result back (at
least with current CNN implementations afaik), so my guess would be that
such playout would take time in order of seconds. So there seems to be a
tradeoff, CNN playouts are (probably much) better (at "playing better
games") than e.g. distribution playouts, but whether this is worth the
implied (probably much) lower height of the MC tree is a question.

Maybe if you had really a lot of GPUs and very high thinking time, this
could be the way.

Josef

On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:

>   Hi!
>
>   In case someone is looking for a starting point to actually implement
> Go rules etc. on GPU, you may find useful:
>
>
> https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
>
>   I wonder if you can easily integrate caffe GPU kernels in another GPU
> kernel like this?  But without training, reimplementing the NN could be
> pretty straightforward.
>
> On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> > Hello Detlef,
> >
> > I've got a question regarding CNN-based Go engines I couldn't find
> > anything about on this list. As I've been following your posts here, I
> > thought you might be the right person to ask.
> >
> > Have you ever tried using the CNN for complete playouts? I know that
> > CNNs have been tried for move prediction, immediate scoring and move
> > generation to be used in an MC evaluator, but couldn't find anything
> > about CNN-based playouts.
> >
> > It might only be feasible to play out the CNN's first choice move for
> > evaluation purposes, but considering how well the performance of batch
> > sizes scales, especially on GPU-based CNN applications, it might be
> > possible to setup something like 10 candidate moves, 10 reply
> > candidate moves and then have the CNN play out the first choice move
> > for those 100 board positions until the end and then sum up scores
> > again for move evaluation (and/or possibly apply some other tried and
> > tested methods like minimax). Given that the number of 10 moves is
> > supposed to be illustrative rather than representative, other
> > configurations of depth and width in position generation and
> > evaluation would be possible.
> >
> > It feels like CNN can provide a very focused, high-quality width in
> > move generation, but it might also be possible to apply that quality
> > to depth of evaluation.
> >
> > Any thoughts to share?
> >
> >
> > All the best
> >
> > Michael
> >
> > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker  wrote:
> > > -BEGIN PGP SIGNED MESSAGE-
> > > Hash: SHA1
> > >
> > > Hi,
> > >
> > > as somebody ask I will offer my actual CNN for testing.
> > >
> > > It has 54% prediction on KGS 6d+ data (which I thought would be state
> > > of the art when I started training, but it is not anymore:).
> > >
> > > it has:
> > > 1
> > > 2
> > > 3
> > >> 4 libs playing color
> > > 1
> > > 2
> > > 3
> > >> 4 libs opponent color
> > > Empty points
> > > last move
> > > second last move
> > > third last move
> > > forth last move
> > >
> > > input layers, and it is fully convolutional, so with just editing the
> > > golast19.prototxt file you can use it for 13x13 as well, as I did on
> > > last sunday. It was used in November tournament as well.
> > >
> > > You can find it
> > > http://physik.de/CNNlast.tar.gz
> > >
> > >
> > >
> > > If you try here some points I like to get discussion:
> > >
> > > - - it seems to me, that the playouts get much more important with such
> > > a strong move prediction. Often the move prediction seems better the
> > > playouts (I use 8000 at the moment against pachi 32000 with about 70%
> > > winrate on 19x19, but with an extremely focused progressive widening
> > > (a=400, a=20 was usual).
> > >
> > > - - live and death becomes worse. My interpretation is, that the strong
> > > CNN does not play moves, which obviously do not help to get a group
> > > life, but would help the playouts to recognize the group is dead.
> > > (http://physik.de/example.sgf top black group was with weaker move
> > > prediction read very dead, with good CNN it was 30% alive or so :(
> > >
> > >
> > > OK, hope you try it, as you know our engine oakfoam is open source :)
> > > We just merged all the CNN stuff into the main branch!
> > > https://bitbucket.org/francoisvn/oakfoam/wiki/Home
> > > http://oakfoam.com
> > >
> > >
> > > Do the very best with the CNN
> > >
> > > Detlef
> > >
> > >
> > >
> > >
> > > code:
> > > if (col==Go::BLACK) {
> > >   for (int j=0;j > > for (int k=0;k > >   {
> > > for (int l=0;l > > data[l*size*size+size*j+k]=0;
> > > //fprintf(stderr,"%d %d %d\n",i,j,k);
> > > int

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Michael Markefka

Hello Detlef,

I've got a question regarding CNN-based Go engines I couldn't find
anything about on this list. As I've been following your posts here, I
thought you might be the right person to ask.

