Open AI Championship

Post #25

Finals: before start

Project

Sandbox

Round 1

Round 2

Finals

Recent comments

12 december 13:31: alex3d wrote comment on post Some more good news ;)
12 december 12:07: newpuk wrote comment on post Finals
12 december 09:50: leloy wrote comment on post Finals
11 december 23:48: VexfxjC wrote comment on post Some more good news ;)
11 december 23:38: alkozel wrote comment on post Some more good news ;)
11 december 18:26: Jeners wrote comment on post Some more good news ;)
11 december 16:59: VexfxjC wrote comment on post Some more good news ;)
11 december 16:57: VexfxjC wrote comment on post Some more good news ;)
11 december 16:08: OREZ wrote comment on post Some more good news ;)
11 december 11:58: Eugene713 wrote comment on post Some more good news ;)

VK Group

Telegram chat

Some more good news ;)

In the first half of Round 2, approximately the same number of waves were tested as in each half of the previous tournament stage. Of course, you expected more, so as we.

At the beginning of Round 2, there were some failures in the testing system. We quickly fixed them, but the process did not accelerate much. The main reasons for this are:

  1. Complicating the rules of the game, compared to Round 1, and in particular adding facilities and the ability to produce new units.
  2. Selection of strategies. Top strategies, of course, produce much more calculations and more efficiently utilize the CPU resources provided to them. Round 2 involves a significant number of strategies that operate at the time limit, and sometimes exceed it.
  3. Small number of games per single wave.

The last two reasons reinforce each other. It often turns out that games with long-running strategies remain at the end of the wave. The most participants have to wait until one or two games are tested.

We are working on solving the problem in two directions at once. First, the number of testing machines will be further increased (compared to the beginning of Round 2). Secondly, we will immediately create several waves of games. This will greatly reduce the amount of resources wasted. This solution has a disadvantage: during the testing of these batch waves the standings table can be not relevant. However, we consider this a lesser evil than a small number of games.



icxon

icxon

Saint Petersburg

At the stage of introduction of “pay for tank battle” it was easier to shorten the game up to 10K-12K ticks. That would solve all of these cases including. Although stating that performance is not a problem.

Dec 9, 2017 8:09:08 PM
mixei4

mixei4

КГУ

“Often it turns out that the game of long-running strategies remain at the end of the wave.”

Yes. But sometimes it happens that the game in the top of the queue, but is tested for 20 minutes. The system freezes, or testing one game for 20 minutes is this normal?

Dec 9, 2017 8:27:40 PM
Adler

Adler

Dzyarzhynsk

>This solution has a drawback: during testing, the wave table with the rating of participants may be very different from reality.

offer to make then the website page in which you will complete a log of tested games. FORTS like this: take content result.txt(which creates a local-runner ohms), then create a json string like this format:

{  “gameid”:9000,  “palyers”:[“user_a”,“user_b”],  “result.txt”: “OK 1283\nSEED 22904832135050\n2 388 CRASHED\n1 396 CRASHED\n” } 
then add in the end of the file . do fflush and so for each game when the event comes that the game is tested

well, to this file(or copy of one minute ago) was available via http

we did it as something to do resprim :)

Dec 9, 2017 8:46:36 PM
JIo6cTep

JIo6cTep

Samara

Adler

Adler

Dzyarzhynsk

This solution has a drawback: during testing, the wave table with the rating of participants may be very different from reality. offer to make then the website page in which you will complete a log of tested games. FORTS like this: take content result.txt(which creates a local-runner ohms), then create a json string like this format: …

It does mean that if, for example, has 3 waves, it may be a situation in which someone has played all 3 games and already got 6 points, and someone played the 0. Accordingly, the staging table can not be trusted. And the result you get will not be able in any form.

