ymgaq / aq Goto Github PK

In last game during test found that AQ commit suicide during filling dame.
Please check last 2 moves in next game:
http://www.yss-aya.com/cgos/viewer.cgi?19x19/SGF/2017/10/16/297028.sgf

If any kind of such situations (suicide/missed ladder/etc.) will be found in future - are you interested to have this issues to be reported?

Hello!
If it possible - could win-rate be added to putput? Or 'value' is 100% representation if gerenal win-date?
Accodring next example it can't be it.

Also asking because other engines provide several types of win-rate.

Could you clarify what exactly the meaning of next vars:
value | roll | prob

For example engine has next output:
|move|count |value|roll |prob |depth|
|C15 | 11300| 16.3| 38.2| 4.7 | 15|
|D15 | 6293| 15.5| 38.1 | 29.3 | 24|

prob is higher for D15 but engine played C15.
value is 15.5-16.3 but it's only 10 move of standart opening, so it can't be win-rate.

Attached screenshot.

Thank you!

Does AQ support handicap game?

In my test, AQ show illegal move in handicap game.

And when I change Komi, only roll out ratio change , Value count doesn change. Is it correct?

Linux all.
On all PC without video cards with support of last CUDA versions AQ doesn't work.

Is it possible to make AQ Linux release compatible with most PCs?

Thanks in advance.

The author releases AQ2.0.2, which is not as good as the version tested on CGOS?
It's hard to beat zen6
my hardware i7 6700K + GTX 1080

hi, ymgaq,
can give some tips, thanks.

Hello all,

I have the latest AQ v2.0.3.1 and Sabaki v0.31.4.
Can someone help me how to add the engine to Sabaki?
I downloaded the AQ_win.zip file and extracted it.
I downloaded Sabaki. Then I go to: File -> Preferences -> Engines -> Add -> added the AQ.exe from the previous .zip
Then I go to: Engines -> Attach -> and put the AQ in either Black or White field.

• If I put it in Black then it 'thinks' indefinitely, changing the cursor to a circle but it doesn't use any CPU.
• If I put it in White, I play a move as Black then it passes the turn every time.

edit:

I tried GoGui as well and didn't manage to start it. I copied the AQ folder in GoGui's one. Then I went

Program - New Program - Setting command as 'C:\Games\GoGui\AQ\AQ.exe', Working directory as 'C:\Games\GoGui\AQ' (both with the last ' \ ' and without it) and when I press 'Ok' I got the error:

'The go program terminated unexpectedly. The Go program did not respond to any commands. One possible reason is that the commend for executing the program in GTP mode was not correct.'

I have NVidia 670 with the latest drivers updated.

What am I doing wrong?

I want to run the engine, not caring about the GUI.

Thank you

John

the problem exists, how to fix it?

I saw AQ-2.1.1-4t1g were number 5 at cgos.

I have compiled an AQ.exe file with Release and x64 option in visual studio 2015(windows 7 64bit).
But a compiled AQ.exe's size is very smaller than a released AQ.exe's size(263,483KB) and naturally failed to execute.

Help me~

Hello !
Is there a way to make AQ go into emergency mode when there's 60 seconds left (instead of 15 sec.) I think it would solve my problem.
I have a much better computer than before, but my bot is losing more and more Elo points on CGOS , because of the losses on time, (and probably bad configuration, too)
My Internet connexion is average.
Thanks for this beautiful program !

In this corner, IsSelfAtariNadake(S19) is true as expected. What I wonder about is the fact that IsEyeShape(S19) is true. Is this intended? If there was a black stone at T18, IsEyeShape(19) is false as it should be.
In fact, in this corner white should play S19 to kill black, but AQ excludes S19 as a candidate child node because it is considered an eyeshape.

Hi,

I followed the instructions to install gogui on my Ubuntu 14.04 OS but it didn't work (sudo bash ./install.sh -p /usr/local -j /usr/lib/jvm/java-6-sun to execute the installation script *** I replaced by java-9 cause it's the one i'm using) but there were some errors).

Is there any other softwares than Gogui to run AQ? I tried with Sabaki but when I selected the engine of AQ in the repertory it didn't work.

Maybe I should run it on Windows? But I think on Linux it works better, can calculate more simulations because Linux uses less ressources, I am wrong? At least with Leela it works as it.

Thanks for the help.

Is Aq support cuda9?

Can I run AQ on a CPU only computer?

