Very interesting idea. Had not thought about this, but it's a very good idea. I'll give it some more thought.

Thanks!!

Well done on bringing this to market - it's a very neat product.

Understanding what the program thinks we should do in a given situation is less useful than knowing why. For example, why does it suggest checking 95% of the time in this situation:

I raise $3.50 from the HJ with TT. "Maria" from the BB calls. Flop is JhJs4c. I bet $5 on the flop, Maria calls. Turn is a 4h and goes check/check. River is a 6d and Maria checks the river to me. This seems like an easy value bet?

The program's stats suggest Maria has a full house or better only 12% of the time, and 2 pair 88%. Assuming no two pair is better than mine (because she would have 3bet QQ+ pre) and she never folds a full house or better, she would have to fold 76 percentage points of the 88% for her calling / raising range to be ahead of TT.

If she's really folding 76% of the time then I can bluff every single time, because she is horribly unbalanced. But I don't think the program assumes people are that unbalanced (rightly). Allowing her to c/r a % of the time does have to limit my value range a bit but not too much I think, because her value range is too narrow.

Interested in your thoughts.

Could you send me a hand history? Feel free to disguise your name of course.

Looks like for some reason, it's thinking that she's not calling the flop bet w/ over cards. Seems like that would happen at least a portion of the time.

In any event, looks like the only value we get from a bet is pocket 99's, 88's or 77's. (vs. the downside of J, 66's or 44's). Seems like a close decision, but a small bet might be in order. I suspect I would bet a small amount.

At this moment, the program evaluates a range of bets sizes in calculating an EV of bet. We are improving that so that so that it will initially evaluate a range of bet sizes, then take the best bet size and continue evaluating that bet size. This will be an improvement in (particularly river) bet sizing and bet EV calculation.

Thank you very much for posting this.

Sorry, I don't have the hand history - I was playing against the computer and have now quit. You're right about not calling with overcards on the flop, but given I raised from the HJ this isn't crazy.

My broader point is that it would be good to see more of the why. Seeing a detailed breakdown of the calculation would be really useful. How does it assume each opponent makes a decision? I presume it doesn't do simulations within simulations, because the game tree is too large?

One further question on 3-betting. I played around 100 hands and it didn't suggest one 3bet bluff, but it did suggest 3betting stuff like KQ and 99. Shouldn't its range be a more polarised?

I played against the best AI for about 20 minutes. I was not importing hands while I was playing, but then I realized that if I do not import hands, the AI might not be able to adjust to me. Do we need to import the hands while playing in order for the AI to adjust? If so, then it might be good for us to choose between using different Heros, so that we can have one to use while we look at the computer's advice, and another that we use while not looking at the advice.

So I started importing, and I was advised to steal from the button with 98o, so I did. Same exact hand a second time, same advice, so I took down the blinds again. But then the same hand happened a third time, and I was advised to fold 98o from the button. Did it do that because I had opened from the button so much that it was starting to tighten up on the button stealing advice?

If it is easier to do, then perhaps you could do it before tournaments so it would not have to wait? Shortest job first is a method frequently used by computers you know.

You are welcome.

Yes, we need to add the why. For now the best "why" is to look at what it's assuming your opponent has in the probabilities. We will add some "why", but it's actually pretty hard. We can certainly do it, but it's lower priority than adding HU and tournaments.

Exactly, it can't do sims within sims. We have an opponent model that predicts how any players will play any hand in any situation. Here's what we could do pretty easily....

Today, if I see a hand that's played poorly by our AI here's what I do. I turn on some logging that isn't currently part of the software. The software then writes to an html file all its assumptions about the hand including

• the probability that your opponent will take the actions he took with each of 1326 hands (or 169 preflop)
• the probability he would think you would take your actions with every possible card
• and the results of the simulation

From this I can see... Oh it assumed he folded all over cards. Okay, why did it assume that. Okay maybe I need to fix some element of the Opponent Model.

We could share this with customers. It's not currently very user friendly.

At some point, I'd like to provide our data about a hand to an EV calculator. If you are familiar with these (e.g., Stox / Card Player have one) they allow you to analyze a hand, but they are exceedingly complex and time consuming. We could populate that tool with all of our assumptions in seconds. Then you could go into that tool and change assumptions and recalculate.

