On 3 juin 2011, at 20:07, Harald Johnsen wrote:

> Hi,
> 
> I've been busy writing some code to extract feature from position to feed the 
> MM tool from Rémi.
> As I was reading the paper one more time I stopped on that sentence :
> 
> "Each value that these features can take is considered as a separate 
> “individual”, and is associated to one strength parameter i. Since values 
> within one feature are mutually exclusive, they were all updated together 
> within one iteration of the minorization-maximization algorithm".
> 
> For some reason I was feeding the MM tool differently, i.e. I have one 
> feature per spatial pattern or game feature i.e. I don't have one feature for 
> distance to border, I have 4 features for that (one feature for each 
> distance).
> Also Rémi said he was using like 16k shapes, this would give me 16k features 
> the way I doing atm and the MM tool would take days to output any results.
> 
> To make it short, how to you feed the MM tool (I've joined the beginning of 
> my input file).

This is the wrong approach. What I call an "individual" is what some authors 
would call a "binary feature", ie a bit of information indicating that a 
property is true or false. A feature is a collection of mutually exclusive 
individuals. So "distance to previous move" is a feature, and "distance 1", 
"distance 2", ... are different levels of that feature. "distance to border" is 
another feature. It must be another feature, because it is not mutually 
exclusive with "distance to previous move", ie you can be at distance 1 from 
the previous move, and at distance 2 from the border at the same time. 3x3 
shape is another separate feature, with as many levels as there are 3x3 shapes.

I hope this makes things clearer.

Rémi

> 
> Additional question : I have millions of lines in that file since we have 1 
> mm game per real game move, each mm game has like 360 teams fighting on an 
> empty board. Is that ok ?
> 
> HJ.
> 
> 
> 
> ! 419
> 419
> 1 f.0.none
> 1 f.1.cfg_distance_1
> 1 f.2.cfg_distance_2
> 1 f.3.cfg_distance_3
> 1 f.4.ft_mc_owner_0
> 1 f.5.ft_mc_owner_1
> 1 f.6.ft_mc_owner_2
> 1 f.7.ft_mc_owner_3
> 1 f.8.ft_mc_owner_4
> 1 f.9.ft_mc_owner_5
> 1 f.10.ft_mc_owner_6
> 1 f.11.ft_mc_owner_7
> 1 f.12.ft_reduce_to_1_lib
> 1 f.13.ft_reduce_to_2_lib
> 1 f.14.ft_reduce_to_3_lib
> 1 f.15.ft_atari_1_stone
> 1 f.16.ft_atari_2_stones
> 1 f.17.ft_atari_3_stones
> 1 f.18.ft_capture_1_stone
> 1 f.19.ft_capture_2_stones
> 1 f.20.ft_capture_3_stones
> 1 f.21.save_1_stone
> 1 f.22.save_2_stones
> 1 f.23.save_3_stones
> 1 f.24
> 1 f.25
> 1 f.26
> 1 f.27
> 1 f.28
> 1 f.29
> 1 f.30.border_distance_0
> 1 f.31.border_distance_1
> 1 f.32.border_distance_2
> 1 f.33.border_distance_3
> 1 f.34.selfatari
> 1 f.35.bad_selfatari
> 1 f.36.own_eye
> 1 sp.37.0
> 1 sp.38.1
> 1 sp.39.2
> 1 sp.40.3
> 1 sp.41.4
> 1 sp.42.5
> 1 sp.43.6
> 1 sp.44.7
> 1 sp.45.8
> 1 sp.46.9
> 1 sp.47.10
> 1 sp.48.11
> 1 sp.49.12
> 1 sp.50.13
> 1 sp.51.14
> 1 sp.52.15
> 1 sp.53.16
> 1 sp.54.17
> 1 sp.55.18
> 1 sp.56.19
> 1 sp.57.20
> 1 sp.58.21
> 1 sp.59.22
> _______________________________________________
> Computer-go mailing list
> [email protected]
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to