Hi,

Thank you for the nice program!

I have a question about how AQ plays. It seems to me that when the value goes above 90% (AQ predicts it is very likely to win), AQ will start to play instantly. It will think for a while at a certain move and then make multiple instant follow-up moves before it thinks for a while again. I was wondering if it was a behavior by design or accidental. The problem of this behavior is that I saw AQ made a quick but wrong move that directly ruined its winning game. At the early stage of a game, it seems that, although AQ can sometimes make an instant move, it happens much less frequently -- AQ usually uses the full byoyomi before making a move.

Thanks,

iteachcs

I have seen some impressive results from the new AQ version. Just wondering if I can get the Windows binary.

Thank you.

(;FF[4]CA[UTF-8]AP[GoGui:1.4.5]
KM[7.5]RE[B+Resign]
;B[cd];W[pp];B[dq];W[pd];B[ec];W[do];B[co];W[cn];B[cp];W[ce]
;B[de];W[cf];B[df];W[bd];B[cg];W[cc];B[dd];W[bg];B[bh];W[bf]
;B[ch];W[dm];B[bb];W[bc];B[cb];W[ae];B[en];W[eo];B[nc];W[qf]
;B[dn];W[dc];B[fn];W[lc];B[db];W[fc];B[fb];W[ed];B[eb];W[jq]
;B[hq];W[qn];B[nq];W[oq];B[np];W[jo];B[kp];W[jp];B[nm];W[dg]
;B[fe];W[eg];B[ff];W[dj];B[bj];W[fg];B[ei];W[di];B[dh];W[eh]
;B[fi];W[gg];B[ck];W[dk];B[cl];W[fk];B[dl];W[fd];B[hf];W[hd]
;B[he];W[id];B[hi];W[hg];B[je];W[jg];B[jf];W[jd];B[le];W[kg]
;B[ne];W[md];B[me];W[nd];B[od];W[oc];B[oe];W[pc];B[ji];W[lh]
;B[nh];W[lj];B[jl];W[ni];B[oi];W[oh];B[og];W[ph];B[mi];W[ng]
;B[nj];W[mg];B[pe];W[of];B[qe];W[rc];B[li];W[rg];B[rf];W[qg]
;B[rd];W[ld];B[rb];W[qc];B[sc];W[qb];B[nb];W[ra];B[sb];W[ob]
;B[ho];W[ih];B[ii];W[hp];B[gp];W[hr];B[gr];W[iq];B[ip];W[lp]
;B[lr];W[nn];B[po];W[on];B[op];W[pq];B[qo];W[pn];B[oo];W[mo]
;B[rq];W[qj];B[ln];W[lq];B[mr];W[io];B[mn];W[hn];B[go];W[no]
;B[lo];W[ko];B[kr];W[kq];B[ql];W[pl];B[pk];W[qk];B[mp];W[jm]
;B[ol];W[rl];B[kn];W[il];B[kl];W[ik];B[gl];W[gk];B[jn];W[in]
;B[kj];W[el];B[em];W[gn];B[fl];W[ek];B[fo];W[hl];B[gm];W[or]
;B[rn];W[rm];B[ro];W[jr];B[hs];W[is];B[gs];W[qr];B[rr];W[km]
;B[lm];W[jk];B[kk];W[nr];B[mq];W[ps];B[ns];W[ah];B[ag];W[ac]
;B[hj];W[hk];B[hb];W[hp];B[ir];W[bi];B[ai];W[ci];B[ah];W[hr]
;B[ip];W[qp];B[rp];W[hp];B[ir];W[gq];B[fq];W[hr];B[ip];W[hm]
;B[fm];W[hp];B[ir];W[cj];B[bk];W[cm];B[bm];W[hr];B[ip];W[mb]
;B[jb];W[hp];B[ir];W[bn];B[an];W[hr];B[ip];W[jc];B[kb];W[hp]
;B[ir];W[aj];B[ak];W[hr];B[gq];W[qm];B[pi];W[qi];B[pj];W[ok]
;B[nk];W[om];B[oj];W[ib];B[ia];W[hc];B[gc];W[gd];B[gb];W[ab]
;B[ba];W[ms];B[ls];W[rs];B[qq];W[sn];B[so];W[sm];B[ms];W[os]
;B[sr];W[sq];B[ss];W[ml];B[nl];W[ks];B[qs];W[bl];B[al];W[rs]
;B[js];W[bo];B[bp];W[ks];B[qs];W[ao];B[ap];W[rs];B[js];W[gi]
;B[gj];W[ks];B[qs];W[fj];B[gh];W[rs];B[js];W[ir];B[qs];W[af]
;B[aj];W[rs];B[jh];W[fh];B[hh];W[ig];B[kh];W[ks];B[qs];W[mm]
;B[im];W[rs];B[js];W[jm];B[qs];W[km];B[ll];W[rs];B[mh];W[lg]
;B[qs];W[kd];B[pr];W[ha];B[ic];W[jj];B[ib];W[ma];B[lb];W[la]
;B[ka];W[ks];B[gi];W[])