Three and four betting is our programs biggest weakness. We have an awesome solution for this that's time consuming (it will truly rock). Because we will make this major improvement, we aren't making minor changes to it anymore. So you have identified one improvement and there are a lot more.

Having said that (other than exploitive moves) the preflop was developed working closely with a guy who averages well over $1000/hour playing online poker (i.e., one of the top 5-10 in the world!). This doesn't mean it can't be better, but the core preflop strategy should be quite sound. He won't let me share his name w/o paying him a lot more than I'm willing to pay him.

Precisely. I think you must be importing so it'll adjust to you. I'll ask the guys and if I'm wrong I'll get back to you.

In any event, you can do two things:

• your idea (i.e., have two names and use one for getting advice and one for not getting advice)
• or you can select only import hands when advice screen is closed

I wish that was why. Right now, we don't have a table image function so if you've been raising a lot, it won't change the advice.

I'm confident that it wasn't a steal opportunity. I bet someone else raised and you didn't see it. We need to make opponent actions clearer -- maybe adding chips or making the best amounts on the table another color. I've seen people have this problem -- precisely what you are saying... (i.e., why wouldn't I steal with A9o!? -- because someone else raised)

The only other reason (which doesn't match what you describe) could be that these blinds started defending more (i.e., they became less tight to steals attempts.) It would alter the advice for this; however, I doubt it ever tightens up so much to fold 98o on the button w/ a steal opp.

The program dose not allow for more than one Hero. There is a drop-down for the player names that says "Hero," but it is disabled and so cannot be changed to another Hero. Maybe that's because it is the beta version?

Sorry if this has already been answered, does your software work in limit games?

Thanks.

hi, this is the great minds think alike guy tee-hee, will definitely check it out, thanks bro, you the bestest, that's coming from the top, asbestos!

No it wasn't asked. But it does not. Sorry.

Nope, I am still with NC which is STILL awesome.

I am presently in "learning curve" mode improving my ability to script.

I'll be baaaack!

T

This program is interesting

I am still trying to understand how it "thinks" and whether, when we disagree, it is a flaw in the program, my awful play or a difference in assumptions.

I found a few deviations today and some of them I agreed with 100%. So, even a long term winning player like me can use the program to pick up clear errors in their own play

However, this one was a bit strange:-

I PFR Jc8c OTB and BB calls. Flop is Js9d8s and HGP suggests checking this back as a "slow play". To be fair its EV was only a small amount in favour of checking and I know that you run the simulations a few times and that it could have been different another time.

However, I could never imagine checking here a a "slow play". We may have a terminology difference but this nowhere near meets the defintion of a SP as discussed in TOP.

My hand is probably best (HGP says the opponent has a better hand 9%) but I am very vulnerable and will get called/raised by many more worse hands so I have clear value here (IMO).

I fully appreciate this might be my bad poker thinking so I would not mind comments from either the developers or other poker players to explain why checking is so close here.

However, having bet and been called, the opponent donks 1/3 pot on a turn K that produced a 2nd flush draw. I raised ('cos I'm a fish) and HGP again suggested not putting extra money in. To be fair it now thought we were behind 34%. I thought the small bet was more likely to be a "price setting" bet with a draw. All the K had really done to change my flop view of his hand is to put KJ in his range to get past me (K9 and K8 were not likely). Just because he fires 1/3 PSB does not suddenly take him from 9% to 34% IMO. He called and we checked it down when the flop flush draw hit and he showed T2s having turned a flush draw to go with his 1 card oesd (perhaps K9 and K8 were in his range after all)

I am not being results oriented; just trying to understand what the AI is doing.

T

TJD,

That sounds wrong to me too. The only explanation I can think of is balance. That is, a super wet flop like that is one I'd probably just give up on with certain hands, so a check back with some strong hands balances those giving up checks. I don't think that's actually a good enough reason to check, but perhaps HGP does make those kind of considerations.... Authors?

Would be cool if there was a way to set for example your hole cards, and flop cards, and then play the hand over and over multiple times.

I like the continuation bet here as well. Can you
(a) right click the hand (in the hand window) and re-import. Maybe the simulation was off by a few standard deviations.
(b) post the hand history (feel free to disquise names), but if you do please post as regular text so I can re-import and analyze.

The way HGP balances play is the following:
(a) by recommending to check x% of the time when bet is better (for example) and
(b) doing so more frequently with strong hands that aren't quite as good as a bet (despite the fact that bet is better)
(c) the simulation naturally has some variance.

