Another problem is it always seems to be a hand behind the hand that i am actually playing. For instance at this very moment it says i have 23 off. Those are the cards i had last hand.. I infact have 92 off at this moment.

Just to be even more clear, when you buy this program you are contributing $ to a botter. then when you suggests improvements to the program after using it you are contributing $ to botter, than when the botter figures out how to use the money you bought the program with, and the intelligence you offered to improving the program to apply it to his bots at the real tables, illegally, you are contributing $ to bots, since they don't tilt or make errors.

This is by far the scummiest display of a person trying to destroy online poker and scam people out of there money IMO.

not sure how this program is suppose to help me at the tables when it is always a hand behind and doesn't actually give any advice. It just shows me my cards and the board cards.

How did ZeeJustin, Aba20 cheat?

A lot of pious people on this thread

was there a 3 months free trial thing for 2p2 ppl? if so, is that still available?

Hi. It doesn't give you real time advice that's why you only see the last hand.

What makes poker hard is that you rarely get good feedback w/o hiring a full time poker coach. However, this software will tell you if you make a mistake immediately after playing a hand -- similar to having a poker coach.

yes, we had one -- but we aren't running it any longer. You can get a 2 week free trial here.

The advice window only provides advice when you've replayed the hand to the point in which the user needs to make a decision.

In the monitor window can you see Deviations? You might need to scroll to the right to see it. you can also select what information you want to see in the window with the little arrow on the top left of the page below "Play".

You can also see deviations or errors in the actions window. Deviations will show up as blue and will quantify the error. If you click on them, then you'll see that situation in the replay table and you'll be able to see advice and probabilities (i.e., what you should think your opponent hands).

I'd also recommend watching the demo video here (if you have not already done so). I think it will be quite helpful.

Is there as definition anywhere of "okay"?.....and if "okay" means the hand could have been played better is there any way of learning how please?

I just dont find "okay" very helpful...unless I am understanding it wrongly.

"Okay", means the advice differed, but by only by a little (i.e., The EV difference was less than 5% of the pot or the action you took was recommended > 30% of the time). In those instances the program doesn't define that as a "Deviation". Said differently, the program is saying it thinks it was a close call.

If you replay the hand, you can see the advice in the Advice window.

I think i totally misunderstood the purpose of your program i'll fiddle with it later and try to get it to work.

also is there anyway to import a pt3 or hem DB?

ok i got it working but i already see a flaw as far as nl is concerned. It clearly does not consider stack sizes. I just reviewed a hand where i A2 off on the button and it was folded around to me. Grail says i should fold 90 percent of the time. That is bad advice because my and sb both had over 240xbb. Also sb will stack off with any over pair or better. With out even running stoxev i'm pretty sure it is +ev for me to call and try to flop 2p+. Or is grail meant strictly for limit?

It considers stack sizes as short, medium, and deep. But deep is 100bb's or more. So you are right, it should play more hands 200 big blinds deep, but we haven't added that capability yet.

Not yet. We plan to add an interface to make such imports much more user friendly -- but we aren't sure when it will be complete.

I am sorry to say that the new update does not solve the major issue I have with the software and once again I have stopped testing it because it is is a waste of my time.

I doubt very much whether any experienced player would use this program in the "Monitor" state. As was explained by an earlier poster, having some weaker player looking over your shoulder and constantly saying "that was wrong" is not conducive to concentration at the table especially when the "advice" that is given is incorrect (or at best marginal) a large % of the time.

So, how may a stronger player wish to use this program? During play, we all make mistakes and in addition we may have some leaks. If someone we could trust reviewed our hands and came up with solid advice, we could look at their comments in the cold light of day and see whether there were any lessons for us to learn.

During play using Monitor mode, you obviously have to give very quick feedback because that is the whole point of the exercise. The "value" of this advice is often doubtful but it will frequently be better than the poor play of a relative beginner. However, in post play analysis there is no need to be so frugal with time spent calculating.

In this version you have added different simulation levels and I ran a block of hands using "Slowest/Lowest Variance".