Hi, could you consider to add this in next version.
It would be convenient when doing engines sparrings with batch files.
Also "playouts" parameter should be useful.

AQ can be attached by GoGui but always collapses when it plays a move. Here is the GTP shell's info

genmove w
thinking...
AQ: malloc.c:2405: sysmalloc: Assertion `(old_top == initial_top (av) && old_size == 0) || ((unsigned long) (old_size) >= MINSIZE && prev_inuse (old_top) && ((unsigned long) old_end & (pagesize - 1)) == 0)' failed.

Besides, CUDA 9.0& JAVA 8.0 have been installed correctly in my PC.

AQ is very strong.
AQ always wins Leela0.11.
I run AQ in Ubuntu terminal, but I can't run AQ in Gogui or sabaki.
How can I run AQ in Gogui?

When attaching AQ to Sabaki in the middle of the game, the order of the game is loaded differently from the actual one, so that AQ suggest "Ko postion" as in the picture. (The last move is W to J10)

Hi.

I suffer from "C1013 too many include files // File : tensor // Line :1" error during compiling.

How did you solve it?

I am suffering from a build problem for several weeks.

Plz someone upload AQ's CMakeLists.txt file for Visual Studio 2015 and Cmake.

What should be time settings for playing on CGOS.
I've tried two variants:
-main time[sec] =900
-byoyomi[sec] =0
-time controll =on
or
-time controll =off
in both cases AQ loosing sometimes by time

I've tried to analyze game with AQ. To do so I did next:

• open game with Sabaki viewer with support GTP
• scrool to position
• use attach engine
  But after thinking AQ just resign. In any position when game is even.
  Tried default config and with japanese rules.
  Tried -never resign =on but AQ still resign.

When I use Sabaki to play against AQ - it works fine (well haven't tested deep yet but at brief look it works). So assume may be the issue with loading existing game

Attached logs, game and screenshot. Added txt to sgf files -because github doesn't support sgf.

1.txt
game_1.sgf.txt
1.sgf.txt

From the AQ battle LOG found
AQ winning percentage is usually the leading
Middle game start winning percentage down

Analysis mode means that we can see what AQ think at the moment. Maybe you can update it once per second. I expect earnestly about analysis mode because sometime it is hard to tell whether aq meets a bug or aq finds a tesjui(手筋). Would you consider to make it?

Thank you for publishing great go program!
I was able to compile it using visual studio. However when I run the program I get a stack overflow exception. Is there a solution for it?

The paper introduces AlphaGo Zero, the latest evolution of AlphaGo, the first computer program to defeat a world champion at the ancient Chinese game of Go. Zero is even more powerful and is arguably the strongest Go player in history.
Previous versions of AlphaGo initially trained on thousands of human amateur and professional games to learn how to play Go. AlphaGo Zero skips this step and learns to play simply by playing games against itself, starting from completely random play. In doing so, it quickly surpassed human level of play and defeated the previously published champion-defeating version of AlphaGo by 100 games to 0.

What is possible? What is impossible?
How is ur opinion?

Do you have a any plan about build your AI using OpenCL?
I heard that you can build the OpenCL version using tensorflow-cl.
I hope that the OpenCL version will be distributed with the Windows version.
Thx.

1. Does AQ support simultaneous multi GPU?(not just set gpu)

2. Can u explain brief explanation about AQ command? (ex..worker)

3. Is there option for pondering( On or Off)?

Really thx for your efforts.

Hello,

Many friends of mine have computers with strong CPUs like i7, Ryzen or even Xeon but non-Nvidia GPU.

Is it possible to create a stronger version of AQ-mini by removing the thread = 2 only limit?

When I was trying to run AQ on GoGui,
After I ran the New Program, an error message popped up stating:

"The Go program did not respond to any commands. One possible reason is that the command for executing the program in GTP mode was not correct."

what should I do?