Have you ever tried using the CNN for complete playouts? I know that
CNNs have been tried for move prediction, immediate scoring and move
generation to be used in an MC evaluator, but couldn't find anything
about CNN-based playouts.

It might only be feasible to play out the CNN's first choice move for
evaluation purposes, but considering how well the performance of batch
sizes scales, especially on GPU-based CNN applications, it might be
possible to setup something like 10 candidate moves, 10 reply
candidate moves and then have the CNN play out the first choice move
for those 100 board positions until the end and then sum up scores
again for move evaluation (and/or possibly apply some other tried and
tested methods like minimax). Given that the number of 10 moves is
supposed to be illustrative rather than representative, other
configurations of depth and width in position generation and
evaluation would be possible.

It feels like CNN can provide a very focused, high-quality width in
move generation, but it might also be possible to apply that quality
to depth of evaluation.

Any thoughts to share?

All the best

Michael

On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker  wrote:
> -BEGIN PGP SIGNED MESSAGE-
> Hash: SHA1
>
> Hi,
>
> as somebody ask I will offer my actual CNN for testing.
>
> It has 54% prediction on KGS 6d+ data (which I thought would be state
> of the art when I started training, but it is not anymore:).
>
> it has:
> 1
> 2
> 3
>> 4 libs playing color
> 1
> 2
> 3
>> 4 libs opponent color
> Empty points
> last move
> second last move
> third last move
> forth last move
>
> input layers, and it is fully convolutional, so with just editing the
> golast19.prototxt file you can use it for 13x13 as well, as I did on
> last sunday. It was used in November tournament as well.
>
> You can find it
> http://physik.de/CNNlast.tar.gz
>
>
>
> If you try here some points I like to get discussion:
>
> - - it seems to me, that the playouts get much more important with such
> a strong move prediction. Often the move prediction seems better the
> playouts (I use 8000 at the moment against pachi 32000 with about 70%
> winrate on 19x19, but with an extremely focused progressive widening
> (a=400, a=20 was usual).
>
> - - live and death becomes worse. My interpretation is, that the strong
> CNN does not play moves, which obviously do not help to get a group
> life, but would help the playouts to recognize the group is dead.
> (http://physik.de/example.sgf top black group was with weaker move
> prediction read very dead, with good CNN it was 30% alive or so :(
>
>
> OK, hope you try it, as you know our engine oakfoam is open source :)
> We just merged all the CNN stuff into the main branch!
> https://bitbucket.org/francoisvn/oakfoam/wiki/Home
> http://oakfoam.com
>
>
> Do the very best with the CNN
>
> Detlef
>
>
>
>
> code:
> if (col==Go::BLACK) {
>   for (int j=0;j for (int k=0;k   {
> for (int l=0;l data[l*size*size+size*j+k]=0;
> //fprintf(stderr,"%d %d %d\n",i,j,k);
> int pos=Go::Position::xy2pos(j,k,size);
> int libs=0;
> if (board->inGroup(pos))
> libs=board->getGroup(pos)->numRealLibs()-1;
> if (libs>3) libs=3;
> if (board->getColor(pos)==Go::BLACK)
>   {
>   data[(0+libs)*size*size + size*j + k]=1.0;
>   //data[size*size+size*j+k]=0.0;
>   }
>   else if (board->getColor(pos)==Go::WHITE)
>   {
>   //data[j*size+k]=0.0;
>   data[(4+libs)*size*size + size*j + k]=1.0;
>   }
>   else if
> (board->getColor(Go::Position::xy2pos(j,k,size))==Go::EMPTY)
>   {
> data[8*size*size + size*j + k]=1.0;
>   }
> }
> }
> if (col==Go::WHITE) {
>   for (int j=0;j for (int k=0;k   {//fprintf(stderr,"%d %d %d\n",i,j,k);
> for (int l=0;l data[l*size*size+size*j+k]=0;
> //fprintf(stderr,"%d %d %d\n",i,j,k);
> int pos=Go::Position::xy2pos(j,k,size);
> int libs=0;
> if (board->inGroup(pos))
> libs=board->getGroup(pos)->numRealLibs()-1;
> if (libs>3) libs=3;
> if (board->getColor(pos)==Go::BLACK)
>   {
>   data[(4+libs)*size*size + size*j + k]=1.0;
>   //data[size*size+size*j+k]=0.0;
>   }
>   else if

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Petr Baudis

  Hi!

  Well, for this to be practical the entire playout would have to be
executed on the GPU, with no round-trips to the CPU.  That's what my
email was aimed at.

