well it does not assumea range like we do but in some other way forexample he knows utg openismuch tighter so he 3bet less than the button open

While I am interested in hearing answers to the above questions and more, this statement bothers me. A bunch of random people playing what is essential free money poker isn't a measure of how strong Snowie is or isn't. If there is no money at risk there is no poker.

Hi Everyone:

Let's welcome Roberto and Stephan to our forums. I know there's a lot of questions relative to PokerSnowie and this should be a good way to possibly get some answers.

Best wishes,
Mason

Hi zachvac:

We've had communication with the PokerSnowie people and this sort of thing is not in the cards.

Best wishes,
Mason

Hi mike1270,

your comment gives me the opportunity to touch two important topics: the challenges and the random generator. I fully agree with you, we can't use the challenges as a strong or final proof about the AI being strong, and therefore I understand why it bothers you.

In our opinion it's just one of the facts currently showing some level of strength of the AI, but in fact I agree that a real impartial challenge should probably be for real money and organized in a different way.

We are internally evaluating a more solid format for the challenge and we would welcome any idea or suggestion by you and the other 2+2 readers about how we could run it.

I realize that having the random generator on our client (btw we are simply using the standard C++ random generator) could create some issues to peple suspecting that we might be manipulating it, as well as I realize that technically people could think that Snowie is "made aware" of the cards of the opponent or the cards that are coming in the boards and your have no control on it.

Therefore we think that a "real challenge" probably should have the following characteristics:

- it should be for real money
- it should preferably happen on a third party website/server/room
- there should be a way to monitor the RNG

As said, we are evaluating it and we would welcome any suggestion around it coming from the 2+2 community.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

Suggestion: since you openly admitted to us that you have no idea of where PokerSnowie is in respect to perfect GTO play, remove all BS marketing verbiage on your site saying or implying otherwise.

@PokerSnowie:
I would assume you tested your methodology on simpler abstract games to verify the approach is working correctly. Can you disclose some general information about the most complex poker-related game that you ran the Snowie algorithm on where you have actual data on the exploitability of the calculated solution?

@PokerSnowie:

I am sure everyone would be interested to know how Snowie performs in the annual computer bot competition: http://www.computerpokercompetition.org/.

Submissions are still open for this year's competition. Here we could see how well it performs against state-of-the-art bots currently developed by computer poker research groups.

What are your thoughts on this challenge?

^ this

Definitely the most important question, and a recurring one through this thread.

i read the entire thread up until the point where i couldn't take it anymore. nice post man.

the gto claim is quite ridiculous.

now that we have established that a number is not a pi candidate unless it shows a zavac of digits after the decimal point we could turn our attention to issues more marginal in nature.

to me this thing looks like a shortcut for botters - why code your own engine when it may be simpler to look for ways to "extend" the functionality of this bot. they are a bit scarce on info what means they have in place to prevent this. i mean ..not starting when a poker client is open is more like a joke you think is funny after a couple of shots with your buddies, no?

Hi all,

before answering to some more poker-related question, I would like to update you about our ideas on improving the challenges, in term of transparency. Let me try to summarize what we are going to release pretty soon.

- When a user will start a challenge session, he will have the option to export into a file the cards randomly generated for a certain number of hands, prior to start playing. This file will be used by Snowie during the challenge, instead of the random generator. We think that this feature will remove any doubt about Snowie manipulating the random generator in order to get the best cards.

- Even with this deck exported, if Snowie knows the hole cards, it could theoretically choose the best move which maximize its equity. For example, if the Snowie evaluation is 80% FOLD and 20% bluff-RAISE, if Snowie knows that the opponent has the nuts, then Snowie could always choose to FOLD. To avoid this, together with the deck, we will also export a set of random numbers which will also define which move Snowie will have to do when the mixed evaluations come. And this will be on a file, stored on the user machine.

- Even with these two changes, someone could think that Snowie will manipulate its evaluations, in order to get advantage by knowing the opponent hole cards and the cards that will be dealt on the board. In the example above, Snowie could say that the evaluation is 100% FOLD. About this, you just need to go to scenario, set up the position, and you will immediately see that in a given situation, the Snowie advice is always the same.

- I'd also like to remind that the histories of the played hands are stored in the PokerCoach folder, and therefore any external statistical analysis can be done directly by the user or by any third party.