Hi ymgaq,
I try to build the AQ, system is ubuntu 16.04, bazel config with cuda, cuda 8.0 cudnn 6.0, seems the final link failed,
the way I do it is: copy AQ-master into tensorflow-master/tensorflow/examples/ dir, then run
./configure
bazel build /tensorflow/examples/AQ-master/...
in tensorflow-master dir

Thanks.
following are the last couple lines of output:

validate.cc:(.text._ZN10tensorflow5graph16ValidateGraphDefERKNS_8GraphDefERKNS_19OpRegistryInterfaceE+0x92): undefined reference to tensorflow::ValidateNodeDef(tensorflow::NodeDef const&, tensorflow::OpDef const&)' validate.cc:(.text._ZN10tensorflow5graph16ValidateGraphDefERKNS_8GraphDefERKNS_19OpRegistryInterfaceE+0xaa): undefined reference to tensorflow::CheckOpDeprecation(tensorflow::OpDef const&, int)'
bazel-out/local_linux-opt/bin/tensorflow/core/util/tensor_bundle/libnaming.a(naming.o): In function tensorflow::MetaFilename[abi:cxx11](tensorflow::StringPiece)': naming.cc:(.text._ZN10tensorflow12MetaFilenameB5cxx11ENS_11StringPieceE+0x19): undefined reference to tensorflow::strings::Printf[abi:cxx11](char const*, ...)'
bazel-out/local_linux-opt/bin/tensorflow/core/util/tensor_bundle/libnaming.a(naming.o): In function tensorflow::DataFilename[abi:cxx11](tensorflow::StringPiece, int, int)': naming.cc:(.text._ZN10tensorflow12DataFilenameB5cxx11ENS_11StringPieceEii+0x1f): undefined reference to tensorflow::strings::Printf[abi:cxx11](char const*, ...)'
collect2: error: ld returned 1 exit status
Target //tensorflow/examples/AQ-master/src:AQ failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 6393.312s, Critical Path: 911.08s
FAILED: Build did NOT complete successfully

I'm a student of Intelligence science as well as an Amateur 5 dan Go player. I come here for reading and studying your awesome works to improve my coding skill, sometimes playing with it to learn Go. But recently, when I play Go on platforms like Tygem, I can find some people, obviously, using programs to raise there level (I'm pretty sure about that because i've tested a lot of programs, like Leela or Rn), which is unfair and gives really bad experince to there opponents. I understand that releasing a Windows version will help promoting AQ, but too many people download the program only for using it on Tygem, and for win other people.Is that your original intention? Or do you think that's acceptable?
Thanks.

I noticed a situation that AQ wrongly recognizes as seki.

Obviously black is dead, but IsSeki() recognizes D3 and E2 as seki.

This has great impact on rollout. Because in rollout, either by PlayoutLGR or PlayoutRandom(), a detected seki point will never be filled, so AQ will never realize that black is dead, outputing incorrect result.

hi ymgap
i'm confused about this option
thanks for your work and time

Thx for making great AI go program.

I have some questions.

1. Is there a methods to extend tree search, or move count?
   If I raise up max_node_limit, is it effective?(and max node limit = 65537. 2^16=65536. What is +1's meaning?)

2. AQ use max about 10 Giga memory. Can I raise up memory usage?

3. What is AQ's maximum CPU thread counts?

4. Could u open AQ training methods or training tool in detail?

....... Sorry for silly question.

Did anyone try to compile AQ on Max OS?

I tried tensorflow 1.2 & 1.4 to compile AQ, and neither of them worked.
But I can compile the example you gave (matmul.cpp) and it works.

The following is the error message.

