While I was building the hand history advisor, I got through about fifty SNGs, or 2,000 hands.  I started awfully, picking up prizes here and there but losing my initial $30 deposit in a couple of weeks.  Undeterred (read foolishly), I learned a little more, thought a little more, and stuck in a second $30.  This time things went better – I played tighter, stuck to $2 games, and picked up some decent returns.  I could correctly apply pot odds and better identify when I was behind.  But I was still a losing player.

This is when I started to focus more on building my app.  I still dipped into the odd game, but had enough information to know what I wanted to build and how I would piece it together.  I started reading up on poker bots, and came across a recurring theme.  Apparently most bots are very simple, and are built to play ring (cash) games rather than tournaments.  For the uninitiated, a ring game doesn’t follow the same structure as a tournament.  There is no start or end time and the blinds don’t increase.  You buy some chips, sit down, and play until you have run out of money or something good comes on the Food Network.  I thought I should give it a go.

The suitability for bot play is immediately apparent – you don’t have to factor in the stage of the game, size of the blinds or number of players remaining.  Every hand starts out the same.  I’ll write another time about what this means for my project, but for my poker game, it has worked miracles.  The tables are generally tighter than in a tournament but there are invariably a couple of loosy gooseys that chuck money in to half the pots.  The key is identifying these players and learning how to draw the most money from them when you hit a hand.  Some players  will try to call every hand pre-flop, but a raise will scare them off.  Others can’t bear to let go of that initial investment and will throw good money after bad.  Some opponents call a lot of blinds, but very rarely raise – when they do you have to take them seriously.  Others ONLY raise or fold.  These guys you can wait for a big hand and re-raise.  There’s a whole post’s worth of stuff I’ve learned over the 4,000+ hands I’ve played.

Ultimately, I got to break even point.  Not on my original investment, but in the ring games I played.  I hovered in profit and deficit, but so many hands is a fair sample to draw the conclusion that I wasn’t a losing player anymore.  And then I went back to the tournaments…

I feel like I have been through an abusive crash course and returned fitter and ready to conquer.  It is much easier to identify the sharks and the fish at the table, and my new observational skills pay out much more compared to the big boy cash table, where I never really fit in.  I’ve cashed in nine of the sixteen SNG tournaments I’ve entered and if anything felt like I had bad luck.  Even including rakes, the numbers look astonishing – an investment of $31 and a return of around $60, or 100%.  The sample size is small, but If I maintain even one tenth of this form over the next week or two I will be a winning poker player.

This is all a fine yarn of course, but I know what you’re all thinking.  SHOW ME THIS IN GRAPH FORM.

These images come from Poker Copilot, which I definitely recommend for hand analysis. Here’s that first $30 when I didn’t really know what I was doing:

The odd prize (and even a couple of decent wins) but a definite trend downwards – I was typically cashing in one in five games or so.  The data under this shows that I was seeing too many flops, not letting go of underdog hands, and making a lot of mistakes.  So I read some poker strategy online, slowed down a little and returned with a much better game.

These twenty tournaments are nothing to write home about, and I’m not quite breaking even, but at least I’m taking a longer time to lose.  I’m applying pot odds, trying to put a player on a hand, and most importantly sticking to low buy-ins instead of jumping up a level whenever I win a couple of SNGs in a row.

Then I hit the ring games, where I finally learned to play poker.  The sample size here is almost 4,500 hands so is pretty reliable.  I lost some big hands early on and had the looseness beaten out of me until I was more-or less even.  There’s a lot of swing up and down, and I was in the red more often than the black, but the quality of opponent at these tables is pretty decent.

Back to the tournaments, and here are the twenty I played post-ring game.  A definite trend, though an admittedly small sample size. Tournaments are hard to analyse over weeks because each one is over an hour long and hundreds of hands, but a single result – win or lose.  I don’t expect to keep this form but it’s promising, and I don’t feel like it’s all regular variance.  Watch this space to see me crash and burn.

So here’s what’s new:

• The tournament is selectable from a drop-down, and the players in that tournament are displayed down the left hand side.  The rest of the view details the pre-flop play for the selected player.
• Each action for that player is given some context.  What position were they playing from?  How many chips did they start the hand with?  This is important – an all-in raise from late position with 30 chips isn’t the same thing as one from under the gun with 20,000.
• With this in mind, the bet amount is also displayed as a percentage of chips and as a multiple of big blinds.  A raise of 300 is more aggressive when the blinds are 15/30 than when they are 150/300.
• Position is also displayed as a percentage index.  Furthermore, this is colour-coded for early/mid/late position.
• The card data is formatted to look a little bit more like, well, cards.

Just about all of this is achieved using bindings.  More than just content can be bound to data.  The list of tournaments, for example, is sorted by date by setting the sort-order binding.  It’s probably one of the most impressive features of cocoa and one I’ll miss when I have to switch back to another environment.