- Finally, in order to avoid giving to the user an extra advantage by knowing in advance Snowie's cards and the board cards, the file which will be exported at the beginning of a challenge will be encrypted with an open source algorithm and the key to decrypt it will be given at the end of the session.

Comments are welcome.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

Snowie can't express the reasons why something is right or wrong, but our interpretation is this: by raising you want to achieve that the opponents fold a significant portion of their hands.

If you raise in early or middle position to 2.25 big blinds, your raise is strong enough to make the big blind player fold quite some hands.
If you raise on the button, however, your hand range is quite wide. This fact, combined with the good pot odds you are giving by raising to only 2.25 bb, gives the big blind player a good reason to defend his blind very much. Therefore, a higher bet size on the button makes a lot of sense.

The PokerSnowie Team

It's difficult to answer to this question without knowing the concrete situation (stack sizes, first raise amount, dead money in pot etc.). It seems like Snowie chooses the bet size full pot and you would prefer a smaller amount. But could it be, that raising full pot has the advantage to cripple your re-raise power? Then Snowie could profitably call with his bluffing hand, being itself already pot committed.

Raising only half pot would allow you to make a powerful re-raise, where you can add bluffs yourself. So despite your argument, raising full pot could still be right.

The PokerSnowie Team

really? on a side note: you're dodging the important questions.

Yes, the EV is determined assuming the opponents to play also a GTO strategy. Snowie always chooses the option with the highest EV.

The PokerSnowie Team

You are surprised with good reason. You actually pointed out a current weakness of Snowie: dealing with an out of position bet on multiway pots (especially against donk bets). The training algorithm had a weakness in this kind of situation, which has been fixed. With the next brain update this weakness will be gone.

The PokerSnowie Team

Are EV's expressed in $ and cents? Because even if the training algorithm was wrong it seems strange that it displays an EV of -0.69 when it only costs $0.10 to call. Calling for $0.10 can't have an EV of less than -$0.10.

The EV (expected value) is displayed in big blinds

Snowie will maximize EV of whole range with the bet size that it recommends in certain situation, right? Even if some hands can have higher EV with different bet size?

I'm currently testing out PokerSnowie as a learning tool and in order to avoid bringing any of my own experience/bias in, I'm practising HU cash - which I think is sufficiently different from my usual preferred game, which is 45 man turbo SNGs.
(Incidentally over the last 30 years I've won significantly more than I've lost playing stud & holdem, so I'd like to think I'm a good learner).

What I would like to do with Snowie is to learn HU to a sufficiently high standard before I enter live games, because I hate learning lessons the expensive way.

But while I've been able to get comfortable with the preflop play, I am getting increasingly frustrated with the post flop learning.

As a teacher, Snowie leaves a lot to be desired because it doesn't communicate any specific strategies or tactics (in the same way that watching say a Phil Galfond video would).

Instead of "lightbulb moments" of insight, I'm left scratching my head when Snowie tells me my cbet on the flop (that gets Snowie to fold) is wrong because my EV is 0.66 versus an EV of 0.69 for checking. But when Snowie then inevitably bets after my check, I'm the one whose is expected to fold! WTF

So after weeks of using this programme, more or less the only things I've learned about post flop play is that if I call in the Big Blind, I should check the flop roughly 98% of the time. Or if I open the button and the Big Blind calls and then checks the flop, I should also check about 65% of the time and raise 35%.

There are almost infinite permutations of hole cards and flopped community cards that I could test - but life is too short.
I need to know WHY I should be taking a specific action in each spot and on each street. (Maybe I'm missing something blindingly obvious here...)
Perhaps if Snowie let me download, into a spreadsheet, a million hands that it had played, I could identify all the betting patterns, etc., much quicker.


In terms of proof of Snowie's capabilities, I don't see how it would benefit Snowie to "cheat" it's students, since presumably the purpose of Snowie is to help people play better.
In which case, the best way of proving their case is by producing a load of winning players (rather than having a pissing contest with other AI programmes).

So perhaps they could start with showing subscribers the specific methods that were used to turn a bunch of newbies (Team PokerSnowie) into mid to high stakes, and international tourney, players in just a few weeks.

Try to callculate ev your self. You know snowie ranges so its not that dificult that way you will learn faster.

You guys throw claims in relation to "GTO" around like confetti. If you want to be taken seriously, you really should be a lot more careful in your use of this term and stop with all the marketing BS trying to exploit players who have heard the term, but don't really know what it is.

I get the impression that it's either their entire goal or else they're part of those players.