1> Generating Code...
1>search.obj : error LNK2019: unresolved external symbol "public: __cdecl google::protobuf::internal::LogMessage::LogMessage(enum google::protobuf::LogLevel,char const *,int)" (??0LogMessage@internal@protobuf@google@@qeaa@W4LogLevel@23@PEBDH@Z) referenced in function "protected: class tensorflow::NodeDef * __cdecl google::protobuf::internal::RepeatedPtrFieldBase::Mutable<class google::protobuf::RepeatedPtrField::TypeHandler>(int)" (??$Mutable@VTypeHandler@?$RepeatedPtrField@VNodeDef@tensorflow@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@IEAAPEAVNodeDef@tensorflow@@h@Z)
1>search.obj : error LNK2019: unresolved external symbol "public: __cdecl google::protobuf::internal::LogMessage::~LogMessage(void)" (??1LogMessage@internal@protobuf@google@@qeaa@XZ) referenced in function "protected: class tensorflow::NodeDef * __cdecl google::protobuf::internal::RepeatedPtrFieldBase::Mutable<class google::protobuf::RepeatedPtrField::TypeHandler>(int)" (??$Mutable@VTypeHandler@?$RepeatedPtrField@VNodeDef@tensorflow@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@IEAAPEAVNodeDef@tensorflow@@h@Z)
1>search.obj : error LNK2019: unresolved external symbol "public: class google::protobuf::internal::LogMessage & __cdecl google::protobuf::internal::LogMessage::operator<<(char const *)" (??6LogMessage@internal@protobuf@google@@QEAAAEAV0123@PEBD@Z) referenced in function "protected: class tensorflow::NodeDef * __cdecl google::protobuf::internal::RepeatedPtrFieldBase::Mutable<class google::protobuf::RepeatedPtrField::TypeHandler>(int)" (??$Mutable@VTypeHandler@?$RepeatedPtrField@VNodeDef@tensorflow@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@IEAAPEAVNodeDef@tensorflow@@h@Z)
1>search.obj : error LNK2019: unresolved external symbol "public: void __cdecl google::protobuf::internal::LogFinisher::operator=(class google::protobuf::internal::LogMessage &)" (??4LogFinisher@internal@protobuf@google@@QEAAXAEAVLogMessage@123@@z) referenced in function "protected: class tensorflow::NodeDef * __cdecl google::protobuf::internal::RepeatedPtrFieldBase::Mutable<class google::protobuf::RepeatedPtrField::TypeHandler>(int)" (??$Mutable@VTypeHandler@?$RepeatedPtrField@VNodeDef@tensorflow@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@IEAAPEAVNodeDef@tensorflow@@h@Z)
1>search.obj : error LNK2019: unresolved external symbol "private: void __cdecl google::protobuf::Arena::AddListNode(void ,void (__cdecl)(void *))" (?AddListNode@Arena@protobuf@google@@AEAAXPEAXP6AX0@Z@Z) referenced in function "public: void __cdecl google::protobuf::Arena::Own<class std::basic_string<char,struct std::char_traits,class std::allocator > >(class std::basic_string<char,struct std::char_traits,class std::allocator > *)" (??$Own@V?$basic_string@DU?$char_traits@D@std@@v?$allocator@D@2@@std@@@arena@protobuf@google@@QEAAXPEAV?$basic_string@DU?$char_traits@D@std@@v?$allocator@D@2@@std@@@z)
1>search.obj : error LNK2001: unresolved external symbol "class google::protobuf::internal::ExplicitlyConstructed<class std::basic_string<char,struct std::char_traits,class std::allocator > > google::protobuf::internal::fixed_address_empty_string" (?fixed_address_empty_string@internal@protobuf@google@@3v?$ExplicitlyConstructed@V?$basic_string@DU?$char_traits@D@std@@v?$allocator@D@2@@std@@@123@A)
1>D:\AQ2031\win\x64\Release\aqgo.exe : fatal error LNK1120: 6 unresolved externals
========== Rebuild All: 0 succeeded, 1 failed, 0 skipped ==========

I think it is owing to the tensorflow version, so which version of tensorflow did you use to compile AQ?

Have the chance?

i loaded sgf .
and i played game with AQ in tabaki.

but,....

Why am I getting an error?

Game is completely won, opponend passes, AQ plays on.

1. When I raise up node_limit to 132109, It works fine.
   But 262147failed. Is max node_limit is around 2^17?

2. When using 132109, memory usage is still about 10giga.
   How can I raise up memory usage? What do I modify?

3. What,s the meaning of expand_cnt?
   To expand search ,Should I go up expand_cnt ? Or go down?

Thank you.

1. Is it number of extended board? Or just outer number of 'v'? In color[v] outer boundary's meaning?

2. Aq has severe flower five shape life death bug.

3. In seki you need to add more condition(ex lib_cnt = 3 or 4)

Hello,

./AQ: error while loading shared libraries: libcudart.so.8.0: cannot open shared object file: No such file or directory

Is it possible to have a pre compiled version for Ubuntu without GPU support ? I do not want to install CUDA Nvidia on my Ubuntu Linux.

Thank you.

Regards...