However, the issues of lack of "consistent" advice remain. I played 1600 hands and reviewed them afterwards. HGP came up with 34 Deviations.

I looked at these. Some were of course mistakes by me; I AM human. Also, some of the plays I made may well have been specifically geared towards an opponent's tendencies that I believe I knew better the HGP. However, when I looked at those where I thought I had played in a "standard" manner and was doubtful about the deviation I re-imported the hand several times. I got exactly the same problem as before and that is that the re-import often said my action was NOT a deviation.

This is a complete waste of time! If HGP thinks an action is incorrect (deviation) and I disagree and HGP STILL thinks it is right after I re-test multiple times then either I have a leak so HGP was useful OR we have found a "problem" with the AI that we can address to make future advice better.

The current situation does neither. If I think HGP is wrong; I re-import and NOW it agrees with me. That is no bloody use at all. All I have done is wasted my time.

I do not WANT to check through numerous "deviations" only to find out that it was not a deviation at all just too small a sample.

I realise this may be an insurmountable problem in which case you are simply producing a product that would never be bought by me because the time spent is too great for the benefit gained. However, if it CAN be resolved then this could be an excellent tool for serious players. It would take time before the AI was improved enough to get it "right" often enough to satisfy competent players but all products (should) get better as time goes by.

I see that this thread has not received much interest recently. That could be that 2+2 members have decided to have nothing to do with you or it could be that , like me, they have come to the conclusion that what may at 1st glance may seem to be a useful piece of software, is in fact of no practical use at all.

Cheers

T

PS The Analyze screens do not deal with preflop. If I filter for preflop all these screens are blank. This is what I would expect BUT the fact that we cannot filter the Replay screen to include "Show only hands with deviations" means that to see whether it disagrees preflop only, I need to scroll through the entire hand list - not good!

The way I would use this, if it were possible, is to request the Replay screen to only show hands with post flop deviations and/or differences of opinion preflop. I would then replay those hands.

Once again, the feature where you attempt to differentiate by classifying mistakes in the reports tab may be of some interest to a beginner but is just annoying to me. It seems as if General, Action and Strength include all deviations but to see and replay them I need to choose a tab and then a minimum of 5 "Hands" buttons (Action Tab) to get hands. I resize window; replay hands; close window and then click the next "Hands" button. THEN if I want to see what HGP thought preflop I need to go back to the Replay screen and scroll down - ugh!

T

Can you send a hand history that switches advice during re-import. This should very very rarely happen with Slowest/Lowest Variance (and only when the decision is super close).

Thanks.

We are adding a filter for deviations. For now, one can use replay and then sort the Deviation column and review only deviations. You sort by clicking on the column.

Let me try to clarify, because it will happen sometimes when decisions are very close. Not all deviations are the same. I you are reviewing hands, look at the $ amount of the deviation. If it's a big $ amount (relative to the pot), then it's a big deviation. If it's very small, then it's a very little deviations (i.e., not much of a difference in value between two different actions).

If it's small think of it as your coach says, hmmmm it's close... you could do either, but I slightly prefer a check. Then maybe when you review the hand later with the coach he says, hmmm it's close I said I slightly prefer a check before but upon reflection I think your bet is fine here.

Maybe we should allow the users to avoid seeing small deviations so this doesn't happen....

We have specific cuttoff points between a "deviation", "Okay", and "not a deviation". We could increase the size of "Okay".... However, no matter how big we make Okay, there will always be the point (i.e., the penny) when "Okay" becomes "deviation'. To help with that we show Expected Value, so that you don't need to worry about deviations that are small -- but focus more on deviations that are larger.

BINGO!

If the "deviation" IS really close then it is extremely likley that small sampling errors will switch the result from one side to the other. In the end it is NOT a deviation at all when all these samples are amalgamated.

It is of no use (IMO)to a user to see that HGP thinks it is a deviation and then a moment later it is not.