Dec 9, 2017 9:51:03 PM
VexfxjC

VexfxjC

БГТУ «Военмех»

JIo6cTep

JIo6cTep

Samara

It does mean that if, for example, has 3 waves, it may be a situation in which someone has played all 3 games and already got 6 points, and someone played the 0. Accordingly, the staging table can not be trusted. And the result you get will not be able in any form.

so someone will be “significantly” more games for more precise position in the table

Dec 9, 2017 9:54:44 PM
JIo6cTep

JIo6cTep

Samara

VexfxjC

VexfxjC

БГТУ «Военмех»

so someone will be “significantly” more games for more precise position in the table

No, the final number of games for all participants will be the same. It is now: created X-waves. Wait for them to play X matches, one from each wave. Create X waves. Etc.

Dec 9, 2017 9:58:39 PM
Adler

Adler

Dzyarzhynsk

I’m not hard to write code which is all to steam, considering how many games played, how many waves fully tested, make a staging table with only those games that belong fully tested waves.

Dec 9, 2017 10:00:15 PM
Adler

Adler

Dzyarzhynsk

corrected format:

{  “wave_id”:10,  “gameid”:9000,  “players”:[“user_a”,“user_b”],  “result.txt”: “OK 1283\nSEED 22904832135050\n2 388 CRASHED\n1 396 CRASHED\n” }

Dec 9, 2017 10:19:30 PM
tyamgin

tyamgin

Simferopol

Adler

Adler

Dzyarzhynsk

corrected format: { “wave_id”:10, “gameid”:9000, “players”:[“user_a”,“user_b”], “result.txt”: “OK 1283\nSEED 22904832135050\n2 388 CRASHED\n1 396 CRASHED\n” }

Easier already to correct the label to show only fully played the waves, than to fence API, which will use half human.

Dec 9, 2017 10:28:59 PM
mixei4

mixei4

КГУ

Rating is now charged in the games in order. If you have created three games and tested 1 and 3, then the rating will be counted only for the first game, and for the third count only after the second. That is the biggest problems with a label is not due to a large number of waves. At least not more than now.

The problem is that tested can be a lot of games, and the rating is counted only half.

Dec 9, 2017 10:37:02 PM
selya_

selya_

Гимназия №1

What’s the difference? Fun, of course, to look at your rating. But to do something still nothing you can) I looked at my rating and be sad. After all, some identical to my strategy, which, moreover, is much worse than going with a winrate of 90%+, and my digs due to the fact that often catches distributed strategy) is Better then do not look at the rating)

Dec 9, 2017 11:04:43 PM
VexfxjC

VexfxjC

БГТУ «Военмех»

selya_

selya_

Гимназия №1

What’s the difference? Fun, of course, to look at your rating. But to do something still nothing you can) I looked at my rating and be sad. After all, some identical to my strategy, which, moreover, is much worse than going with a winrate of 90%+, and my digs due to the fact that often catches distributed strategy) is Better then do not look at the rating)

I agree ! here all the will of randomness )

Dec 9, 2017 11:07:18 PM
Adler

Adler

Dzyarzhynsk

Dec 10, 2017 12:43:00 AM
Jeners

Jeners

Moscow

In the game to introduce a priori the function, if the strategy for the tick neither of which can do (RemainingActionCooldownTicks <> 0) just skipat move as such

Dec 10, 2017 12:01:24 PM
dbf

dbf

A

But if the strategy of counting something every tick, without making moves?

Dec 10, 2017 12:03:10 PM
Jeners

Jeners

Moscow

Although, there are some nuances, to update the provisions of troops -_- come even at the time when the strategy has already spent all of the action.. anyways I XS)

Dec 10, 2017 12:04:01 PM
Jeners

Jeners

Moscow

dbf

dbf

A

But if the strategy of counting something every tick, without making moves?

Well so if she thinks something or what not doing, hence the action she had left. And skipat is not necessary essesno. Another issue is updating the situation of the troops on the map, if it was a core API function, it would be possible not to bother, the Update would have happened in the API. And I would BLACULA action and calculating strategy with the inability to make any progress.

