I'm a 20 year old aussie currently studying Software Engineering & Finance.

Poker History
I started playing poker when I was 17, trying to emulate the success of Athene (Chiren80). I initially approached poker as a game to make money and following Athene, tried to play as many tables as possible to maximize my hourly. After 18 months of playing (at the end of 2012) I was doing well at 100nl and had taken several failed shots at 200nl, which i found much harder. Around this time I had started to play more and more often, and more and more tables, and I eventually burned out and lost interest. I basically gave up poker for 18 months, putting my roll towards moving out and other projects (e-commerce websites and software).

Sports Modelling and Betting
Around a year ago I began to read up on Betfair trading and creating models for betting, which turned out to be of great interest to me. About 6 months ago I started collecting data and working on a model that would be used for automated trading on soccer, I am now in the final stages of testing and will have it working for the next EPL. In doing this I further developed my programming skills and dipped into a lot of stats and machine learning. My sudden interest in stats and modelling led me into picking up Finance this year along with my Software Engineering degree.

When I heard out about the Dota 2 International (a $10 million tournament) I decided to create a model for betting on the games. This is something that I have only had operational for the last week, but I've had some success with the picks. I'll be posting some picks from the model in this thread.

Return to Poker
Looking back it was an awful decision to mass multi-table like Athene. I would be playing 12-18 tables of 6max regularly and I felt that this stunted my growth as a player, and made me lose interest. I decided to force myself to play 4-6 tables and think through every decision. I started out at 5nl a month ago and have moved up to 10nl, here's my graph so far:



I plan on slowly rebuilding my roll whilst studying full time and working on my other projects.

Here's the output of my dota model for tonight's first match: Newbee vs Titan.



The only value pick is Newbee winning the BO3, currently at 1.377 on pinnacle.

As an avid dota watcher i'm definitely curious to follow this! Will make some bets alongside you if you keep posting in this thread. What is your model based off?

Wow interesting, I'm also trying to learn sports modeling to force myself to learn programming. Could you give some tips for newbies on sports modeling? It would be amazing if you could do a little step by step blog on your next soccer model

EDIT: If you want I can give you coaching on 6max (100nl player) and you coach me on modeling. My skype sn is same as 2p2.

Here's the output and prices from yesterday's games. Since they followed on from each other (the winner of the first plays in the 2nd) the lines don't go up until after the 1st game. It makes it hard for me to bet on the games when they start at 2am/5am/8am :P

I set alarms to wake me up and managed to get some bets on though. Here's the full output and results for yesterday's games.



I've been using the kelly criterion for the bet sizing, you can read about it here: http://www.bettingexpert.com/blog/th...h-should-i-bet
The "Edge" column represents the full kelly bet percentage.

Because of the compounding, I've started the bankroll off at $100 to make the return over the thread's life easy to track.

I've been using half kelly for the bets, for the Nb vs NaVi game I actually used 1/3 kelly (just because the edge seemed too high).

Because of the poor timing (lines going up and games taking place when I'm asleep) I'm going to work my model program into getting updated stats and betting automatically on pinnacle before each game.

I might post a little tutorial on how I can do this automatically.

The next dota games are 3 days away, time for programming and grinding.

Hey Roy, I'm a big fan of your blog! I've been following it for the last few years

I'll keep posting picks in the thread at least until the end of the international. It makes it hard that they have games that depend on the winner of the first game, but I'll post at least the first one of the day.
I could post the output for the possible games (depending on who wins the lead up games), but it wouldn't have taken into account the games that lead up to it. I'll figure something out for this.

Initially I rated teams, using something similar to elo. This looked like quite a good fit, especially for the more established teams, but I realised that players tend to jump teams a lot.

The solution was to model players using markov chains. The model essentially ranks players using their win/loss history. It takes into account their team, i.e. if they're a bad player playing on a good team then they won't get a big reward for a win, and their opponents team, i.e. if they defeat a good team then they'll get a bigger reward. It also has weights for crucial games, i.e. the last game in a BO3 or the finals of an international. It uses some stats magic to convert the rankings of players into rankings of teams and into probabilities of winning.

One kinda big assumption of the model is that it assumes that the players for each team in the game you're predicting are the same for each team's last game. This is fine for The International, but something I'll have to fix.

What exactly is dota? Interesting model, I will be following.

I miss when Chiren used to grind.. although i wish he would have tried to master heads up or something more entertaining than mass tabling supernova elite at 100nl lol

GL with the dota betting Ive never watched or played it but I know its big now

The best way to learn programming in my opinion isn't to just do a bunch of random tutorials on the internet. It's to have a really big goal, like "create an automated soccer trading program" and use the knowledge of the internet (just google literally every problem you run into). When I creating my soccer program I probably spent 70% of my time googling how to do stuff that I had never done before.

Programming for R (a free stats program) is pretty simple. You could just find some data and see what you can do with it (google is your friend). A really good way to learn programming is to create a program to collect data for you from Betfair/Pinnacle's API (there's also some tutorials that you can follow).