If you look at threads in the lower stakes forums that discuss hands that take up the most space you will almost always see one side of the argument or the other being defended as 100% right. Sometimes, once they have dragged on for a while a more experienced player comes into the mix and says something along the line of "it's too close to call".

As you are well aware, decison making could be 100% perfect if we knew with certainty what his range is to date and what he would do with every part of that range. When discussing this with students I use the word "parameters". The estimation of these parameters is at the heart of good poker and the really good news is that we can never be 100% right or the game would be too easy and we would be over run with bots.

When a similar EV result is acheived for a "reasonable" range of paramaters we have a meeeh! situation. The errors in our estimation of parameters are all reasonable and the result is either +EV or -EV depending on the parameters we use.

Unless a mistake is CLEAR I do not think it is worth while mentioning; it is just hot air.

What we need to see is a LARGE difference that remains even over a "reasonable" range of parameters. Now we can say to our student - nope that is a leak! We can also justify it by looking at the results at any set of paramters in this "reasonable" range and show that the decision was wrong for all of those.

T

Nicely played. When importing our pros hands (Default sim), they get a deviation rate (postflop deviations/hands) of 2.3%. They are two of the all time best $200 buy in players (i.e., better than our AI) so some of this is our AI's errors, some is due to mixing up their play, some is due to them making meta plays that our AI does not do, and some is due to them knowing their opponent better than our AI, and some are their errors).

Yours was 2.1%!

You must know what you are doing -- or you make the same errors our AI does

If the price was much lower I would say "nice piece of software" and would buy it right now. But as it is I don't find much use to it. It doesn't really work well in micro and small stakes because it assumes villain's play is one thing when in fact is the opposite.
I tried for one day and I saw lots of bad advice when it was monitoring my play. It also recommends a very tight range pre flop, telling me to fold almost all connectors in CO or BTN. What's up with that?

For the price:
.It should tell why it is recommending certain actions.
.We should be able to export hands so we can analyze them in HEM and tune our game. (I don't even know if I'm wining against the bots when following the AI advice or if I could make more in certain spots playing it a different way. I played a bit a -EV aggressive style and started making more $).
.We should be able to configure the AI every way possible using HEM stats, like make a bot who never folds to 3bets pre flop and tries to steal the pot when opponent misses CBets, etc, etc. (This way it could simulate other type of opponents against who we play against on a daily basis. I mean, 90% of my opponents don't play a +EV game). Either this or hire/get the databases of very good players at different stakes and let the AI study and decide what is the best play at NL25, NL100, and so on.
.I also noticed the AI takes way too long to adapt. I think I was 3 betting 80% of my hands for hours and the AI just kept folding except when it had AK or better).
.It should also present odds on screen. Like, you are getting 3 to 1 on your money, meaning you need to be right x out of y times so you should fold/call. (great way to memorize certain spots, specially for people who would want to buy this type of training software)
.The graphics are horrible, it could really use something like that mod shown some posts back.

My point is, if the software is still in beta stage and there is so much to improve upon the price should be drastically lower for now. People can use the much cheaper HEM or PT and play the player perfectly at different stakes and make so much more $.

But yeah, this software has great potential and I believe it will be worth the price or more when it gets tweaked but for a beta software it's price is unrealistic, specially when the type of players who need it are probably still learning the game so they are micro/small stakes players or higher stakes players who are break-even or losing.

Anyway, let us know if or when anything changes. I'm definitely interested if either it gets heavily tweaked or if the price changes until it does. Stay cool. : )

We recommend stealing with a lot of hands; however, we adjust it by stack size, fold to steal % of opponents, and of course position.

Deep stacked against tight opponents on the button we recommend stealing w/a ton of hands. Even when the opponent has average fold to steal attempt numbers we recommned stealing with hands like 87o, 64s, 96s. You are right that 76o, 65o and 54o don't make that list. We'll recommmend stealing with some of them against tight blinds, but not in normal cases.

Here's the frequency with which our pros (who make about $100/hour multi-tabling at $200 buy in) play on the button w/ some weaker hands...