Dec 10, 2017 12:06:44 PM
MikeWazowski

MikeWazowski

Moscow

Jeners

Jeners

Moscow

In the game to introduce a priori the function, if the strategy for the tick neither of which can do (RemainingActionCooldownTicks <> 0) just skipat move as such

it seems there is a limit on CPU time within the same tick, and thus can be strategies that have cheated, even if you have exceeded the current limit of moves

Dec 10, 2017 12:09:34 PM
Jeners

Jeners

Moscow

MikeWazowski

MikeWazowski

Moscow

it seems there is a limit on CPU time within the same tick, and thus can be strategies that have cheated, even if you have exceeded the current limit of moves

Um that’s not the question if the strategy is at the moment no free actions, then it is considered as if not what you need. Because these calculations are not applicable to the game, it’s like water in a mortar to pound

Dec 10, 2017 12:12:12 PM
MikeWazowski

MikeWazowski

Moscow

Jeners

Jeners

Moscow

Um that’s not the question if the strategy is at the moment no free actions, then it is considered as if not what you need. Because these calculations are not applicable to the game, it’s like water in a mortar to pound

in the presence of fog of war for a very clever strategies that count will always be …

Dec 10, 2017 12:20:54 PM
Jeners

Jeners

Moscow

MikeWazowski

MikeWazowski

Moscow

in the presence of fog of war for a very clever strategies that count will always be …

-_- think whatever you like, but in any case, these calculations are not applicable to the TEC while the strategy does not appear free action, and thus a particularly clever strategy, just will consider it several times in succession until a free action. The picture of the battlefield is changing every tick, but to influence the situation is possible only when there is a free action

Dec 10, 2017 12:25:13 PM
Jeners

Jeners

Moscow

So IMHO, the decision is Do not send data strategy with the tick until the player has no available actions ie a tick for the strategy will not be changed until the strategy is not going to free up action points. For example, the strategy in the first 12 ticks made actions. information about the next 48 ticks strategy just does not receive (tick to stretegii at this point is always equal to =12 ). But after that, she sent all the information about the changes of the world. Something like that

Dec 10, 2017 12:30:10 PM
Siont

Siont

УГАТУ

Jeners

Jeners

Moscow

-_- think whatever you like, but in any case, these calculations are not applicable to the TEC while the strategy does not appear free action, and thus a particularly clever strategy, just will consider it several times in succession until a free action. The picture of the battlefield is changing every tick, but to influence the situation is possible only when there is a free action

For example clustering is performed regardless of the presence of action. The calculation of the movement vector of enemy clusters, just convenient to do it while there are no points of action.

Dec 10, 2017 12:30:57 PM
Jeners

Jeners

Moscow

Jeners

Jeners

Moscow

So IMHO, the decision is Do not send data strategy with the tick until the player has no available actions ie a tick for the strategy will not be changed until the strategy is not going to free up action points. For example, the strategy in the first 12 ticks made actions. information about the next 48 ticks strategy just does not receive (tick to stretegii at this point is always equal to =12 ). But after that, she sent all the information about the changes of the world. Something like that

Although there are some nuances, such as the calculation of the vector of the enemy baguet -_- in General, everything is controversial

Dec 10, 2017 12:32:02 PM
mixei4

mixei4

КГУ

Jeners

Jeners

Moscow

Although there are some nuances, such as the calculation of the vector of the enemy baguet -_- in General, everything is controversial

It is not debatable. You can’t do that. Not least because over the last 48 ticks the opponent can get out of the fog and disappearing back, for example.

Dec 10, 2017 7:05:34 PM
Jeners

Jeners

Moscow

mixei4

mixei4

КГУ

It is not debatable. You can’t do that. Not least because over the last 48 ticks the opponent can get out of the fog and disappearing back, for example.

