Quote:
Originally Posted by DooDooPoker
I'm going to post what Matt Marinelli posted on his twitter (#1 online cash game player in America).
He simulated 6 bots at 0 rake to see the discrepancies in winrates over certain hand periods. Here are the results.
This means you could play 500k hands and break even over that time and you wouldn't know if you were a 6.5bb winner or -6.5bb loser
This is why Thought Process>Winrate.
Discuss.
I didn't read the thread, as it is already too big, so if this was argued already, I'm sorry.
I think it is a completely different beast to evaluate a bot, playing versus his fellow bots playing the same strategy (I assume this was the case right?), where we know for certain that his true EV is 0 bb/100, than evaluating a reg, hopefully playing a winning strategy, versus human players, and with some sort of normal game selection, who gets a winrate of, let's say, 5 bb/100 over 300k hands.
For the reg, there are statistical tools for you to determine what are the odds of him being a 5 bb/100, or being a winner, or even being a crusher, given the observed winrate, all the while you will never know for sure his true winrate.
For the bot playing vs bots, you know his true winrate from the beginning, and going over observed winrates is just an exercise on seeing what lucky/unlucky looks like. Futile exercise, and a waste of time, if you ask my opinion about it.