On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
> Regarding full CNN playouts, I think that problem is that a playout is a
> long serial process, given 200-300 moves a game. You need to construct
> planes and transfer them to GPU for each move and read result back (at
> least with current CNN implementations afaik), so my guess would be that
> such playout would take time in order of seconds. So there seems to be a
> tradeoff, CNN playouts are (probably much) better (at "playing better
> games") than e.g. distribution playouts, but whether this is worth the
> implied (probably much) lower height of the MC tree is a question.
> 
> Maybe if you had really a lot of GPUs and very high thinking time, this
> could be the way.
> 
> Josef
> 
> On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
> 
> >   Hi!
> >
> >   In case someone is looking for a starting point to actually implement
> > Go rules etc. on GPU, you may find useful:
> >
> >
> > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
> >
> >   I wonder if you can easily integrate caffe GPU kernels in another GPU
> > kernel like this?  But without training, reimplementing the NN could be
> > pretty straightforward.
> >
> > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> > > Hello Detlef,
> > >
> > > I've got a question regarding CNN-based Go engines I couldn't find
> > > anything about on this list. As I've been following your posts here, I
> > > thought you might be the right person to ask.
> > >
> > > Have you ever tried using the CNN for complete playouts? I know that
> > > CNNs have been tried for move prediction, immediate scoring and move
> > > generation to be used in an MC evaluator, but couldn't find anything
> > > about CNN-based playouts.
> > >
> > > It might only be feasible to play out the CNN's first choice move for
> > > evaluation purposes, but considering how well the performance of batch
> > > sizes scales, especially on GPU-based CNN applications, it might be
> > > possible to setup something like 10 candidate moves, 10 reply
> > > candidate moves and then have the CNN play out the first choice move
> > > for those 100 board positions until the end and then sum up scores
> > > again for move evaluation (and/or possibly apply some other tried and
> > > tested methods like minimax). Given that the number of 10 moves is
> > > supposed to be illustrative rather than representative, other
> > > configurations of depth and width in position generation and
> > > evaluation would be possible.
> > >
> > > It feels like CNN can provide a very focused, high-quality width in
> > > move generation, but it might also be possible to apply that quality
> > > to depth of evaluation.
> > >
> > > Any thoughts to share?
> > >
> > >
> > > All the best
> > >
> > > Michael
> > >
> > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker  wrote:
> > > > -BEGIN PGP SIGNED MESSAGE-
> > > > Hash: SHA1
> > > >
> > > > Hi,
> > > >
> > > > as somebody ask I will offer my actual CNN for testing.
> > > >
> > > > It has 54% prediction on KGS 6d+ data (which I thought would be state
> > > > of the art when I started training, but it is not anymore:).
> > > >
> > > > it has:
> > > > 1
> > > > 2
> > > > 3
> > > >> 4 libs playing color
> > > > 1
> > > > 2
> > > > 3
> > > >> 4 libs opponent color
> > > > Empty points
> > > > last move
> > > > second last move
> > > > third last move
> > > > forth last move
> > > >
> > > > input layers, and it is fully convolutional, so with just editing the
> > > > golast19.prototxt file you can use it for 13x13 as well, as I did on
> > > > last sunday. It was used in November tournament as well.
> > > >
> > > > You can find it
> > > > http://physik.de/CNNlast.tar.gz
> > > >
> > > >
> > > >
> > > > If you try here some points I like to get discussion:
> > > >
> > > > - - it seems to me, that the playouts get much more important with such
> > > > a strong move prediction. Often the move prediction seems better the
> > > > playouts (I use 8000 at the moment against pachi 32000 with about 70%
> > > > winrate on 19x19, but with an extremely focused progressive widening
> > > > (a=400, a=20 was usual).
> > > >
> > > > - - live and death becomes worse. My interpretation is, that the strong
> > > > CNN does not play moves, which obviously do not help to get a group
> > > > life, but would help the playouts to recognize the group is dead.
> > > > (http://physik.de/example.sgf top black group was with weaker move
> > > > prediction read very dead, with good CNN it was 30% alive or so :(
> > > >
> > > >
> > > > OK, hope you try it, as you know our engine oakfoam is open source :)
> > > > We just merged all the CNN stuff into the main branch!
> > > > https://bitbucket.org/francoisvn/oakfoam/wiki/Home

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Josef Moudrik

Yes, that's why I wrote with current CNN implementations. But I still
wonder whether my estimate for the round-trip length is at least of the
correct magnitude.

