BL2 Bot Player
Re:BL2 Bot Player
One comment: The plot above is not a fit, in the sense that I did not optimise the parameter cc. The cc was calculated using excel. The fit would be better for higher number of tests and a higher value instead of 25000.<br><br>Post edited by: Lieven, at: 2008/02/09 10:53
Re:BL2 Bot Player
Ok, Cleaned up the excel sheet, attached. Sorry for spamming the forum. Feel free to merge these three posts [file name=poisson-ac480908ff0c37c4b866f37cd79030f0.7z size=64352]http://www.blockout.net/components/com_ ... 79030f0.7z[/file]
Re:BL2 Bot Player
Hi,
Thank you very much for the model :) I\'m going to have a look
at this stuff more in depth. I\'ll try to code this in C++ which
should be a bit faster.
Ok, Cleaned up the excel sheet, attached. Sorry for spamming the forum. Feel free to merge these three posts
Where is there some spam ? All your interventions are of a very
good quality ;)
a part les vérifier toutes pour chaque block et de les stocker
je ne vois pas de méthode globale
c\'est complexe comme problème !
Sur les 41, combien de pièces ont 24 orientations différentes ?
OK Thanks for your answer, I was just wondering if it exists
a simple method which allows, when you know the number of
planar symmetry, to deduce the number of transform (with det=1)
that conserve the shape and the orientation of the block.
I think that all chiral blocks has the 24 orientations so
it is already much. I think that for Flat Fun the algo can
be greatly boosted, but for OOC (even 3DMania?) I don\'t know
if we can greatly improve the stuff. (definitely a speed
up ratio lower than 2).<br><br>Post edited by: jlp_38, at: 2008/02/09 13:21
Thank you very much for the model :) I\'m going to have a look
at this stuff more in depth. I\'ll try to code this in C++ which
should be a bit faster.
Ok, Cleaned up the excel sheet, attached. Sorry for spamming the forum. Feel free to merge these three posts
Where is there some spam ? All your interventions are of a very
good quality ;)
a part les vérifier toutes pour chaque block et de les stocker
je ne vois pas de méthode globale
c\'est complexe comme problème !
Sur les 41, combien de pièces ont 24 orientations différentes ?
OK Thanks for your answer, I was just wondering if it exists
a simple method which allows, when you know the number of
planar symmetry, to deduce the number of transform (with det=1)
that conserve the shape and the orientation of the block.
I think that all chiral blocks has the 24 orientations so
it is already much. I think that for Flat Fun the algo can
be greatly boosted, but for OOC (even 3DMania?) I don\'t know
if we can greatly improve the stuff. (definitely a speed
up ratio lower than 2).<br><br>Post edited by: jlp_38, at: 2008/02/09 13:21
Re:BL2 Bot Player
Unfortunately i don\'t have excel at home :blush:
I\'m downloading it. ~8h00 of download remaining :(
Lieven: Could you explain me a bit better how do you
fit the Gaussian curve to the experimental data ? Do you
use it to estimate the value of cc ? What is the error
on cc after n tries ?
Anyway, I\'m still investigating on the symmetry problem.
Not much on this (for polycube) on the web...
May be the algo can be boosted up more than i thought.
I\'m wondering if it exits a polycube (up to nbCude=5)
that don\'t have symmetry.
All seem to have a least one planar symmetry or a
central symmetry, so the \"transform group order\" of each
block should be >=2. Am i wrong ?
So with the 41 polycubes, i think that 12 orientation
should be enough but it\'s just a supposition...
http://en.wikipedia.org/wiki/Octahedral_symmetry
I\'m downloading it. ~8h00 of download remaining :(
Lieven: Could you explain me a bit better how do you
fit the Gaussian curve to the experimental data ? Do you
use it to estimate the value of cc ? What is the error
on cc after n tries ?
Anyway, I\'m still investigating on the symmetry problem.
Not much on this (for polycube) on the web...
May be the algo can be boosted up more than i thought.
I\'m wondering if it exits a polycube (up to nbCude=5)
that don\'t have symmetry.
All seem to have a least one planar symmetry or a
central symmetry, so the \"transform group order\" of each
block should be >=2. Am i wrong ?
So with the 41 polycubes, i think that 12 orientation
should be enough but it\'s just a supposition...
http://en.wikipedia.org/wiki/Octahedral_symmetry
Re:BL2 Bot Player
I made a brute force algo to check invariant transform.
