That's why I said it deviates from GTO but the principle is the same.Check for EV's results in solver and Snowie and then lock solver to 100% check and compare the results (also ranges).

No, well, kinda. It means that Snowie has a strategy for the UTG player that Snowie thinks is best against various strategies (including sub-optimal ones) of the 3-bettor, because the neural net trained against "random" agents, not perfect/optimal players.
e.g. If the 3-bettor had a particularly tight/strong range (i.e. he played tighter than Snowie would), then donking into him might be very bad.

Solvers like GTO+ find (near) perfect solutions for specific ranges. Snowie offers ranges that supposedly do well against "randoms" as well. This is one of the many problems with the term "game theory optimal" itself. A solver can produce an "optimal" strategy for how to play against a specific and known range/strategy, but if your opponent has a different range/strategy, the solver's recommendation is no longer perfect/optimal, since it won't maximise EV against every possible strategy.
Snowie offers strategies that should do well on average vs nits and lags, but this might not be ideal when it's up against a nemesis, or a version of itself.

I don't use snowie, *are* those really the same range? For JJ you have JcJd but there are 6 JJ variations. The reason this might be important is that there are 6 pair variations but only 4 XYs variations, so if you include just a single example of each hand type, then the overall distribution is skewed (it makes the XYs hands appear more common than they actually are)

(Also, snowie has not actually solved GTO. I don't know in the solver's case whether it can be considered solved either - From my understanding you can get an optimal (or near optimial) solution from "this point forward" but that doesn't actually mean that the node you're currently on is even part of the GTO solution at all. In essence it lets you find a solution to a modified game.

There is an update.Snowbot goes all in on training.Goodbye Challenges!!!
...and hello tournaments train mode.

On the day my annual subscription expired? It's so rigged!
Any info/screengrabs on the new features?

I haven't checked yet all the improvements but mainly ... http://www.pokersnowie.com/blog/2018...ining-platform
Customized spots r good improvement i think.

Weird coincidence Arty as mine expired today as well. I got an email from them telling me they've reactivated my free trial though. Not sure what that's about.

When in training mode using the Cash game training, the auto rebuy is not working. I've tried 100% and 90% but the stack keeps dwindling so that needs to be looked at.

Thanks for the link, disident. I ended up finding it shortly after posting.
Husk, I remember you tipping me off about the discount price a year ago, so that must have been why I renewed. I'm trying to resist the urge to sub again. The new features sound quite cool.

I haven't bothered this year. It's GTO+ for me these days.

Off topic but how are you finding gto+? I can’t seem to figure out why it’s much cheaper than pio and SP. are the solves as accurate?

Exactly the same results like PIO if the trees/situation that u r solving r same.Perfect for the money.Also is user friendly(easy).I was lucky to have bought CREV so I got it for free.
U can go to gto+ main site and check the videos.It is all explained.

Anybody tried new training features on snowie recently?

Hiya

Does anyone know how the "EV" of a move in training mode is calculated?

I.e. if i make a river raise worth 15 EV but calling is worth 16 EV, how does it work this out? Does it calculate what Snowie's whole range would do and then work out the EV for my specific hand? Or some other method i.e. how well Snowie's raise or call worked vs the agents it played against?

I think the software just looks at an analogous situation in its database and looks up what the final result was when it took each option with the particular combo you're looking at. This is basically how the "bot" makes its decisions. It looks to maximise its EV, based on previous similar situations.
e.g. If its database says all combos of AA won 6.04bb (or whatever it says) on average when it was raised UTG, and that's more than it made by limping, then it would raise AA UTG, since it has an expectation of making 6bb by doing that.
Any hand with a EV of <0.0bb (as a raise or call) is folded pre.

In some spots you'll find that betting one size will have a higher EV than another size, and it will still pick the size with the lower EV, but that's because its worked out which size generates the most EV for the range as a whole.
Where the EV of raising or calling is exactly the same, a mixed strategy is used, but if one option is definitively better (e.g. the 15EVbb raise vs a 16EVbb call), it will always do the one with the higher EV (a so-called "pure strategy").

EDIT: On a related note, Snowie's EV numbers are often quite different to what you might experience in real life. For example, in every database I've ever looked at, aces won at least 10bb per hand as a raise first in. Snowie's pseudo GTO play implies the rockets only win at about 6-7bb in the nittier games that Snowie "learned" to play.

Just bought pro subscription so I am looking for some kind off study plan for next couple months in order to fully take advantage of it.
I play line 2/5 and midstakes (200-1500) live tournaments.