Josef

On Tue, Dec 8, 2015 at 6:03 PM Petr Baudis  wrote:

>   Hi!
>
>   Well, for this to be practical the entire playout would have to be
> executed on the GPU, with no round-trips to the CPU.  That's what my
> email was aimed at.
>
> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
> > Regarding full CNN playouts, I think that problem is that a playout is a
> > long serial process, given 200-300 moves a game. You need to construct
> > planes and transfer them to GPU for each move and read result back (at
> > least with current CNN implementations afaik), so my guess would be that
> > such playout would take time in order of seconds. So there seems to be a
> > tradeoff, CNN playouts are (probably much) better (at "playing better
> > games") than e.g. distribution playouts, but whether this is worth the
> > implied (probably much) lower height of the MC tree is a question.
> >
> > Maybe if you had really a lot of GPUs and very high thinking time, this
> > could be the way.
> >
> > Josef
> >
> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
> >
> > >   Hi!
> > >
> > >   In case someone is looking for a starting point to actually implement
> > > Go rules etc. on GPU, you may find useful:
> > >
> > >
> > > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
> > >
> > >   I wonder if you can easily integrate caffe GPU kernels in another GPU
> > > kernel like this?  But without training, reimplementing the NN could be
> > > pretty straightforward.
> > >
> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> > > > Hello Detlef,
> > > >
> > > > I've got a question regarding CNN-based Go engines I couldn't find
> > > > anything about on this list. As I've been following your posts here,
> I
> > > > thought you might be the right person to ask.
> > > >
> > > > Have you ever tried using the CNN for complete playouts? I know that
> > > > CNNs have been tried for move prediction, immediate scoring and move
> > > > generation to be used in an MC evaluator, but couldn't find anything
> > > > about CNN-based playouts.
> > > >
> > > > It might only be feasible to play out the CNN's first choice move for
> > > > evaluation purposes, but considering how well the performance of
> batch
> > > > sizes scales, especially on GPU-based CNN applications, it might be
> > > > possible to setup something like 10 candidate moves, 10 reply
> > > > candidate moves and then have the CNN play out the first choice move
> > > > for those 100 board positions until the end and then sum up scores
> > > > again for move evaluation (and/or possibly apply some other tried and
> > > > tested methods like minimax). Given that the number of 10 moves is
> > > > supposed to be illustrative rather than representative, other
> > > > configurations of depth and width in position generation and
> > > > evaluation would be possible.
> > > >
> > > > It feels like CNN can provide a very focused, high-quality width in
> > > > move generation, but it might also be possible to apply that quality
> > > > to depth of evaluation.
> > > >
> > > > Any thoughts to share?
> > > >
> > > >
> > > > All the best
> > > >
> > > > Michael
> > > >
> > > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker 
> wrote:
> > > > > -BEGIN PGP SIGNED MESSAGE-
> > > > > Hash: SHA1
> > > > >
> > > > > Hi,
> > > > >
> > > > > as somebody ask I will offer my actual CNN for testing.
> > > > >
> > > > > It has 54% prediction on KGS 6d+ data (which I thought would be
> state
> > > > > of the art when I started training, but it is not anymore:).
> > > > >
> > > > > it has:
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs playing color
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs opponent color
> > > > > Empty points
> > > > > last move
> > > > > second last move
> > > > > third last move
> > > > > forth last move
> > > > >
> > > > > input layers, and it is fully convolutional, so with just editing
> the
> > > > > golast19.prototxt file you can use it for 13x13 as well, as I did
> on
> > > > > last sunday. It was used in November tournament as well.
> > > > >
> > > > > You can find it
> > > > > http://physik.de/CNNlast.tar.gz
> > > > >
> > > > >
> > > > >
> > > > > If you try here some points I like to get discussion:
> > > > >
> > > > > - - it seems to me, that the playouts get much more important with
> such
> > > > > a strong move prediction. Often the move prediction seems better
> the
> > > > > playouts (I use 8000 at the moment against pachi 32000 with about
> 70%
> > > > > winrate on 19x19, but with an extremely focused progressive
> widening
> > > > > (a=400, a=20 was usual).
> > > > >
> > > > > - - live and death becomes worse. My interpretation is, that the
> strong
> > > > > CNN does not play moves,

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Álvaro Begué

I don't think the CPU-GPU communication is what's going to kill this idea.
The latency in actually computing the feed-forward pass of the CNN is going
to be in the order of 0.1 seconds (I am guessing here), which means
finishing the first playout will take many seconds.