The hard part is figuring out what to apply to your data to get a fit for a model. For instance, a poisson model can be used to fit the scores of soccer matches. Find articles on predicting what you're trying to predict to figure out what to apply to it. Then you can use R to fit your data to a model automatically often using packages/tutorials that other people have made.

I have far more programming skill than modelling or stats skill. My soccer model only covers around 30% of the markets that I had originally want to model, probably something I will improve on later.

I'll be making a post about how I get my dota program to place bets automatically with pinnacle. I might post some stuff later about the soccer program.

Dota is a 5v5 team computer game. You can watch the games from the tournament yesterday here: http://www.dota2.com/international/watch/July14/

Me too

Cheers man

Couldnt resist and clicked to see what's dota, and saw 10M prize pool for some animated guys going around and kick each other, hah, are there often these big tourneys??

gl with your goals op

Haha, yeah. There's several tournaments a year, this is by far the biggest though.

The biggest reason (other than for the experience) for creating the model was how little competition there is. Compared to soccer/tennis where you have sports hedge funds and insane amounts of money being bet on them.

I like to be a big fish in a small pond.

Very interesting, i'll definitely be following this thread!

Ranking players rather than teams seems like a good way to go. It sounds like there's potential to add lots more data and refine the model, and a lot of that data is out there and available on dotabuff etc.

Good stuff man.

I've played a bit of poker and done some more programming. I'll have a poker update later in the week.

I added a button to automatically get the pinnacle odds and put them in the table with the model odds, along with the edge.

Here's the output for the first two of tomorrow's games:



There's also an Under/Over line for the number of games in the series (Under or Over 2.5 games in the best of 3). I didn't put them up because there was no value on either of them.

Unfortunately there is one game following these between the winner of the above two games. I could give the output for that one but it would depend on not only the winners of the games, but also if the games were a 2-0 or a 2-1. I could give the output for all possible combinations but I think it's too much. Instead I will set an alarm to wake up and bet on the game, I'll post the output and any bets then too.

Here's my bets on these games for tomorrow:



I was planning to use Pinnacle's API to place bets automatically, but I won't end up doing this as I can't get it to reliably tell me crucial things like is this match a best of 3? And is this even a Dota2 match?
For now I'll leave the automatic betting to Betfair.

Awesome stuff. I had $68.74 in my pinny account that i'd forgotten about, so I went all in on EG +1.5. If they win i'll parlay on newbee, gl us!

lets hope western teams stay in the tourney all the way, it will be pretty boring (at least for me) to watch chinease teams battle it out, like, every day about a month ago there were newbee-dk-ig-vg bo3 and bo5 games on minor tourneys

EE-sama have no mercy pls. eg c9 in the final would be kewl

pls tell more about the trading on betfair. whats your favourite market? whats your strategy?
was doing this back in time too..
unfortunatly im from germany

(and wow @ this dota thing.. never heard about it.. sometimes i wish i would be young enough to stay up to date :-) )

Yolo. Lol... I'm watching it now, I think the Newbee game is first? GL!

I hope the western teams stay in just because I think everyone overestimates the ability of western teams to win. This makes it good for betting :P

One of my "hypotheses" going into using the model's predictions for betting was that people will overestimate the ability of western teams because they follow the western teams through streams/forums/podcasts/etc. (as I observed in SC2 where the koreans just dominate). I think that this bias dies off the longer the tourney goes on as I think that people expect a Chinese teams to win overall.

I would like to see the return on a betting system where you just bet on the Chinese team in an east vs west match :P

Yeah the newbee game was/is first, i had it wrong but I deposited another 100 to put on newbee lol - thank god! They killed it in game 1. KILLLLLLLLLLED IT.

The only success I've had is in trading soccer. I developed a model that I will use with a program to automatically trade the matches before they start. The program will initially focus on trading the Over/Under X.5 Goals markets. I have pretty much finished the program, I just need to rigorously test it before the next English Premier League.

Because of some weird laws Australians aren't allowed to bet/trade on in-play markets (except Horse Racing). Otherwise I would create a model for betting on in-play basketball (I believe I could get a big edge).

I would like to create a model and program for automatically betting/trading on horse races. I think this is probably pretty hard to do because I think horse racing is pretty corrupt relative to Soccer/Basketball/Dota. But, because there is so much money on horse racing, and there are so many races every day - all year long, even a small edge can add up to big profits.

Woooo newbee

Haha, it paid off. Let's hope EG picks up the first game

DK is actually the highest rated team in my model. I was surprised when they said that VG was the top seed. I can show the ratings for the teams but I don't want people to reverse engineer my model lol.

After this series I'll put up the output/bets before the last series.

Here's the output for the next series. It's not up on pinnacle yet.



If you're going to bet on the handicap (+/- 1.5) make sure the + and - match up with pinnacle (the manual picks like this one doesn't match up with pinnacle if the favorite differs between the two)

I'll edit this post with the pinnacle odds when they go up.

EDIT: Here's the output with pinnacle odds:

There's no handicap up for the series.

Unfortunately no value to be found

EG up to 1.719 now, with the way they just played and how strong PPD drafted last game I think i'm gonna throw some money down on that.