I’m going to finish up my bindings crash-course by creating a dashboard view with many inter-connecting tables.  Once this is accomplished I am looking forward to banking the experience and starting on something new.

I previously mentioned the clearest advantage a computer playing poker has over a human counterpart.  No matter how many simultaneous games it is playing and how much it’s had to drink, a bot never forgets.

In a brief dalliance with multi-table play, I immediately recognised the disadvantage you are at when your attention is divided.  Even less forgivable, my mum Skyped me while I was playing recent game and I didn’t feel like explaining that I’d have to call her back because I was busy gambling against other jobless folk for $2.  Didn’t work out well.  So don’t hold transatlantic telephone conversations while you’re trying to play poker is the (admittedly obvious) lesson, nor let any other engaging distraction draw your attention from the action.  Incidentally, Beatles Anthology DVDs are ok, but only if you’ve already watched them all a million times.

There is no real danger of missing your hand or action of course – FullTilt is not shy to summon you back with a serious of increasingly ‘friendly’ chirps.  “Coo-ee!”, goes the 15 second warning message, “don’t forget you’re losing over here!”.  What you’re supposed to be watching for, though, is table image.  Who is calling almost every blind and hoping for luck on the flop?  What did DrPoker guy have last time he went all-in?

I find this a real problem online.  Not only are the opponents strangers represented by intangible strings of text and depressingly cheery avatars, but there is little of the conversation or poker banter that associates a real-life game.  These visual and audio cues are the little hooks that we all hang our memory hats on, and without them we’re just confused and adrift and our floors are littered with hats.  In fact, on the rare occasions that one or two online players exchange more than a trite ‘g/l all’, I find that I am much better able to remember what kind of hands they have been involved in as the match progresses.  I am subconsciously assigning rounded character profiles to each that speaks up, and a narrative to the game as a whole.  Most often I’m just remembering how Gooner69xoxo said something racist about Mexicans and that I smiled when he lost a bundle to the slightly less objectionable LOLCATZ4EVA.  Had they played in silence, though, I would have trouble remembering who went in with what, how much they put on the line, and who it was i’m meant to be lmfao at for getting pwned by a n00b.

And this is probably what I’m most happy with about the first version of my poker advisor.  At the simple click of a button (after the less simple typing of a long table ID and an even less decipherable player name) I’m looking at a flawless recollection of how you bet.  Even though at the moment I am only reviewing the pre-flop actions of each player in the current game, it is surprisingly telling.  Look at the two profiles I’ve posted above.  After just 15 hands you can tell these are two very different types of player, despite knowing nothing about their cards or if they have to pick their kids up in half an hour.  Once I can find and import historical hand history data, and work out what I’m meant to do with that information, I will be researching which  dinosaur skull I want to buy for an ashtray.

The next step though is to start logging hole card details.  The information in these screenshots will be much more useful when I can also see what card the player was holding (when played to showdown).

Although my immediate goal is to build something that simply reminds me how my opponents have been betting thus far, I want to make future development as simple as possible.  The last thing I want to do is design, build and test a functioning application only to throw it away and start over again.  This is where my selection of candidate objects and attributes is important.

For anyone not familiar with the object-oriented paradigm, the OO101 example is often of a person.  The idea is to make ‘classes’ of objects that are typically real-world entities and generally nouns (PERSON).  That class then has specific attributes that make it different to other members of the same class (Name, Age, etc).  A class might also represent a collection of other objects – a ‘FAMILY’ might be a collection of ‘PERSON’ objects.  Relationships exist between objects of different classes, so as well as a PERSON object being part of a FAMILY object, one PERSON might be the ‘Mother’ of a second PERSON object.  I use the quotation marks not to denote sarcasm but to highlight a formalised relationship between two objects.  For example, Sarah Palin is the ‘Mother’ of Trig Pain. ‘Honestly’.  Finally, you might want to do something to the object – increase a PERSON’s age once a year for example.  For this we apply a Methods, for example person.increaseAge.  These are our verbs.

So getting the right classes is important.  Here is what I’m thinking for my first poker app (you can click to enlarge if you so fancy):

A POKER HAND is the first class I have built.  This represents a single hand of poker (unsurprisingly), from posting of the blinds to the eventual determination of the winner.  Every poker hand has a small blind, big blind and a collection of players.  On FullTilt, there is also a unique hand number that I want to collect, in case it’s useful later.  The POKER HAND’s array of PLAYERs each have a name, seat position and chip count.  I also want to store a succession of actions that take place.  These ‘actions’ have their own attributes so I have made an ACTION class that stores the player, action type (post/fold/call/raise/check) and the amount associated.  So if JohnnyPoker raises 200, my attributes for the ACTION are JohnnyPoker, Raise and 200.  I am also going to store whether or not he has gone all-in, just because I have a hunch it might be worth noting and it is easily derived.