If my code is correct (I\'m not sure), here is the result:
(So my first supposition seems to be wrong)
=> 1 orientations
=> 3 orientations
=> 3 orientations
=> 3 orientations
=> 3 orientations
=> 12 orientations
=> 3 orientations
=> 12 orientations
=> 12 orientations
=> 24 orientations
=> 12 orientations
=> 12 orientations
=> 24 orientations
=> 12 orientations
=> 12 orientations
=> 24 orientations
=> 24 orientations
=> 12 orientations
=> 24 orientations
=> 24 orientations
=> 3 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 12 orientations
=> 12 orientations
=> 12 orientations
=> 12 orientations
=> 12 orientations
=> 12 orientations
=> 8 orientations
=> 12 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations
=> 24 orientations<br><br>Post edited by: jlp_38, at: 2008/02/09 20:17
If my code is correct (I\'m not sure), here is the result:
(So my first supposition seems to be wrong)
=> 1 orientations => 3 orientations => 3 orientations => 3 orientations => 3 orientations => 12 orientations => 3 orientations => 12 orientations => 12 orientations => 24 orientations => 12 orientations => 12 orientations => 24 orientations => 12 orientations => 12 orientations => 24 orientations => 24 orientations => 12 orientations => 24 orientations => 24 orientations => 3 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations => 12 orientations => 12 orientations => 12 orientations => 12 orientations => 12 orientations => 12 orientations => 8 orientations => 12 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations => 24 orientations<br><br>Post edited by: jlp_38, at: 2008/02/09 20:17Re:BL2 Bot Player
If the above are correct, here is the speed up ratio
we can expect for all mode:
Flat => 2.74286
Basic => 1.82609
Extended => 1.53991<br><br>Post edited by: jlp_38, at: 2008/02/09 22:28
we can expect for all mode:
Flat => 2.74286
Basic => 1.82609
Extended => 1.53991<br><br>Post edited by: jlp_38, at: 2008/02/09 22:28
Re:BL2 Bot Player
I will post here the details, so there is no reason to install excel.
From my first post on this: The cc number is simply the number of average restarts, total number of runs/total of restarts. in the example cc=20000/854=23.41. Therefore, this can be straightforwardly calculated.
So there is absolutely no fitting going on! The fact that the plotted curve matches the experimental curve so well, is good evidence that the model works.
Total restarts, N=854
sqrt(N)=29.2232
Then 68 confidence interval is in N+/-sqrt(N)
or in between 824.7767216 and 883.2232784 total restarts
Converting this back in the average number of restarts we get a cc value in between 22.6443307 and 24.248987.
or thus an average number of blocks between
25000/22.644=1030.970902 and 25000/24.248987=1104.029098
Other confidence intervals might be harder to calculate. Haven\'t look deep into it.
Hope I didn\'t make any mistakes here, it is late :)
From my first post on this: The cc number is simply the number of average restarts, total number of runs/total of restarts. in the example cc=20000/854=23.41. Therefore, this can be straightforwardly calculated.
So there is absolutely no fitting going on! The fact that the plotted curve matches the experimental curve so well, is good evidence that the model works.
Total restarts, N=854
sqrt(N)=29.2232
Then 68 confidence interval is in N+/-sqrt(N)
or in between 824.7767216 and 883.2232784 total restarts
Converting this back in the average number of restarts we get a cc value in between 22.6443307 and 24.248987.
or thus an average number of blocks between
25000/22.644=1030.970902 and 25000/24.248987=1104.029098
Other confidence intervals might be harder to calculate. Haven\'t look deep into it.
Hope I didn\'t make any mistakes here, it is late :)
Re:BL2 Bot Player
Thanks, now I understand better :)
I was just curious to see the distribution. I plotted it
without \"histogram binning\". Strange distribution near 0 ?
Did you get similar results ?
Avg=1076.08 Min=20 Max=13438 nbGame=100000
<br><br>Post edited by: jlp_38, at: 2008/02/10 01:47
I was just curious to see the distribution. I plotted it
without \"histogram binning\". Strange distribution near 0 ?
Did you get similar results ?
Avg=1076.08 Min=20 Max=13438 nbGame=100000
<br><br>Post edited by: jlp_38, at: 2008/02/10 01:47Re:BL2 Bot Player
Hi,
I\'m going to work a bit on the optimization.
I\'ll code the TransformCube() and fround() function
in SSE and add the symmetry optimization.
I plan to get a factor ~2 on OOC.
I\'ll post the code ASAP on sf.net
I\'m going to work a bit on the optimization.
I\'ll code the TransformCube() and fround() function
in SSE and add the symmetry optimization.
I plan to get a factor ~2 on OOC.
I\'ll post the code ASAP on sf.net
Re:BL2 Bot Player
I was wondering if the strange distribution around zero
was linked the number of couple (x,y) solution to 3*x + 4*y = n.
Only 3 or 4cube block in 3DMania. However the red curve should
be corrected as there is only 1 \'3cube\' block and 6 \'4cube\' blocks.
<br><br>Post edited by: jlp_38, at: 2008/02/10 19:42
was linked the number of couple (x,y) solution to 3*x + 4*y = n.
