Testing The Accuracy and Predictions of The Computer

[HomicideHenry]

With the recent 'success' of the computer simulation program that I used to determine winners in the "Modern Day All-Time Heavyweight" tournament, as the computer was proven accurate enough to pick the winner of the Hasim Rahman vs Oleg Maskaev bout last Saturday, I thought, why not 'test' the simulator a few more times?

Currently in what I call "Project Science Faction" myself and a few others are going to look at some up-coming bouts in the near future and have the same computer simulator that predicted the winner of Maskaev/Rahman try and determine the winner.

I have explained before, as well as Ezzard, how the simulator works. Much like Murray Woroner's NCR-315 computer this simulator uses probability and variables, as well as the fighters records and their opponents records, to come up with a 'winner'.

Each fight will be run three times, whoever wins two out of the three simulations that the computer does, will be determined as the 'winner' and the results of the simulators will be evaluated and cross-examined by the actual results.

I plan to have at least 10 fights be run through the computer simulator, and if the computer is at least 75% accurate, meaning that if the ratio of results by the computer is equal or near the actual outcome, then it will prove that this simulator is accurate enough to make a decent, if not exact, prediction in "dream fights".

The goal in mind is that if the computer at least gets 7 fights right in it's predictions to the actual outcomes. So far the computer was right in picking Maskaev to win over Rahman. 1 down, 9 more to go [6 more to achieve the goal].

The next fight(s) that the computer will "predict" is under examination, names and dates being tossed around, nothing is for certain as of yet. If anybody has any fights, no matter the weight class, to throw out for suggestion, please do.

[Controversial]

With the recent 'success' of the computer simulation program that I used to determine winners in the "Modern Day All-Time Heavyweight" tournament, as the computer was proven accurate enough to pick the winner of the Hasim Rahman vs Oleg Maskaev bout last Saturday, I thought, why not 'test' the simulator a few more times?

Currently in what I call "Project Science Faction" myself and a few others are going to look at some up-coming bouts in the near future and have the same computer simulator that predicted the winner of Maskaev/Rahman try and determine the winner.

I have explained before, as well as Ezzard, how the simulator works. Much like Murray Woroner's NCR-315 computer this simulator uses probability and variables, as well as the fighters records and their opponents records, to come up with a 'winner'.

Each fight will be run three times, whoever wins two out of the three simulations that the computer does, will be determined as the 'winner' and the results of the simulators will be evaluated and cross-examined by the actual results.

I plan to have at least 10 fights be run through the computer simulator, and if the computer is at least 75% accurate, meaning that if the ratio of results by the computer is equal or near the actual outcome, then it will prove that this simulator is accurate enough to make a decent, if not exact, prediction in "dream fights".

The goal in mind is that if the computer at least gets 7 fights right in it's predictions to the actual outcomes. So far the computer was right in picking Maskaev to win over Rahman. 1 down, 9 more to go [6 more to achieve the goal].

The next fight(s) that the computer will "predict" is under examination, names and dates being tossed around, nothing is for certain as of yet. If anybody has any fights, no matter the weight class, to throw out for suggestion, please do.

OK....how about James Toney vs Samuel Peter

[Collins2000]

Run it for any / all of the upcoming heavyweight 'title' fights, Rufus.

[BoxBuzz]

I'll be looking forward to comparing it to the BoxRec rating system

[HomicideHenry]

The only draw back, I think with the Toney/Peter fight is that James Toney has fought so many great names in various weight classes, while Peter has relatively faced one major fighter in the flawed Wladimir Klitschko. I think the computer would see Toney's record against guys like Roy Jones, Bernard Hopkins, Evander Holyfield, John Ruiz, and automatically pick Toney to win---if that makes sense to anybody.

I will run it anyways, with the help of others with the simulator as well, and use the best two out of three method---and of course compare the computers findings with the actual out come and as well as this site's computer rankings on an ATG scale.

I would like to test the computer in the Klitschko/Briggs fight, but with the WBO champion pressing Klitschko, who knows if the Briggs bout will be cut. I would also like to test it against the Barrett/Valuev match, and the upcoming Holyfield fight.

That's 3 fights in consideration, so far.

[amohoop34]

Would Miranda vs. Abraham be a good example? It's considered to be somewhat of a pick'em fight, and neither fighter has a loaded resume that would influence the computer too much, i don't believe.

[HomicideHenry]

I think another good fight to be tested by the computer would be Hatton/Urkal....iono if it's still going on as planned though, considering Hatton was looking so forward to facing Gatti in a unification match.

[lvlarc]

Right I can help with this, but I'm no expert on this Title Bout, had it on PC for few months but never really used it.

I've just opened it up, sorted HW's by rating, and the list looks like this...



http://img334. .us/img334/513/titleboutbb5.jpg


Do you have the same ratings? Toney isn't rated as a HW in my version, but a retired SMW...

