| Poker-AI.org http://poker-ai.org/phpbb/ |
|
| Compute Ev http://poker-ai.org/phpbb/viewtopic.php?f=24&t=2704 |
Page 1 of 1 |
| Author: | MrNice [ Fri Feb 14, 2014 1:32 pm ] |
| Post subject: | Compute Ev |
Hi, I would like to compute EV at the first node of my gametree. I have read the source code of opencfr but in my case I'm a bit lost how to calculate the reach value for my first node and for next chance nodes... to calculate the reach of the first node, I have two idea : - calculate the probability to reach the bucket (number of hole combinaison card in the bucket/169) - calculate the probability to get the hole card What about next chance nodes ? thanks for your help. MrNice |
|
| Author: | Pitt [ Fri Feb 14, 2014 1:54 pm ] |
| Post subject: | Re: Compute Ev |
If I understand correctly, you have a set of strategies computed for a particular abstraction and wish to get EV (ie game utility) for each player ? Then you need a full tree traversal. Just pass the payoffs back and ponderate with action weights and chances weights. I don't know how you structured your tree but I don't think this : Code: number of hole combinaison card in the bucket/169 makes sense. Anyway, when using CFRM, you should be able to get an incremental utility sum that you can divide by your iterations number. |
|
| Author: | MrNice [ Sat Feb 15, 2014 8:40 am ] |
| Post subject: | Re: Compute Ev |
Hi Pitt, Thank you for the info. The payoffs is already back propagated and weighted with the action probability but not by the chance weight... This is what I don't understand : how to calculate the chance weight based on my abstraction... By the way I'm using chance sampling so should I weight the payoff by the chance weight as well or the chance weight =1 ? Have a nice week end. MrNice |
|
| Author: | spears [ Sat Feb 15, 2014 10:39 am ] |
| Post subject: | Re: Compute Ev |
Are you trying to calculate ev from your abstracted model without reference to particular cards? If you want to do that you'd have to record the number of times you moved from one bucket to another at a chance node. Alternatively, just sum the evs of every iteration of cfrm and divide by the number of iterations. |
|
| Author: | Pitt [ Sat Feb 15, 2014 11:44 am ] |
| Post subject: | Re: Compute Ev |
Code: By the way I'm using chance sampling so should I weight the payoff by the chance weight as well or the chance weight =1 ? Yes, on sampled nodes in CFRM, utility is passed back as is. |
|
| Page 1 of 1 | All times are UTC |
| Powered by phpBB® Forum Software © phpBB Group http://www.phpbb.com/ |
|