Join Date: Aug 2014
Posts: 11
I am a barely winning live low stakes cash player who occasionally plays at big MTT and min-cashes at a respectable rate. I began my hobby in CO before the state changed the $5 limit. This meant that I learned the game by playing the daily tournaments where I managed around 18% cash rate. The cashes and wins combined enough to make me a winning player (barely).
When I transitioned to cash I got creamed in the beginning. The TAG style I played at that point was fashioned from short stacked turbo tournaments making it easy to set mine and play speculative hands against me in position. I eventually stopped the bleeding and learned to quit stacking off on overpairs and TPTK. This new style played well in Vegas NLHE but broke even at the spread limit games in CO.
At the 2015 WSOP I was a 3:1 table (not event) chip leader in the monster stack event when a talented LAG table changed and sat immediately to my left. He proceeded to play what seemed like a VPIP/PFR of 50/30 and shut down my feeble attempts at being the table bully. These were the last three rounds of the day and I decided I just wanted to bag the chips and play at another table tomorrow. The cost in blinds and missed action ensured that my stack was back to mediocre the next day and I went on to bust a few rounds before the bubble. This was a traumatic event for me and I decided that I had to learn the LAG style.
I discovered Poker Snowie and decided to give it a try. The longest thread on this forum about the program criticizes it for the false claims of GTO play. I agree 100% about that line of criticism, but still believe it has value for players like myself that still need remedial work and a good statistics package and hand replays for deep study. I am in a state without access to online poker and my attempts back in 2008 at NL50 were steadily losing, so I figure that simulation is better than nothing.
My tournament game was probably 15/5 in early position that expands to 25/10 near the button. I started training with Snowie and I have played it for sample sizes of 10K hands around 8 times now. I lost the first 7 attempts at beating Snowie both in results and error rates. The 8th attempt has me showing a profit of $8,169.00 for 9K hands but with an error rate of 17.86 which it scores as intermediate.
When I alter my play to fit into expert or world class play on Snowie’s scale, I find that the results are modestly winning.
However my new LAG style 36/20 scores lower on error rates but shows better profits. The following plays score terribly on the error rate, but are more effective against the extremely tight ranges (16/13) that Snowie plays at 8 player 1/2 cash challenge: calling 3bets with pocket pairs, floating the flop on static dry boards and betting or check/raising the turn, 4bet and 5bet raising ranges with position, raise/folding difficult turn or river boards, cold calling 3bets in position with high implied odds hands.
Snowie recommendations that seem questionable to me include: folding far more than 67% of your 2bet range to a 3bet, outright folding small pocket pairs even when deep stacked in middle position, folding more than 85% of your range to a button raise in BSBB situations, never 3 betting that UTG raiser despite the fact that Snowie seems to do that around 30-40% of the time, calling overpairs against 3 streets of betting on virtually any board, continued 1/2 pot bets on static boards with no draws that result in 2 or 3 barrel bluffs.
I think that Snowie may be off because they don’t really explain the equities calculations shown in its 1/2 pot, pot, and 2pot betting options. I think these ranges come from its huge sample database of bot v bot play where the real players input is still small and skewed toward lesser skilled players like myself.
I expect that the program will progress as more real players contribute to the database. So my analogy is that Snowie is a TAG player from around 2008 that hasn’t really grasped the changes in style and analysis contributed from the online grinders in the last several years. In other words, the GTO approximation resulting from machine learning is flawed because the elite online players are not messing around with this program enough to have trained it correctly.I have noticed several posters on this forum making this exact same point.
So I think that Snowie training provides value to a learning player, but the live advice and recommendations from Snowie cannot be trusted without running your own analysis using Flopzilla or Equilab (or whatever).
My challenge is to post an average of 5 hands a week where I argue with Snowie’s live advice and see if I can logically demonstrate why I feel it is wrong when contrasted with the ideas of Ed Miller, Jonathan Little, and Dan Harrington’s recent works as models of “correct” play.
