Round 5

Here we go again.  Absolute shocker of a week last week, I'm now languishing at the bottom of the Squiggle ladder, and I don't even think the tips were that bad!

Round 5

Gambling:

Round 4 had 4 bets, of which 3 paid out.
Balance: $102.93

Round 5 bets:  StKilda, GWS, and Hawthorn - all to beat the line

Round 4 tips, and Round 3 results

 Like a kid with a bad school report, I've procrastinated and not posted the results of last week!

It was a shocking round for me, tipping just 3 of 9 (could easily have been 5 with two games being less than a kick at the end).

                     Tips  Bits  MAE   Correct | Round by Round
ZELOBot               18   0.13  30.7   66.7%  | 6750                        
HunterBot             17   5.29  27.3   63.0%  | 6830                        
ZaphBot               16   2.04  28.4   59.3%  | 6730     

Note that the scores above to not match Squiggle - my own tracking only considers what I tipped at the start of the round, whereas on Squiggle I did a refresh in the middle of the weekend and it flipped one tip based on the results of the round so far - so on Squiggle I got 4/9.

I've left my sons model in there because it's beating me, and it is absolutely killing it on the Bits score!

Still, I am not changing the algorithm other than rolling back a small change on how it calculates percentages.

Here are this weeks tips.                

Three differences this week: WCE/Stk, Ade/Nor, and Geel/Mel After last week I have no idea which is better :-)

Gambling:

Last week there were no clear bets to be made, so I didnt make any. 
Balance still at $101.59

This week I'm making 4 line bets, each for $1 with payout at $1.90
Those are:  North, StKilda, Essendon, and Freo - not that I expect those teams to win, but that they will do better than the line indicates.

Round 3 tips

Here are the Round 3 tips for the two models:

Round 2 went 'ok' with both bots getting 7/9 taking their total to 13/18 for the season so far.  This week they differ on a couple of games, both of which they pick as 1-point games, so should be interesting to watch.

Round 2 tips and ladder projection

 Here are my tips for Round 2, for both bots.  Formatting sucks, I'll figure out something better for next week

ZaphBot  Car= 3 57% | Gee=22 70% | Syd=29 74% | PA =42 83% | StK=12 63% | GCS=17 66% | Ric=30 75% | WB = 5 58% | Fre= 7 60% | 

ZELOBot  Car= 3 53% | Gee= 8 59% | Syd=10 61% | PA =22 71% | StK= 7 56% | GCS=12 62% | Ric= 4 54% | WB = 4 54% | Fre= 5 56% | 

and here is the projected ladder for the end of the season

Round 1 results

 Round 1 is over, and what a ride that was - some incredible games going down to the line, and some amazing upsets.  Over on Squiggle I'm sitting in 3rd place after 1 round - it could have been worse!

Both models scored 6/9, but got there with slightly different paths

ZaphBot   

✔ Ric=35 78% 
✔ WB = 1 56%
✔ Mel=23 70%
✕ Gee=36 79%
✔ Haw= 2 56%
✕ BL =16 66%
✔ PA =22 70%
✕ GWS= 7 60%
✔ WCE=37 80% 

ZELOBot   
✔ Ric=48 90%
✕ Col= 7 56%
✔ Mel=17 67%
✕ Gee=53 90%
✔ Haw= 2 53%
✕BL =38 86%
✔ PA =44 90%
✔ StK= 6 56%
✔ WCE=42 90% 

Here's the standings after Round 1.  You'll see I've added a third bot in here, my 15yo son came up with a nice simple algorithm so I coded it up and included it in my testing.  Normally I wouldnt include it here, but given the result I'll keep it for a week.  It won't last given his 80-90% confidence in games, but being on top for a week is pretty cool!

                     Tips  Bits  MAE   Correct | Round by Round
HunterBot              6   2.83  18.4   66.7%  | 6                           
ZaphBot                6   0.49  19.9   66.7%  | 6                           
ZELOBot                6  -1.14  23.7   66.7%  | 6                      

This season I am doing some minor gambling and will track the results. Usually I will take either a line bet where my prediction is at least a goal away from the line, or a head-to-head where the odds are in my favor. I have started the season with exactly $100, and will bet in increments of $1 for the moment.

Two bets placed.   Richmond to beat the line, and Melbourne to beat the line.  Both won.

Balance:  $101.80
Profit/Loss:  $1.80

ZaphBot the next generation:
I am working on a tweaked version which may regard the results of the game differently while training and evaluating, it looks promising to start with and I'll share the results once I have it working properly (and will retro-fit the first round or two).

Round 1 Tips & predicted Ladder

Big news - ZaphBot is now being ranked on Squiggle

Here are this weeks tips for ZaphBot, and the new bot I was trialling last year, which for now I will call ZELOBot since it is based on ELO but with a twist:

ZaphBot   

Ric=35 78%
WB = 1 56%
Mel=23 70%
Gee=36 79%
Haw= 2 56%
BL =16 66%
PA =22 70%
GWS= 7 60%
WCE=37 80% 

ZELOBot   
Ric=48 90%
Col= 7 56%
Mel=17 67%
Gee=53 90%
Haw= 2 53%
BL =38 86%
PA =44 90%
StK= 6 56%
WCE=42 90% 

Collingwood and StKilda are the two differences this week.

Ladder Prediction

I have also simulated 1,000,000 seasons with ZaphBots ranking of the teams, and here is the predicted ladder at the end of the home and away season.  The model places Richmond as the minor premiers, with Geelong as runner up, and Adelaide as firm favourites for the wooden spoon.

The table below shows the team, and their probability (%) of finishing in a given ladder position, as well as projected number of wins for the season.  It will be interesting to see how things go during the season:

The Third Bot

I'm working on a third bot at the moment, but it will take a bit longer to figure it out.  Once it is ready I'll add it to the list to compare, and will retro-fit the start of the season (it will be based on history up to the end of last season).

2021: Put your money where your mouth is

So, here we go. Will this year be a 'proper' season, or will it be COVID-19 impacted?  Time will tell

This season I will continue with my tried and tested model as the primary model.  This is despite the new model out performing the old one for the 9 weeks that it ran at the end of last season.  My reasoning is that it was a compromised season and I do not have enough information to indicate whether the new model is truly better or not.

So this season I'll run both against each other, but use the original one in competitions/etc.

I'm also going to do some very tiny gambling and track the results.  I'm starting with a bankroll of a whopping $100, and will bet $1 on game outcomes each week, but not necessarily every game.  I'll choose the games based on the odds being offered compared to the odds my model is predicting.  It is possible some weeks will have zero bets due to no good odds being offered.