So perhaps it would be interesting to do something like this for
correspondence games, but not for regular games.


Álvaro.



On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  wrote:

>   Hi!
>
>   Well, for this to be practical the entire playout would have to be
> executed on the GPU, with no round-trips to the CPU.  That's what my
> email was aimed at.
>
> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
> > Regarding full CNN playouts, I think that problem is that a playout is a
> > long serial process, given 200-300 moves a game. You need to construct
> > planes and transfer them to GPU for each move and read result back (at
> > least with current CNN implementations afaik), so my guess would be that
> > such playout would take time in order of seconds. So there seems to be a
> > tradeoff, CNN playouts are (probably much) better (at "playing better
> > games") than e.g. distribution playouts, but whether this is worth the
> > implied (probably much) lower height of the MC tree is a question.
> >
> > Maybe if you had really a lot of GPUs and very high thinking time, this
> > could be the way.
> >
> > Josef
> >
> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
> >
> > >   Hi!
> > >
> > >   In case someone is looking for a starting point to actually implement
> > > Go rules etc. on GPU, you may find useful:
> > >
> > >
> > > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
> > >
> > >   I wonder if you can easily integrate caffe GPU kernels in another GPU
> > > kernel like this?  But without training, reimplementing the NN could be
> > > pretty straightforward.
> > >
> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> > > > Hello Detlef,
> > > >
> > > > I've got a question regarding CNN-based Go engines I couldn't find
> > > > anything about on this list. As I've been following your posts here,
> I
> > > > thought you might be the right person to ask.
> > > >
> > > > Have you ever tried using the CNN for complete playouts? I know that
> > > > CNNs have been tried for move prediction, immediate scoring and move
> > > > generation to be used in an MC evaluator, but couldn't find anything
> > > > about CNN-based playouts.
> > > >
> > > > It might only be feasible to play out the CNN's first choice move for
> > > > evaluation purposes, but considering how well the performance of
> batch
> > > > sizes scales, especially on GPU-based CNN applications, it might be
> > > > possible to setup something like 10 candidate moves, 10 reply
> > > > candidate moves and then have the CNN play out the first choice move
> > > > for those 100 board positions until the end and then sum up scores
> > > > again for move evaluation (and/or possibly apply some other tried and
> > > > tested methods like minimax). Given that the number of 10 moves is
> > > > supposed to be illustrative rather than representative, other
> > > > configurations of depth and width in position generation and
> > > > evaluation would be possible.
> > > >
> > > > It feels like CNN can provide a very focused, high-quality width in
> > > > move generation, but it might also be possible to apply that quality
> > > > to depth of evaluation.
> > > >
> > > > Any thoughts to share?
> > > >
> > > >
> > > > All the best
> > > >
> > > > Michael
> > > >
> > > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker 
> wrote:
> > > > > -BEGIN PGP SIGNED MESSAGE-
> > > > > Hash: SHA1
> > > > >
> > > > > Hi,
> > > > >
> > > > > as somebody ask I will offer my actual CNN for testing.
> > > > >
> > > > > It has 54% prediction on KGS 6d+ data (which I thought would be
> state
> > > > > of the art when I started training, but it is not anymore:).
> > > > >
> > > > > it has:
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs playing color
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs opponent color
> > > > > Empty points
> > > > > last move
> > > > > second last move
> > > > > third last move
> > > > > forth last move
> > > > >
> > > > > input layers, and it is fully convolutional, so with just editing
> the
> > > > > golast19.prototxt file you can use it for 13x13 as well, as I did
> on
> > > > > last sunday. It was used in November tournament as well.
> > > > >
> > > > > You can find it
> > > > > http://physik.de/CNNlast.tar.gz
> > > > >
> > > > >
> > > > >
> > > > > If you try here some points I like to get discussion:
> > > > >
> > > > > - - it seems to me, that the playouts get much more important with
> such
> > > > > a strong move prediction. Often the move prediction seems better
> the
> > > > > playouts (I use 8000 at the moment against pachi 32000 with about
> 70%
> > > > >

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread David Ongaro

Did everyone forget the fact that stronger playouts don't necessarily lead to 
an better evaluation function? (Yes, that what playouts essential are, a 
dynamic evaluation function.) This is even under the assumption that we can 
reach the same number of playouts per move.