Also in options for external data source link it has "www.boxrec.com/record$$ID$$.html_"

[HomicideHenry]

I'm sorry that I didn't post this up earlier, I went to bed last night rather than drop more 'predictions' so I apologise for that.

I did a 'best two out of three' on Holyfield vs Bates, in which was Holyfield's 'come back' fight since losing to Larry Donald. The computer picked Holyfield 3 out of 3 to win by KO. The first run of it went three rounds, the second went one and the third went two.

The actual result of that fight was a 2nd round KO win for Holyfield.

Myself I don't know if this is 'accurate' considering that the computer is judging, also, on who each man fought---and Bates had relatively fought nobody and Holyfield has faced Tyson, Lewis, Bowe and a slew of other greats, so it could be biased looking at it. Of course, the computer does take into question of age and of course the win/loss ratio, and Holyfield has lost his last few bouts.

So...... :) the computer was right again! First with Maskaev/Rahman and now Holyfield/Bates. Two down and five more to go to make the goal of 75% accuracy.

[Collins2000]

I'm sorry that I didn't post this up earlier, I went to bed last night rather than drop more 'predictions' so I apologise for that.

I did a 'best two out of three' on Holyfield vs Bates, in which was Holyfield's 'come back' fight since losing to Larry Donald. The computer picked Holyfield 3 out of 3 to win by KO. The first run of it went three rounds, the second went one and the third went two.

The actual result of that fight was a 2nd round KO win for Holyfield.

Myself I don't know if this is 'accurate' considering that the computer is judging, also, on who each man fought---and Bates had relatively fought nobody and Holyfield has faced Tyson, Lewis, Bowe and a slew of other greats, so it could be biased looking at it. Of course, the computer does take into question of age and of course the win/loss ratio, and Holyfield has lost his last few bouts.

So...... :) the computer was right again! First with Maskaev/Rahman and now Holyfield/Bates. Two down and five more to go to make the goal of 75% accuracy.

You need to put the computer 'predictions' up BEFORE the fight, mate.

[HomicideHenry]

well, this is based on 10 fights. I suppose if the computer wins at least 7 out of ten then that means it is really accurate---and if you look at it in a mathematical stand point, out of 20 fights [if the computer wins 7 out of 10] then it should win 17 out of 20 (?) but that in and of it's self is based on probability.

The goal is 7 out of 10, and as the old saying goes: "better to learn to walk before you can run."

[Collins2000]

OK, let's see the computer prediction on Samuel Peter vs James Toney which is scheduled to take place in a couple of weeks.

The Holyfield vs Some Bum Called Bates 'fight' was a gimme. We don't need to include any more of those, Rufus. They 'prove' nothing.

[BoxBuzz]

yeah I would think the computer would be driven off course by the record and could not take the "age" and "bell curve" factors into account. So the computer is going to be dead centered on Toney and I think most of us know that Peter at least has a very serious shot at giving "lights out" a new meaning in this upcoming fight.

[HomicideHenry]

The Bates/Holyfield fight I would consider a 'cross-roads' fight, considering Holyfeild took such a beating from Larry Donald and at the state Holyfield is in [slurring in speech and zapped reflexes] it was somewhat a decent test to see if Holyfield was truly done for. But I agree in some ways it proves nothing, other than that.

I will run the Toney-Peter fight, and see who wins the 'best two of three' and post it up when I have the time to.

[Ezzard]

IRM

there is a facility in title bout that allows you to simulate a fight 100 times if you want to, all in a matter of minutes.

There are so many other factors to take into consideration. Are you joining up the trainers and corner men with the fighters or just leaving them on a generic setting?

Do you have the rule set configured to be the same as the actual fight?

Are you using the same referee as the actual fight or a random ref (remember these can be historical refs)?

The judges also have a setting whereby they favour certain styles which itself allows for bad decisons.

Remember that the game itself has documentation pertaining to fighting retired fighters against active fighters. There is a button to press to adjust the ratings for active fighters.

[HomicideHenry]

Ezzard can you run the James Toney vs Samuel Peter fight? You seem to have taken alot more things on the game into consideration than I have.

[Ezzard]

Ezzard can you run the James Toney vs Samuel Peter fight? You seem to have taken alot more things on the game into consideration than I have.

Sorry, I have an old version and Peter doesn't exist. Maybe you could get the good people at boxrec to vote on his ratings!!!!!

Also, Toney is rated as a MW and whilst there is a facility for rating fighters out of their weight classes it is pretty basic. Some guys are rated in more than one division but he isn't one of them. I ghenerally attempt to modify the scores myself but in this instance it wouls simply be a computer interpreting my perception of the 2 fighters...