96s steal 93% (we steal in most cases)
A5o steal 87% (we steal in most cases)
980 steal 85% and fold 15% (we steal in most cases)
87o steal 68% and fold 32% (we steal)
76o steal 65% and fold 35% (we barely recommend folding against regular opponents, but stealing against tight opponents.
65o steal 39% and fold 61% (we usually fold)
54o steal 15% and fold 85% (we fold)
430 steal 1% and fold 99% (we fold)

These pros are slightly profitable on these non-suited connectors, but not by much. So if you are going to make much money with them then you are going to need to be a lot better than your opponents in the blinds and/or they better fold a lot.

Non-suited connectors are overvalued by most people in our opinion. I think Harrington recommends stealing pretty deep with non-suied connectors; however, some of the top pros find these weaker non-suited connector hands to be unprofitable. One of the reasons that they are weaker is that when they make straights the straights are quite visible to anyone.

We are huge fans of Harrington and think he's done more to advance poker education than just about anyone; however, this is one of the only places where we (and our pros) respectfully (and just slightly) disagree (i.e., with some of the weaker non-suited connectors).

Also, given that most of our customers aren't nearly as good as these pros we wish to err on the side of conservatism.

No disrespect, but you realize that your AI is drastically worse than what is out there?

Current 'state of the art' can do opponent profiling far faster (can do basic profiles in one or two orbits), and are far better at exploitative play. If you meant one of the best 'publically available bots' then your claim has a bit more merit.

If it took a team four years to develop that is rather insane.

I developed a bot of similar skill in about 3 months of serious effort (20 hours a week or so?) and that included having to write my own screen scraper and mouse control librarys (I reused some code that you can't - ie the java framework that Aaron released; some LGPLed code for poker calculations and neural networks) and that was more than 10 years ago (shortly after Aaron put out his paper on using ANNs for opponent modeling. Although his paper doesn't make use of a huge amount of context which is why poki doesn't get near as good of results as it should). I guess I'm curious what the heck the other 3 1/2 years was for

As to the guy who was afraid that this will be used for bots. The poorer skilled bot writers might use it - but only on the cheapo sites that have poor bot detection methods (any bot writer whos bot isn't superior to this can't think/code well enough to get around any serious attempt to detect bots). If you are playing on the larger sites that have more sophisticated bot detection methods then you are likely already playing the bots that are superior to this one.

For the curious there are probably around 2000 unique bots out there - although the majority (90%) are the weaker bots - either direct clones of poki and other UofAlberta bots (coders could extract the .class files from Poker Academy and use those directly or write their own implementation); or lookup table/heuristic players (when I wrote my bot; I wrote heuristic players based off of most of the major books at that time to have my bot practice against).

That still leaves about 200 bots out there that are equal to or better than this.

At a guess I'd place around 1% of 'players' as bots (bot owners looking to maximize have multiple bots per site, and bots at every site that they can bypass the anti-bot measures on - so while the number of competent bots is small, they use a bunch of them - ie 10 sites with bot distribution of 30% are net losers; 50% are small winners but the bot owner can't manage more than a small number; 10% are good but the the bot owner can't manage many and don't have good anti-anti-bot skills; 10% are good and the bot owner manages a lot). Also assume 8 sites with trivial security; 2 sites with good security. So .8*2000 = 1600 bots single site. Then the good bot players who are bad at anti-bot measure avoidance .1*2000*1 bot per shift per site*4 shifts*8 sites = 6400. Then the good bot players who are good at all of it .1*2000*10 bots per shift per site*4 shifts*10 sites = 80000.

So 80000+6400+1600 = 87000 bots total online is my best estimate.

100 million online poker players/87000 = 1150. So by my rough estimate above that comes to 1 in 1150 players is a bot. So 1 for every 100 full ring tables roughly. (Hmm that seems really low but the number of online poker players is high, since not that many are likely active; and my number of estimated bots is really conservative since there is no reason for a bot owner to not try and have a bot at every table it is profitable for).