Finally (for now) I have a POKER TABLE class.  This represents a succession of completed hands, stored in an array.  It can also tell me the number of completed hands in that table and spit out the most recent hand if I ask it nicely.  I’m not sure about the name of this class but will stick with it for now.  I’m also unsure whether my collection of PLAYERs should belong to the POKER HAND or the POKER TABLE.  The latter seems more intuitive (players sit at a poker table and play hands) but in some variants of Hold ‘Em (Tournaments for example) the players at the table frequently change.

I am under no illusions and fully expect to lose everything I’ve made very quickly.  I think a range of ten competitions is the absolute minimum I need in order to establish whether or not I am improving.  If I can make my previous ten games at any given time better than the preceding ten, I am heading in the right direction.

What’s most important as I play is recognising the changes I am making while reviewing the results.  Am I playing tighter or looser?  More or less aggressive?  And what do these terms translate to in a given scenario of hole cards and board cards?  Perhaps I’m just paying more attention to what my competitors are doing and have a better read on the likely outcome of a hand.  As I find approaches that seem to pay off, I jot them down and will try to implement them in my application.  I’ll share the techniques I try out and why I think they make sense, as well as other incidental lessons I pick up when normal service is resumed and I am losing heavily.

So after many cigarettes by the window (my time-out spot) I have decided to first build what I will call (for now) a Pre-Flop Hand History Advisor.  Many times when an opponent calls or raises before the flop, I find myself trying to remember if this is out of character for that player, and if so to what extent.  Have they folded almost every other hand, meaning they now have something really really decent, or are they calling the blinds almost every time and hoping to get lucky on the flop?  Here’s where I’m heading (click to enlarge):

An opponent’s pre-flop play is easiest to analyse because there are fewest variables (and therefore the easiest maths).  When I look at a player’s call or raise at this stage of the game, I only need to consider seat position, the size of the bet, the number players and the pot size.  I then want to compare this to previous plays and look for patterns.

It’s also simplest to start with something that merely presents data.  There are three distinct functional areas my efforts are going to focus on.  First, I need to read in the poker game from a variety of sources, including hand history files (which FullTilt helpfully stores locally) and the current hand in play.  Then I need to take this data and decide what action to take – call/raise/bet etc.  This stage will be the trickiest.  Finally, I have to hook that action up to the poker client – click the correct button, or enter the bet amount.  My pre-flop advisor only needs to accomplish the first of these, and furthermore from a single source – those hand history files on my local drive.  Easy.  Sort of.

I’ll be tweeting live results and interesting hands, so it’s probably only one for the enthusiasts.  A useful record of my form though as my skill supposedly improves.

An update on progress so far, now that I am a professional poker player and all.  I have played seven or eight $2 games over the last couple of days, with little financial success (a couple of third places, each worth $3.40) but reasonable positions, normally finishing in the top four or five (of nine).  As predicted, even at such low stakes the quality of opponent is noticeably superior to play money games, and I’ve been caught out a number of times making amateur hour moves.  However, with each loss I am noting the most glaring mistake I made, and will start posting them for your collective Schadenfreude.

The first thing I’ve picked up isn’t really a tip or tactic for certain hands, but a more general approach that seems to be getting me decent gains.  By the way, I’m hoping that it goes without saying that you shouldn’t take any advice I offer too seriously, as I am not a professional (or even competent) player.  That’s kind of the point of this whole project.

Ok, so here it is:  I don’t think it’s that hard to consistently make it into a paying position (top three).  In most of my games so far, by the time you reach the final three the person with fewest chips (short stack) will have something around the 1,000 – 2,000 mark.  Considering you start with 1,500 that looks pretty meagre.  Getting there, however, isn’t as simple as sitting tight and folding every hand.  The fourth placed player is generally knocked out after around 60 or 70 hands, by which time the relentlessly increasing blinds will have rendered you paraplegic.  So you certainly need to play, but just not much.  In the few games where I’ve made third, all that was required was one or two medium-sized hands for 300 chips or so.  In fact, a single 500+ win should do it.   The games where I’ve played more hands, I’ve lost earlier.

The obvious disadvantage is that by the time you get to third place, your remaining opponents will have significantly more chips than you.  But by this time, you have already secured a return on your investment, and can try to turn things around.  I have watched the completion of many of the tables I have been eliminated from, and I’m not sure that the benefit of a chip lead in the final three is as significant as you might think.  Think of it as a baseball season – you’re trying to make it through to the playoffs, at which point an entirely new competition starts up to determine the winner.  Certainly your regular season record (chip count) has a bearing on your chances of success, but most of the time it’s anyone’s game, with regular season results having little relevance.

I’m going to stick with this strategy for a couple of days and see what it returns.