The short answer that it just looks for the "solution" at a specific moment, whether it thinks a hand should be in your range or not*. When it was 'training' against versions of itself (some programmed to act randomly), it learned how to play against all hands on all runouts (and with all action sequences) rather than just the hands it 'expects' you to have. If you have the full/pro version, you can select "Show all hands" in the range advice section, and it will have a strategy for every single combo on every street. e.g. You should normally fold 72o pre-flop, but if the board comes 7x77x, it's not going to advise checking back the river with that hand. It has an understanding that quads is a strong hand, even though it didn't recommend seeing the flop with 72 in the first place.

* That said, if you (or one of your real world opponents) take(s) a line that Snowie wouldn't take with any hands at all, it will pop up an error message when you look for range advice.

Can I use the sottware for HU SNGs?

Is any poker player ever supposed to just look at the hand and think "oh, with this hand on this board on the river I do this" ... or isn't considering the action that lead to the particular spot the heart of the game?

Of course but I was just curious because if it's telling you what to do on river after you had made a bunch of mistakes prior to river and shouldn't actually be in that spot....then the river advice will not be that useful and likely to let you continue on with a ****ty strategy.

I am posting all of this from the perspective that I import hands from hm2 to evaluate (not from scenarios you create) so it's not easy to go back and change the action to match optimal strategy to ensure you're in a river spot you intend be and can apply your strat.

I could see it giving advice based on the strength of your hand, preflop action, position, spr, and blockers. Not taking into account prior postflop action.

It considers more or less the past also (you can see this in many situations, it being programmed). It is also how it is at the holdem manager site, that it evaluates your gto check and bet (and so) ranges, as they should be played based on your range.

Does Snowie play better if you stick to the clickable bet sizes, since that's what it was trained for?

And 2x was part of its training, right? Even though it's not a button.

This is a difficult question to answer, because I can't say how "well" Snowie plays in the first place. I think it will be more "accurate" if you stick with what it knows best. My gut feeling is that it becomes slightly exploitable if you choose slightly smaller sizes with bluffs, and slightly larger sizes with value hands, because I believe it "buckets" the bet-sizes. i.e. If you bet 48% of pot or 52% of pot I think it will respond as if you bet 50% of pot, so you could save/earn some money against it by deviating slightly based on hand strength.
I also have a feeling it doesn't respond particularly well to very large overbets. i.e. There are spots where it calls surprisingly wide vs a 2x pot bet, but if you go with 4x pot, it might fold everything but the nuts.

Any chances to have a version of pokersnowie who work with antes in a future ?

Anyone have a problem with Snowie not importing hands? It seems to refuse to import a ton of hands with the following error code



Emailed Support a few days ago and still waiting on a reply. Just wondered if anyone else has this problem.

+1

^^

oh yes ... +2

not sure if Snowie folks follow this thread but +3

I've had import errors before, but that "duplicate card" error message is a new one for me. Are you playing on Party or Stars?

Stars. I play using the FTP skin, not sure if that makes any difference. Anyway, support got back today to confirm they're working on it after I sent them a new hh file that had 36 hands and Snowie rejected......35 of them, with the above error. Something is very wrong as the profit/loss thing is also way way out for me. In one sample looked at I should've been up a few hundred dollars (at $25nl) but when I imported them Snowie was showing me as being down by over $3000. I'm assuming the problems are related in some way.

I have been using it for a month and really love it although I am not very good.
Even with its flaws it is for sure better than posting a hand on here and getting 5 different responses.

It doesn't calculate it as such. It just looks at "identical" situations from its historical training and sees how much it gained or lost on average.

e.g. In a pre-flop 6-max 100NL spot where it folds to the button who has AA, Snowie's database says that AA will win 8.23bb on average if it opens with a pot-sized raise.

This is analogous to running filters on a HEM/PT database, except that Snowie has billions of hand histories to compare, so it can look up its expectation for any post-flop board texture or action sequence. That said, the game tree for NLH (especially multiway) is effectively infinite, so you should take the EV figure given by Snowie with a grain of salt. It can't have played every post-flop sequence enough times for the EV to have converged on a particularly accurate number.

Why does Snowie often not have any (or very few) bluffs when it bombs river?

its river strat is bad in a lot of spots

For robots, balancing is evidently a hard skill to learn.


But seriously, I don't understand why it is sometimes incredibly unbalanced. It's definitely quite weird, and doesn't seem to match what most poker coaches say about river strategy. (All the stuff Polk preaches about "You need to have X amount of bluffs", for example). I can't decide if Snowie still has a lot to learn about poker theory, or humanity does. :/

Thanks guys. Doesn't this beg the question why trust Snowie's flop and turn strats though? I thought that once you know the ranges going into it, the river is by far the easiest street to solve because no cards to come/all equity realised/fewest combos.
Seems like a big problem developers should have solved by now tbh.

I know Claudico folded a king on AQJrTx to a massive overbet and when WCG asked wtf they said folding a percentage of Kx is Nash/GTO and calling would be "Bayesian".
Maybe Snowie spazes for similar reasons?

If Snowie's river play is sometimes (very) wrong, it could be because the neural net was programmed to cut off negative EV options before the sample size was large enough to assess them correctly in some situations. i.e. It might have tried doing big bluffs on the river, got called and lost a stack, so it decided "Oh, I won't do that again", whereas the big value bets won a lot of chips so it decided "I'll do a lot more of that". In effect, the results would keep reinforcing the "bet the nuts, don't bet 7-high" strategy, and instead of balance arising, a massive imbalance is created. But I'm really not sure what's going on.
I'd like to see some Pio/GTOrb solutions for river spots using the same ranges that Snowie gets to the river with, because it's really annoying when I play a Snowie challenge and I know its river-betting range in a particular spot is 98% boats, and yet I'm still supposed to call with trips.

Snowie's pretty easy to beat at heads up 100bb freezeouts imo. My strat consists of running it over, got rated extra terrestrial in my first match lol (first and only time so far). It says I make errors damn near every hand (generally by betting or raising when it says check, call or fold). The EV of Snowie's choice is often only slightly greater than the more aggressive option. In theory I do often agree with it's suggestion to check mid pair, it's just so content to fold to donk bets though.

Couple of questions about snowie.

How can you assign a seat, mostly in poker games my seat is at the bottom of the table,and not in the top upper right...

Why no sound effect? ,not very important ,but maybe fun.

Why the bots play so quick against each other ,no way to slow it down? Sometimes i fold and see other snowiebots in action ,but going so quick.

For the rest ,great programm and very helpfull with my game.

I don't know if this has already been addressed, but many of the preflop ranges give a different % than when put into flopzilla. This seems to only be the case with multi way scenarios, as it's the first I've noticed, and I always ensure the frequencies match. I've attached a pic with an incorrect % that was the easiest to see and match up. Some ranges are missing over a whole percent, where snowie says 6.4% but the combos it gives only fills 5.34%. Below it's over by .34%

Any clue?

Snowie uses mixed strategies for many combos, so sometimes a hand will be mixed between calling and raising, for example.
In your screengrab, however, it's listing JJ/TT/99 as calls at 100% frequency (no mixing). 1.36% is correct, since those 18 combos make up 1.36% of all possible hands. 18/1326 = 0.0136. What does Flopzilla say?