I agree, and his last position would be good to fix )

Dec 10, 2017 7:13:27 PM
Eugene713

Eugene713

МГГУ, МГУ ВМИК

Good day! Then the confusion in the sandbox, I think: to reset the rating I pressed and went to the opponent. http://russianaicup.ru/game/view/226691

Dec 11, 2017 11:36:51 AM
Siont

Siont

УГАТУ

Eugene713

Eugene713

МГГУ, МГУ ВМИК

Good day! Then the confusion in the sandbox, I think: to reset the rating I pressed and went to the opponent. http://russianaicup.ru/game/view/226691

Judging by his page, he is also pressed

Dec 11, 2017 11:47:50 AM
Eugene713

Eugene713

МГГУ, МГУ ВМИК

Siont

Siont

УГАТУ

Judging by his page, he is also pressed

Really… Not once the blue line appears, that was fun)

Dec 11, 2017 11:58:33 AM
OREZ

OREZ

Kazan’

VexfxjC

VexfxjC

БГТУ «Военмех»

I agree ! here all the will of randomness )

Oh, this random, you swept in the 2nd round, and 1st, and in the sandbox. Yes, it is just randomness.

Dec 11, 2017 4:08:24 PM
VexfxjC

VexfxjC

БГТУ «Военмех»

OREZ

OREZ

Kazan’

Oh, this random, you swept in the 2nd round, and 1st, and in the sandbox. Yes, it is just randomness.

I myself, just blame myself ))) in fact wrote the code in the last hours

Dec 11, 2017 4:57:43 PM
VexfxjC

VexfxjC

БГТУ «Военмех»

OREZ

OREZ

Kazan’

Oh, this random, you swept in the 2nd round, and 1st, and in the sandbox. Yes, it is just randomness.

although the games are all the same little… the 7 participants 1 did not fight among themselves, the winner in fact is not defined

Dec 11, 2017 4:59:46 PM
Jeners

Jeners

Moscow

VexfxjC

VexfxjC

БГТУ «Военмех»

although the games are all the same little… the 7 participants 1 did not fight among themselves, the winner in fact is not defined

Yeah, in the top one “100%” leaders, and this is a lot of what he says

Dec 11, 2017 6:26:25 PM
alkozel

alkozel

Калужский филиал ГУ СДА при Спецстрое России

I hard against randomness, it was possible to organize a regular tournament, having estimated the number of games. In our case, 59 games per person. Divide 360 participants in 60 groups (6 in each) take the results of the places in the sandbox. In each group conduct a round-Robin tournament, sort them from 1 to 6 based on the results. In the end we have 5 qualifying games for a place in the group. Well and further standard 59 games group to group (and within it the last with the first sort results). In the end we get 64 games, but guaranteed everyone will play evenly with everyone.

Dec 11, 2017 11:38:38 PM
VexfxjC

VexfxjC

БГТУ «Военмех»

I don’t know how 1 fight count more than 30 minutes… on my laptop 10тилетнем my strategy against my is considered to be only 10 minutes (and that’s with watching the battle, a runner)… clearly the organizers of Mossa more… and not the same as the computer, they start these battles…

Dec 11, 2017 11:48:25 PM
alex3d
VexfxjC

VexfxjC

БГТУ «Военмех»

I don’t know how 1 fight count more than 30 minutes… on my laptop 10тилетнем my strategy against my is considered to be only 10 minutes (and that’s with watching the battle, a runner)… clearly the organizers of Mossa more… and not the same as the computer, they start these battles…

So for 30 minutes and not one fight is considered. In one wave of the second round of 360/2=180 games. “10 year old laptop” wave will be considered as 30 hours. In the second half of the 2nd round was 40 waves in 12 hours - one wave in 18 minutes. Part of that for us employs about a hundred “10 year notes”. If you consider the fact that not everyone uses timelimit in full, and they did not-10-year-old-not-the laptop turns 2-3 dozen servers.

Dec 12, 2017 1:31:57 PM