Any suggestions how to incorporate snowie into my study plan for live poker?

So the evaluation isn't a measure of how well i am doing against Snowie, but a measure of how well i'm doing against the agents it played against?

So potentially I could have a positive winrate vs the machine but be rated as a "beginner". Now i know this is possible even if EV was calculated against Snowie's actual complete ranges (since you'll need to deviate from Snowie to beat Snowie). But this discrepancy can also occur for another reason: my rating is being measured by how well i do vs the training agents, but my $$$ is being measured by Snowie (which exploits the training agents).

i've emailed support to confirm on how the EV is measured. There will always be flaws but i just want to know in what way to adjust to it.

EDIT: this actually makes the most sense, since it explains why there are discrepancies between the hands in Snowie's range and the moves it recommends. Bit dumb though

Kind of. The EV numbers are how it decides upon the best play. If you make decisions that - according to Snowie - have a lower EV than other options, it will mark them as errors. When the pot is big, what Snowie calls "blunders" are more common. e.g. If the pot is 50bb on the river, and calling is worth 30bb, but raising is worth 35bb, then calling is a significant blunder, since you missed out on 5bb of EV. It's impossible to make a 5bb blunder pre-flop unless you do something really stupid like folding AA or you open jam 32o. Most pre-flop decisions will be measured in fractions of big blinds. Those pre-flop decisions are still very important though, but errors get compounded across the streets. It's just that with Snowie's scoring system (the "error rate" number), it will be the big pots that have the greatest effect. (You can get a very good/low error rate by playing as a massive nit that just avoids playing big pots).

You can play exactly like Snowie and still win (or lose). There's this thing called "luck".
When the bots are playing each other, using identical strategies, half of them win and half lose, natch. FWIW, I did over 25,000 hands of training with Snowie and somehow beat it without trying to actively exploit it. I think I just ran well.
While it's nice to have a "winning session" vs Snowie, you should ignore the $ amounts and focus instead on the errors, blunders and balancing frequencies, if you're trying to learn how to play like the bots. The $ amounts are obviously subject to variance. (e.g. Maybe you get lucky and Snowie spews off with air when you have the nuts, but it only makes that play 1% of the time in the long run).

Any plans adding MTT training option to snowie. Current Freezeout options only allows 10 player max SNG

I play 9 handed cash sessions(2500 hands) with snowie and I beat it but it still shows me at beginner level.
It appears that its recommended ranges are super tight. They recommend folding all pairs bellow TT in middle positions. It appears that such ranges would not realy work in live cash games.

Pokersnowie check raise with a 1/4 pot sizing on many flops BB against Button.

Has anyone ever tried this sizing on real tables? Do you think we can follow these tips?

It may be that something like 40% pot is better, but Snowie doesn't have that raise option.

But in regards to how effective it can be, most players will likely not realize how wide they have to continue vs such as small raise, or if they do they will not be familiar with playing turns with wide ranges after flop x/r.

Generally speaking any strategy that puts your opponent into parts of the game tree that they are not familiar with will lead to +ev situations.

Yes, great decision. My best recommendation is to practice 9 handed vs the challenger 200 hands at a time then upload and review where all your 1bb+ blunders are to find out where you're making the biggest mistakes. Try to get world class over 1,000 hands before you start deviating from it's recommendations. Also write down interesting decisions that you have at the casino in your phone, but be somewhat discreet about it, don't be antisocial with the phone and you don't want to overwhelm everyone with the fact you're constantly reviewing hands. Just a brief summary to help you remember later.

It's ranges work for me in live poker. You would be surprised how much $ I've seen regs lose getting set over set playing small pp's with 7-8 people left to act, especially on deep tables. Snowie is going to figure out whether playing those cards is necessary a lot faster than we will playing only 25/hands per hour live, and it won't be biased by results. Also remember that rake is higher in live games so if anything you want to play tighter than snowie unless you have position on a whale. Best of luck and enjoy the learning curve, don't get too discouraged!

Yup. I think the postflop minraise is good in spots where you're oop and don't want villain to take control of the hand and check back vs you on the next street. Or when you have a monster multiways but practically have the boardlock. Or just when villain has tons of air and expect to take it down right away frequently. I would say those are the three main reasons.

I routinely use 34% of pot (which is sometimes just a minraise) and I'm currently crushing in the micros. In my experience, most opponents play horribly against small bets and raises. (They either float too wide, with absolute no hope hands, or they fold too often).