> On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:
> 
> I don't think the CPU-GPU communication is what's going to kill this idea. 
> The latency in actually computing the feed-forward pass of the CNN is going 
> to be in the order of 0.1 seconds (I am guessing here), which means finishing 
> the first playout will take many seconds.
> 
> So perhaps it would be interesting to do something like this for 
> correspondence games, but not for regular games.
> 
> 
> Álvaro.
> 
> 
> 
> On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  > wrote:
>   Hi!
> 
>   Well, for this to be practical the entire playout would have to be
> executed on the GPU, with no round-trips to the CPU.  That's what my
> email was aimed at.
> 
> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
> > Regarding full CNN playouts, I think that problem is that a playout is a
> > long serial process, given 200-300 moves a game. You need to construct
> > planes and transfer them to GPU for each move and read result back (at
> > least with current CNN implementations afaik), so my guess would be that
> > such playout would take time in order of seconds. So there seems to be a
> > tradeoff, CNN playouts are (probably much) better (at "playing better
> > games") than e.g. distribution playouts, but whether this is worth the
> > implied (probably much) lower height of the MC tree is a question.
> >
> > Maybe if you had really a lot of GPUs and very high thinking time, this
> > could be the way.
> >
> > Josef
> >
> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  > > wrote:
> >
> > >   Hi!
> > >
> > >   In case someone is looking for a starting point to actually implement
> > > Go rules etc. on GPU, you may find useful:
> > >
> > >
> > > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html 
> > > 
> > >
> > >   I wonder if you can easily integrate caffe GPU kernels in another GPU
> > > kernel like this?  But without training, reimplementing the NN could be
> > > pretty straightforward.
> > >
> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
> > > > Hello Detlef,
> > > >
> > > > I've got a question regarding CNN-based Go engines I couldn't find
> > > > anything about on this list. As I've been following your posts here, I
> > > > thought you might be the right person to ask.
> > > >
> > > > Have you ever tried using the CNN for complete playouts? I know that
> > > > CNNs have been tried for move prediction, immediate scoring and move
> > > > generation to be used in an MC evaluator, but couldn't find anything
> > > > about CNN-based playouts.
> > > >
> > > > It might only be feasible to play out the CNN's first choice move for
> > > > evaluation purposes, but considering how well the performance of batch
> > > > sizes scales, especially on GPU-based CNN applications, it might be
> > > > possible to setup something like 10 candidate moves, 10 reply
> > > > candidate moves and then have the CNN play out the first choice move
> > > > for those 100 board positions until the end and then sum up scores
> > > > again for move evaluation (and/or possibly apply some other tried and
> > > > tested methods like minimax). Given that the number of 10 moves is
> > > > supposed to be illustrative rather than representative, other
> > > > configurations of depth and width in position generation and
> > > > evaluation would be possible.
> > > >
> > > > It feels like CNN can provide a very focused, high-quality width in
> > > > move generation, but it might also be possible to apply that quality
> > > > to depth of evaluation.
> > > >
> > > > Any thoughts to share?
> > > >
> > > >
> > > > All the best
> > > >
> > > > Michael
> > > >
> > > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker  > > > > wrote:
> > > > > -BEGIN PGP SIGNED MESSAGE-
> > > > > Hash: SHA1
> > > > >
> > > > > Hi,
> > > > >
> > > > > as somebody ask I will offer my actual CNN for testing.
> > > > >
> > > > > It has 54% prediction on KGS 6d+ data (which I thought would be state
> > > > > of the art when I started training, but it is not anymore:).
> > > > >
> > > > > it has:
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs playing color
> > > > > 1
> > > > > 2
> > > > > 3
> > > > >> 4 libs opponent color
> > > > > Empty points
> > > > > last move
> > > > > second last move
> > > > > third last move
> > > > > forth last move
> > > > >
> > > > > input layers, and it is fully convolutional, so with just editing the
> > > > > golast19.prototxt file you can use it for

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Álvaro Begué

Of course whether these "neuro-playouts" are any better than the heavy
playouts currently being used by strong programs is an empirical question.
But I would love to see it answered...