Only 3 or 4cube block in 3DMania. However the red curve should
be corrected as there is only 1 \'3cube\' block and 6 \'4cube\' blocks.
<br><br>Post edited by: jlp_38, at: 2008/02/10 19:42Re:BL2 Bot Player
OK, I uploaded the new code to sf. I added the symmetry
optimization. I didn\'t have a look at the AI playing
but it should be a bit nicer than before (Nicest
rotation).
The AITester is ~2.3 times faster in 3x3x6 BASIC,so a little
bit more than expected :)
(I also made small memory access optimizations)
I added a file BotPlayerAI.cpp where i concentrated all concerning
the evaluations and a file BlockOrientation.h.
Let me know if you get problems.
optimization. I didn\'t have a look at the AI playing
but it should be a bit nicer than before (Nicest
rotation).
The AITester is ~2.3 times faster in 3x3x6 BASIC,so a little
bit more than expected :)
(I also made small memory access optimizations)
I added a file BotPlayerAI.cpp where i concentrated all concerning
the evaluations and a file BlockOrientation.h.
Let me know if you get problems.
Re:BL2 Bot Player
Some notes, I remarked a bug. The demo cheats a bit ;)
One collision was not detected at the top of tip
and in some cases the demo can win few blocks.
I didn\'t make long test, but the avg seems to be affected
of ~ -5% in 3x3x6 BASIC.
Sorry for that :S
If you want to re-enable the bug, you can make the
following change:
BotPlayer.cpp:376
validMove = IsValidPos();
by
validMove = TRUE;<br><br>Post edited by: jlp_38, at: 2008/02/10 23:57
One collision was not detected at the top of tip
and in some cases the demo can win few blocks.
I didn\'t make long test, but the avg seems to be affected
of ~ -5% in 3x3x6 BASIC.
Sorry for that :S
If you want to re-enable the bug, you can make the
following change:
BotPlayer.cpp:376
validMove = IsValidPos();
by
validMove = TRUE;<br><br>Post edited by: jlp_38, at: 2008/02/10 23:57
Re:BL2 Bot Player
Hi,
I might found out a new OOC secret ;)
[ 81] Cube=26802 Avg=23211.1
[ 82] Cube=4372 Avg=22984.2
[ 83] Cube=1374 Avg=22726.9
[ 84] Cube=5888 Avg=22528.8
[ 85] Cube=141915 Avg=23917
[ 86] Cube=38428 Avg=24083.8
[ 87] Cube=6309 Avg=23881.8
[ 88] Cube=8227 Avg=23705.9
[ 89] Cube=8125 Avg=23532.8
[ 90] Cube=20897 Avg=23503.8
[ 91] Cube=19851 Avg=23464.1
[ 92] Cube=17382 Avg=23398.7
[ 93] Cube=25976 Avg=23426.1
[ 94] Cube=9773 Avg=23282.4
[ 95] Cube=10432 Avg=23148.6
[ 96] Cube=33837 Avg=23258.8
[ 97] Cube=8405 Avg=23107.2
[ 98] Cube=3135 Avg=22905.5
[ 99] Cube=4212 Avg=22718.5
[100] Cube=44773 Avg=22936.9
I might found out a new OOC secret ;)
[ 81] Cube=26802 Avg=23211.1
[ 82] Cube=4372 Avg=22984.2
[ 83] Cube=1374 Avg=22726.9
[ 84] Cube=5888 Avg=22528.8
[ 85] Cube=141915 Avg=23917
[ 86] Cube=38428 Avg=24083.8
[ 87] Cube=6309 Avg=23881.8
[ 88] Cube=8227 Avg=23705.9
[ 89] Cube=8125 Avg=23532.8
[ 90] Cube=20897 Avg=23503.8
[ 91] Cube=19851 Avg=23464.1
[ 92] Cube=17382 Avg=23398.7
[ 93] Cube=25976 Avg=23426.1
[ 94] Cube=9773 Avg=23282.4
[ 95] Cube=10432 Avg=23148.6
[ 96] Cube=33837 Avg=23258.8
[ 97] Cube=8405 Avg=23107.2
[ 98] Cube=3135 Avg=22905.5
[ 99] Cube=4212 Avg=22718.5
[100] Cube=44773 Avg=22936.9
Re:BL2 Bot Player
Avg=20672.9 Min=346 Max=141915 nbGame=1000
114min 34sec, 3007.56 cps
114min 34sec, 3007.56 cps
Re:BL2 Bot Player
Simply impressive!
How does it do on 3x3x6 and 3x3x18? I am curious what the maximum nr of blocks you get out of say 1000 runs. Will be interesting to compare this to the best human scores, around 400 and around 900 blocks respectively.
How does it do on 3x3x6 and 3x3x18? I am curious what the maximum nr of blocks you get out of say 1000 runs. Will be interesting to compare this to the best human scores, around 400 and around 900 blocks respectively.