I think the advice should be to bet most of the time here; however, the program doesn't do things like "if c bet opportunity and very strong hand, and one opponent, then c bet x %". It simulates the hand out many times and makes the recommendation based on that simulation result. When I see moves that I would not make it's usually one of two things:
(a) a mistake in the program's assumptions or
(b) a very interesting learning opportunity.

Here's the way I would recommend thinking about the advice. Assume you have a friend who's a solid winning player at $50 buy in (3+ big blinds/hour) 6max and full ring (would make about 1 big blind / hour at $100). This friend replays every hand you play and you get to compare. He didn't learn poker the way anyone else has (he never read a book or watched a video) -- so sometimes he plays unconventionally.

Yes! We actually have that ability in the program, but we turned it off to get approval from the poker sites. I think they would have been concerned that someone could quickly input a live situation and get real time advice.

Maybe we could put that back in, but make it automatically disabled when the poker client is turned on.

We'll ask them.

You're right HGP does not allow for more than one Hero and you can't change his name from the window where you configure tables. But you can do it directly on the table window: right click on Hero/Edit player and change his name. Be carefull, this will change Hero's name on all tables.

I think we need to let the user choose different Hero's name at different tables.

Thank you for reporting.

Where is this database currently stored? Is it removed automatically upon uninstall?

I also noticed there's no way to purge data from the database - are there plans to add this functionality soon?

The database is SQLite and it's in the application. You can find it here:

C:\Users\paul\AppData\Local\HolyGrail (at least on my PC)

It's called hgp.

I think different operating systems may locate it in a different place. Tell me if you can't find it here, and I'll find out where it is on your PC as well.

Yes, right now the only way to delete a hand is to right click it and press delete. We'll add ability to delete easier / purge.

Under vista the database is located in
C:\Users\<User Name>\AppData\Local\HolyGrail
(The path is something similar on win XP and 7, look for HolyGrail folder)

The database is NOT removed upon uninstall

On Analyze tab, you can manually delete hands from the DB, by right clicking on a hand in the hand list. I agree this is not really user friendly if you want to delete a lot of hands.

If you want to start from a fresh DB, you can simply delete or backup the HolyGrail folder from the above path and then reinstall.

I re-imported as you said and over another 10 occasions it thought flop bet was better 9 times. On some occasions this difference was very large.

It "accepted" my turn raise 3 times out of 10 but each time it still had the EV of a call as slightly higher.

This raises 2 questions:-

1) How far out does a deviation have to be before it decides an action is "wrong"

2) Does it consider bet sizing as a "tell". Clearly the program is saying my turn raise is wrong (and it may well be, especially seeing the probable width of his range) but part of my "read" was his small bet which although it is sometimes a trap is much more often a weak hand. Does the AI consider that?

I prefer not to post a HH on here to protect the innocent. However, if I could send it to you directly for investigation and you undertook not to use/release it for any other prupose, I am fine with that.

Cheers

T

Hi, I'm in charge of the GUI for Holy Grail. Not sure what you mean, by "two Heros". What are you trying to accomplish?

Best,
John

When playing against simulated game against the AI, we have three import options: 1 all hands, 2 no hands, and 3 only when playing with advice hidden. I wanted to play using Hero while I import all hands (or only when playing with advice hidden), and using Hero2 while I import no hands. I think I figured out how to do it, but it was a bit confusing (I had to right-mouseclick when I was playing at the table and create a new hero that way).

The goal is to be able to play using one of them while I look at the advice, and to play using the other one while I do not look at the advice. I will look at the advice when I play using Hero2; but I will not look at it when I play using Hero and import at the same time. When I play using Hero, and do not look at the advice, and import, that gives HGP the ability to adjust to my play. But if I start looking at the advice while playing and importing with Hero, then HGP will not adjust correctly to my play, so when I want to look at the advice I switch over to Hero2.

As far as the GUI is concerned, here is how I think it would work well:


That drop-down box that shows Hero playing is disabled, but it would be better if it was enabled and allowed the user to choose Hero, Hero2, Hero3, etc. (There also needs to be a screen/window for creating those other heros too.)

By the way, here is how I ended up creating (and switching to) Hero2, but I do not think it is good to have to do it like this: