Poker Alfie
Any thoughts on this?
Is it beneficial?
What win rates would translate at which levels in NL vs people?
Do the bots react to you, will they respond to different play styles?
Do the different bots maintain the same play styles on each profile or is it randomized each session?
I'm only a few hundred hands in but winning 1.06 bb per hand (which is obviously a heater) but I don't really feel half as challenged as if I were playing even NL2
8 Replies
I haven't reported anything about PokerAlfie for about a year.
These are the improvements that have been made during this time:
- There are now 3 AIs to choose from in the app:
1. Poker AI v3.1
This is the AI that Rickroll played against, but with a few very small improvements.
2. Poker AI v4.0
This is the complete opposite of v3.1 – very aggressive
The skill level is about the same as version 3.1 or maybe even slightly weaker.
3. Poker AI v5.0
It is a more balanced and stronger player than v3.1 and v4.0. For many users, v5.0 presents a much greater challenge than older versions.
- Analysis - Feature:
Analysis offers the possibility to replay any poker hand you want and ask the AI for advice and suggestions in every situation of the hand. Any hand means: a
hand played against PokerAlfie, or a hand you have played elsewhere, or a hand played by others in a live poker tournament, or …
Android Version
https://play.google.com/store/apps/detai...
Windows Version
https://www.microsoft.com/store/apps/9N7...
Windows Version – direct link to Windows Store
ms-windows-store://pdp/?productid=9N7Q05F4SLZ0
How strong is Poker AI v5.0? Here are some comments from other users:
1. “
I've played 1111 hands so far (spread over 10 tournaments with 2 of them being wins and these are my initial impressions:
• I agree with you that v5 seems a lot more balanced, especially compared to v4, which I thought was too aggressive. In v4, the strategy that seemed to work best was to rarely or never bluff (fold to most aggression if you have nothing) but then when you have stuff, you can be passive and let them bet into you to make the pot large. At the end you can bet large or raise and they'll often call. With v5, I found that I needed to bluff to win, since it often (correctly) folded when I bet a good hand (and I couldn't rely on their aggression to build the pot).
• So far, I think v5 is more fun to play against. When I played against v3 and v4, my play style was somewhat formulaic, but with v5 I have to put effort into thinking about whether I'm making a good bluff. Maybe if I play several thousand more hands I'll start to find more robotic patterns like I did for v3 and v4.
• There are still some strange issues with calling patterns
”
2. “
Hey there! Nice app.
Played 5k hands and won at 6bb/100, started losing big until I found it's weaknesses.
It will make crazy calls against overbets on the river x3 or x4, a lot of them on paired boards with things like K high on a board where plenty of hands were possible. So easy to get a lot of value from your strong hands.
If you found a way to fix that, it would be amazing and harder to beat.
Great UI, regards!
”
https://forumserver.twoplustwo.com/167/p...
Future development:
I’ll try to improve the algorithm of v5.0 - especially issues with calling. If I make progress I’ll add a new improved version v6.0 to app.
nice, got a new laptop so downloading now to try out the new version
I just published the new version v5.2, it has new Poker AI v5.2.
AI v5.2 improves the calling issue on the river.
hi alex, just toyed around with it a little
ai still does some funky things
it doesn't appear to properly adjust to the size of the wager
ie MP will open to 3 it folds around to bb who minclicks it up to 7 and mp will fold
this is almost never a fold
ai also tends to struggle with relative hand strength
this was a tiny pot that emma decided she had the nuts on with K high flush on a double paired board and not only reraised but also called an overjam on the river
there's also basically zero multiway pots, it just never happens, not even with cooler situations
i stand corrected
yeah you've definitely fixed the no multiway pots problem of the past, seems I just ran into an issue of low sample size earlier where it was only going heads up like in the past
this was interesting, mike raised Daniel's cbet on the flop and when Daniel shoved mike called with K high
seems like a punt in a vacuum but I like it, shows that Mike was willing to bluff and then understood he was pot committed to call off the remainder - I think older version here doesn't make that play, or if it does, snap folds for just a little more
