search

xlMuRlx / Texas-holdem

Obravnava verjetnosti zmage igralca v igri Texas hold 'em poker glede na njegove karte in število nasprotnikov.
0 stars 0 forks source link

Ustreznost programa in časovna zahtevnost #2

Open xlMuRlx opened 4 years ago

xlMuRlx commented 4 years ago

Potem, ko mi je uspelo zapisati glavni funkciji ovrednoti in model, sem s projektom, kot sem ga imel v mislih, v grobem končal. Seveda moram urediti še shiny, prav tako pa bi lahko uvedel nekaj manjših popravkov.

Zanima me predvsem, če je projekt v takšni obliki ustrezen ali moram programu še kaj dodati. Poleg tega je izvajanje funkcije model precej časovno zahtevnost, vendar nevem kako bi to lahko izboljšal, v kolikor je to sploh potrebno.

systats commented 4 years ago

Hi xlMuRlx - interesting project. Although I can not follow your project description, google translator helped a lot ;). Are you a professional poker player or whats your motivation? What is your domain expertise? We are currently experimenting with neural networks and self play in a pretty complex holdem environment. So I would be delighted to exchange some bits of knowledge. Cheers, Simon

xlMuRlx commented 4 years ago

First of all I'm sorry that the description wasn't more understandable. I didn't really expect it would go international and therefore didn't write anything in English. My project is basically an assignment I got during my studies, which is also the main reason it is a more or less a simple program. As far as I know the Monte Carlo method in poker that I'm using for my calculations only becomes effective from upward of a million iterations. I did, however, also have to write a user console and came to the conclusion that it is not user friendly to have to wait more then 5 minutes for the calculations to finish.

As far as the question of my expertise, I'm a financial mathematics student. As I already mentioned the motivation for the project is as assignment I got, in which I had to choose any mathematical subject and write a program in that field. Since I most of all like solving probability problems I decided to chose calculating poker probabilities because it combines that with a game I enjoy playing.

jO-Osko commented 4 years ago

Za kratko predstavitev se mi zdi program že kar dobro narejen, bi bilo pa lepo dodati nek indikator kako daleč je že simulacija in kakšen je trenuten rezultat. Bi se dalo mogoče oceniti napako simulacije?

Pri branju obstoječih konfiguracij je smiselno preveriti, ali sta dve stanji enaki (5križ in 9 križ lahko uporabi že poračunano stanje 9 križ in 5 križ, če je vse ostalo enako).

Pri branju datoteke je lepše, če se uporabi funkcija file.path, saj tako program avtomatsko prilagodi pot do datoteke glede na operacijski sistem (problematična sta predvsem \\ in /).

Za dolgo predstavitev bi bilo smiselno dodati še možnost izbire kart nasprotnikov (ali pa jih pustiti nepoznane kot do sedaj) in nekaj natančnih izračunov.

Vsaj določene stvari se da izračunati natančno. Če vem, kakšne so moje in nasprotnikove karte, ali lahko pametno izračunam verjetnosti zmage, ob nekaj še ne odkritih kartah? Ko to znam, ali lahko, kljub temu, da nasprotnikovih kart ne poznam kaj povem o verjetnosti?nasprotnik ima v najslabšem primeru 54*53 možnih kombinacij, kjer so nekatere ekvivalentne, kar ne presega 2000. Če lahko dokaj hitro izračunam prvi del, imam lahko dokaj hitro tudi točen odgovor.

systats commented 4 years ago

Pretty interesting stuff though. Do you think there is value in feeding a neural net some pre computed probabilities? Concepts like Equity are based on strong assumption of your opponents cards so rather useless as input feature.