On Tue, Dec 8, 2015 at 1:31 PM, David Ongaro 
wrote:

> Did everyone forget the fact that stronger playouts don't necessarily lead
> to an better evaluation function? (Yes, that what playouts essential are, a
> dynamic evaluation function.) This is even under the assumption that we can
> reach the same number of playouts per move.
>
>
> On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:
>
> I don't think the CPU-GPU communication is what's going to kill this idea.
> The latency in actually computing the feed-forward pass of the CNN is going
> to be in the order of 0.1 seconds (I am guessing here), which means
> finishing the first playout will take many seconds.
>
> So perhaps it would be interesting to do something like this for
> correspondence games, but not for regular games.
>
>
> Álvaro.
>
>
>
> On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis  wrote:
>
>>   Hi!
>>
>>   Well, for this to be practical the entire playout would have to be
>> executed on the GPU, with no round-trips to the CPU.  That's what my
>> email was aimed at.
>>
>> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
>> > Regarding full CNN playouts, I think that problem is that a playout is a
>> > long serial process, given 200-300 moves a game. You need to construct
>> > planes and transfer them to GPU for each move and read result back (at
>> > least with current CNN implementations afaik), so my guess would be that
>> > such playout would take time in order of seconds. So there seems to be a
>> > tradeoff, CNN playouts are (probably much) better (at "playing better
>> > games") than e.g. distribution playouts, but whether this is worth the
>> > implied (probably much) lower height of the MC tree is a question.
>> >
>> > Maybe if you had really a lot of GPUs and very high thinking time, this
>> > could be the way.
>> >
>> > Josef
>> >
>> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis  wrote:
>> >
>> > >   Hi!
>> > >
>> > >   In case someone is looking for a starting point to actually
>> implement
>> > > Go rules etc. on GPU, you may find useful:
>> > >
>> > >
>> > >
>> https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html
>> > >
>> > >   I wonder if you can easily integrate caffe GPU kernels in another
>> GPU
>> > > kernel like this?  But without training, reimplementing the NN could
>> be
>> > > pretty straightforward.
>> > >
>> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
>> > > > Hello Detlef,
>> > > >
>> > > > I've got a question regarding CNN-based Go engines I couldn't find
>> > > > anything about on this list. As I've been following your posts
>> here, I
>> > > > thought you might be the right person to ask.
>> > > >
>> > > > Have you ever tried using the CNN for complete playouts? I know that
>> > > > CNNs have been tried for move prediction, immediate scoring and move
>> > > > generation to be used in an MC evaluator, but couldn't find anything
>> > > > about CNN-based playouts.
>> > > >
>> > > > It might only be feasible to play out the CNN's first choice move
>> for
>> > > > evaluation purposes, but considering how well the performance of
>> batch
>> > > > sizes scales, especially on GPU-based CNN applications, it might be
>> > > > possible to setup something like 10 candidate moves, 10 reply
>> > > > candidate moves and then have the CNN play out the first choice move
>> > > > for those 100 board positions until the end and then sum up scores
>> > > > again for move evaluation (and/or possibly apply some other tried
>> and
>> > > > tested methods like minimax). Given that the number of 10 moves is
>> > > > supposed to be illustrative rather than representative, other
>> > > > configurations of depth and width in position generation and
>> > > > evaluation would be possible.
>> > > >
>> > > > It feels like CNN can provide a very focused, high-quality width in
>> > > > move generation, but it might also be possible to apply that quality
>> > > > to depth of evaluation.
>> > > >
>> > > > Any thoughts to share?
>> > > >
>> > > >
>> > > > All the best
>> > > >
>> > > > Michael
>> > > >
>> > > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker 
>> wrote:
>> > > > > -BEGIN PGP SIGNED MESSAGE-
>> > > > > Hash: SHA1
>> > > > >
>> > > > > Hi,
>> > > > >
>> > > > > as somebody ask I will offer my actual CNN for testing.
>> > > > >
>> > > > > It has 54% prediction on KGS 6d+ data (which I thought would be
>> state
>> > > > > of the art when I started training, but it is not anymore:).
>> > > > >
>> > > > > it has:
>> > > > > 1
>> > > > > 2
>> > > > > 3
>> > > > >> 4 libs playing color
>> > > > > 1
>> > > > > 2
>> > > > > 3
>> > > > >> 4 libs opponent color
>> > > > > Empty points
>> > > > > last move

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

2015-12-08 Thread Hideki Kato

As NNs basically learn the frequency of each move, using the value as 
its probability to be chosen in a simulation could be ok.