[Ezzard]

There is a computer programme that was invented to mimic the human mind and if you asked it a question word association would drive it into giving you the "correct" answers and it would actually learn the information given by people who tried it out. For example if you said "Hello, what do you think of Bill Clinton" you would get a diatribe about oral sex in the Oval Office. On the other hand when you said "How are you today?" the computer blew a fuse. One of the more high-tech ones was diagnosed with depression and began saying very bleak things and lapsing into silence, they thought it had develped consciousness but it turns out every night the depressed cleaning lady was switching it on and moaning about how awful her life was/is. She completely ruined the programme and the programme became an extension of her hopes/fears/inclinations!

Isn't/wouldn't it be the same with this? How can you analyse a fighter by stats, that is why Boxrec has such laughable ratings most of the time. When they get it right it seems to be an accident. Or am I misunderstanding this computer programme?

Think of it like this... There's a game engine that attemtps to model a boxing match. it takes into consideration styles, physical attributes, skills, etc... It also looks at the way in which a fighter utilises strategies, how consistent they are, how well they train, how good they are at carrying out strategies etc... It also allows for freak incidents, broken hands, cuts froma ccidental headbutts, poor judges, premature stoppages, etc...

We can argue on and on about how well the game simulates boxing... Does it make cut stoppages too likely? Does it bring about too many bad decisions? Does it place too much emphasis on cornermens' instructions etc...

Then there are the fighters. Everyone is going to argue about how the fighters are rated. I mean thisd is basically what goes on on this board.

I had the game as a kid and loved it so much that I bought the computer game for nostalgia's sake. When IRM talked about it I fired it up again. It's great for the laptop at airports and train stations etc... But I've never been enamoured by their present day ratings, mainly due to the fact that many fighters' weaknesses aren't apparent until they lose a fight. I don't think it's anything more than a bit of fun. If you could use it as a betting tool it would have been done a long time ago.

BTW Terence, I like yout story. Is it for real? A friend of mine is doing a PhD on artifical intelligence. i might run it past him.

[JC]

The only draw back, I think with the Toney/Peter fight is that James Toney has fought so many great names in various weight classes, while Peter has relatively faced one major fighter in the flawed Wladimir Klitschko. I think the computer would see Toney's record against guys like Roy Jones, Bernard Hopkins, Evander Holyfield, John Ruiz, and automatically pick Toney to win---if that makes sense to anybody.

Not to be pedantic, but doesn't that prove the accuracy is flawed in the first place?

[Ezzard]

Ezzard, right I get it. It sounds good. I'd wonder how the computer can understand styles but one thing I do like is that if you feed it the correct judging criteria it, in theory, eliminates bad decisions. Like all computer programmes it is as good as the information it gets, reading fighters styles may be a problem unless you have broadly generic styles such as counter puncher, slugger etc and the computer can then match the physical attributes (say a small slugger versus a long-armed counterer) and work out a good win. I'd like a strategy-based version of that if I was a trainer!

Those stories are true. Well the "How are you today" one certainly is as I was the one who asked the machine (I think it was very basic though, a "Jibber-jabber" as we called them). Your friend will be at a higher level AI wise, I studied it as part of a 'Philosophy of Mind' unit and enjoyed it. It was not my usual area but was good to study for a year. Run that cleaning lady one by your friend. It sounds apt to me, the AI system basically was a mimic and became this lady. It makes sense as you were restricted to ten-minute bursts and this lady was using this machine for whole nights!

Yes, they are generic styles... It's a good way to wile away an hour when you've nothing better to do.

[Collins2000]

OK, so the computer can't pick a winner in Peter vs Toney.

So that's 1 'right' and 1 'wrong' so far.

[Ezzard]

OK, so the computer can't pick a winner in Peter vs Toney.

So that's 1 'right' and 1 'wrong' so far.

Surely this one's a no decision

[cubedrum]

Ezzard can you run the James Toney vs Samuel Peter fight? You seem to have taken alot more things on the game into consideration than I have.

Sorry, I have an old version and Peter doesn't exist. Maybe you could get the good people at boxrec to vote on his ratings!!!!!

Also, Toney is rated as a MW and whilst there is a facility for rating fighters out of their weight classes it is pretty basic. Some guys are rated in more than one division but he isn't one of them. I ghenerally attempt to modify the scores myself but in this instance it wouls simply be a computer interpreting my perception of the 2 fighters...

Toney is rated at MW, SMW and HW in the latest database from June 2005. Check the ratings tab on a fighter who fought at multiple weight classes and you will find several ratings sets for some fighters.

[Collins2000]

OK, so the computer can't pick a winner in Peter vs Toney.

So that's 1 'right' and 1 'wrong' so far.

Surely this one's a no decision

Irish wanted to test it's ability at predicting. If it can't even make a prediction that's a 'loss'.

Anyway, someone else says it can do Toney vs Peters. So let's see the 'prediction'.

[HomicideHenry]

I'll let Cubedrum do it, since he says he has a 2005 version. Mine is based more around the 2003-2004 ratings I hope it can be done soon, or I'll have to use mine and see what the computer says.