This week I have have something little different to talk about. This is not so much a man Man vs. Machine scenario like that of the famous Kasparov vs. Deep Blue series, but more of a melding of Man & Machine. I’ve been reading about different ways that ‘expert’ knowledge & regression algorithms (formulas to predict outcomes, basically Punter is just a self learning regression algorithm). I’ve been reading about ways that expert knowledge can be added to results from regression algorithms to compliment the final result.
An interesting fact out of this was that more often that not, humans are over confident in their general predictions, and especially when experts were confronted with the fact that the algorithms might hold balance in the situation of a tie breaker decision. One approach taken to limit this human over confidence was to limit the number of times that a human could override the decision of the algorithm. 10% was a number stated in on place, but a sensible controlling factor was also put into place. If the incorrect number of overridden decisions exceeded 50%, then the number of overriding decisions should be reduced. Seems logical really, if the computer wins more often, then let it choose more often.
So, here is how this is going to work. Each week I’ll put up my tips, and then use the following rule set to possibly make some overriding decisions, and then look and see if the combined decisions improve the overall performance. This is not a replacement for the improvement of Punter, but instead might actually help identify trends to target the learning of Punter. For the 07 home & away season I’m averaging about 5 per week, so if I can even improve on 1 per week, that will be a big boost.
Rules:
- A maximum of 2 overrides can be made each week. That is 25% of the tips.
- At the end of the season, the performance will be reviewed, and we might adjust how many overrides can be made.
Now, I’d hardly call myself an expert tipper (or I wouldn’t build Punter, would I). The challenge that I have is if anyone reckons that can go one up on me using the rules above, then feel free to post you’re alternatives as comments to the tips page each week. Ok, now the tips.
My overrides for this week:
- Magpies will beat the Kangaroos
|
|
Probability
|
Punter
|
Bookie
|
|
Gold
|
Silver
|
|
Kangaroos:
|
53.785%
|
$1.86
|
$2.75
|
|
Magpies:
|
46.215%
|
$2.16
|
$1.46
|
|
|
|
|
Cats:
|
57.334%
|
$1.74
|
$1.58
|
|
Western Bulldogs:
|
42.666%
|
$2.34
|
$2.40
|
|
|
|
|
Tigers:
|
50.180%
|
$1.99
|
$1.60
|
|
Bombers:
|
49.820%
|
$2.01
|
$2.35
|
|
|
|
|
Lions:
|
58.228%
|
$1.72
|
$1.10
|
|
Eagles:
|
41.772%
|
$2.39
|
$7.50
|
|
|
|
|
Saints:
|
49.944%
|
$2.00
|
$2.80
|
|
Hawks:
|
50.056%
|
$2.00
|
$1.45
|
|
|
|
|
Blues:
|
28.260%
|
$3.54
|
$2.35
|
|
Swans:
|
71.740%
|
$1.39
|
$1.62
|
|
|
|
|
Power:
|
14.068%
|
$7.11
|
$2.05
|
|
Crows:
|
85.932%
|
$1.16
|
$1.78
|
|
|
|
|
Dockers:
|
67.032%
|
$1.49
|
$1.44
|
|
Demons:
|
32.968%
|
$3.03
|
$2.85
|
|
|
|
|
Premiership: 76 correct tips from 120 games. 63.33% accuracy.
|
|
Pre-Season: 10 correct tips from 15 games. 66.67% accuracy.
|
|
Total: 86 correct tips from 135 games. 63.70% accuracy.
|
|
|
39/51 76%
|
|
4/7 57%
|
|
43/58 74%
|
|
|
20/40 50%
|
|
3/4 75%
|
|
23/44 52%
|
|