Hideki

David Ongaro: <6c2ff906-2a00-45c1-b892-2b14bef35...@hamburg.de>:
>Did everyone forget the fact that stronger playouts don't necessarily lead to 
>an better 
>evaluation function? (Yes, that what playouts essential are, a dynamic 
>evaluation function.) 
>This is even under the assumption that we can reach the same number of 
>playouts per move.
>
>
>> On 08 Dec 2015, at 10:21, Álvaro Begué  wrote:
>> 
>> I don't think the CPU-GPU communication is what's going to kill this idea. 
>> The latency in 
>actually computing the feed-forward pass of the CNN is going to be in the 
>order of 0.1 
>seconds (I am guessing here), which means finishing the first playout will 
>take many seconds.
>> 
>> So perhaps it would be interesting to do something like this for 
>> correspondence games, but 
>not for regular games.
>> 
>> 
>> Álvaro.
>> 
>> 
>> 
>> On Tue, Dec 8, 2015 at 12:03 PM, Petr Baudis > > wrote:
>>   Hi!
>> 
>>   Well, for this to be practical the entire playout would have to be
>> executed on the GPU, with no round-trips to the CPU.  That's what my
>> email was aimed at.
>> 
>> On Tue, Dec 08, 2015 at 04:37:05PM +, Josef Moudrik wrote:
>> > Regarding full CNN playouts, I think that problem is that a playout is a
>> > long serial process, given 200-300 moves a game. You need to construct
>> > planes and transfer them to GPU for each move and read result back (at
>> > least with current CNN implementations afaik), so my guess would be that
>> > such playout would take time in order of seconds. So there seems to be a
>> > tradeoff, CNN playouts are (probably much) better (at "playing better
>> > games") than e.g. distribution playouts, but whether this is worth the
>> > implied (probably much) lower height of the MC tree is a question.
>> >
>> > Maybe if you had really a lot of GPUs and very high thinking time, this
>> > could be the way.
>> >
>> > Josef
>> >
>> > On Tue, Dec 8, 2015 at 5:17 PM Petr Baudis > > > wrote:
>> >
>> > >   Hi!
>> > >
>> > >   In case someone is looking for a starting point to actually implement
>> > > Go rules etc. on GPU, you may find useful:
>> > >
>> > >
>> > > https://www.mail-archive.com/computer-go@computer-go.org/msg12485.html 
>
>> > >
>> > >   I wonder if you can easily integrate caffe GPU kernels in another GPU
>> > > kernel like this?  But without training, reimplementing the NN could be
>> > > pretty straightforward.
>> > >
>> > > On Tue, Dec 08, 2015 at 04:53:14PM +0100, Michael Markefka wrote:
>> > > > Hello Detlef,
>> > > >
>> > > > I've got a question regarding CNN-based Go engines I couldn't find
>> > > > anything about on this list. As I've been following your posts here, I
>> > > > thought you might be the right person to ask.
>> > > >
>> > > > Have you ever tried using the CNN for complete playouts? I know that
>> > > > CNNs have been tried for move prediction, immediate scoring and move
>> > > > generation to be used in an MC evaluator, but couldn't find anything
>> > > > about CNN-based playouts.
>> > > >
>> > > > It might only be feasible to play out the CNN's first choice move for
>> > > > evaluation purposes, but considering how well the performance of batch
>> > > > sizes scales, especially on GPU-based CNN applications, it might be
>> > > > possible to setup something like 10 candidate moves, 10 reply
>> > > > candidate moves and then have the CNN play out the first choice move
>> > > > for those 100 board positions until the end and then sum up scores
>> > > > again for move evaluation (and/or possibly apply some other tried and
>> > > > tested methods like minimax). Given that the number of 10 moves is
>> > > > supposed to be illustrative rather than representative, other
>> > > > configurations of depth and width in position generation and
>> > > > evaluation would be possible.
>> > > >
>> > > > It feels like CNN can provide a very focused, high-quality width in
>> > > > move generation, but it might also be possible to apply that quality
>> > > > to depth of evaluation.
>> > > >
>> > > > Any thoughts to share?
>> > > >
>> > > >
>> > > > All the best
>> > > >
>> > > > Michael
>> > > >
>> > > > On Tue, Dec 8, 2015 at 4:13 PM, Detlef Schmicker > wrote:
>> > > > > -BEGIN PGP SIGNED MESSAGE-
>> > > > > Hash: SHA1
>> > > > >
>> > > > > Hi,
>> > > > >
>> > > > > as somebody ask I will offer my actual CNN for testing.
>> > > > >
>> > > > > It has 54% prediction on KGS 6d+ data (which I thought would be state
>> > > > > of the art when I started training, but it is not anymore:).
>> > > > >
>> > > > > it has:
>> > > > > 1
>> > > > > 2
>> > > > > 3
>> > > > >> 4 libs playing color
>> > > > > 1
>> > > > > 2

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

[Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

15 matches

Site Navigation

Mail list logo

Footer information