AD's daily selections
- Thread starter AustinDillon75
- Start date
AustinDillon75
Colt
I've been testing it to BSP throughout.
AustinDillon75
Colt
So yesterday lost 8pts or so, it feels as though continuing the model as it is won't serve us as it's going disappointingly stale. When the strike rate is pretty stable but the average BSP of the selections themselves is 1pt less than when we started you already know that the market has factored things in.
I have now updated the model in a couple of ways.
1. It still goes back over 4 years of results but what I noticed was it simply regarded results from year 1 as equal to those from year 4 and there will now be a decay which places greater value on recent races. It still looks at recent strike rates and current form but this should stop getting drowned by outcomes that could be redundant in the present.
2. Ive introduced a seasonal set which goes back 10 years for the date 28 days either side of the current day so it has a. It more information to work with on any seasonal trends. It could help a bit for distinguishing what wins summer jump races, for example.
I re ran this model on 2 May as a day where we had 9 winners from 50 and it selected quite a few non runners but got 3 more winners from 5 less selections and turned an 8pt loss into an 8pt gain. The model takes a little longer to run (about 5 minutes) and it will still try and identify value bets too. The Spearmans and Pearsons were lower so it hopefully isn't just hooking itself to the market like glue now. I'd looked at Pearson 0.80 as ideal but it feels like its just copying the market whereas 0.6 is too inaccurate so around 0.7 may prove the point of accuracy vs disagreeing where we need to.
I am going to test this incrementally, post today's races but start re-testing it from previous days too and see if it improves on the current model. It will also be important to see if it is looking at different features. I will also keep a record of the RMSE as each day runs too.
I don't mind recording every race but we will just enjoy any success and try to tackle the failures head on. The main challenges are seeming to be bigger fields and the jumps handicaps and bigger fields > lower strike rates > bigger prices.
Here's hoping.
I have now updated the model in a couple of ways.
1. It still goes back over 4 years of results but what I noticed was it simply regarded results from year 1 as equal to those from year 4 and there will now be a decay which places greater value on recent races. It still looks at recent strike rates and current form but this should stop getting drowned by outcomes that could be redundant in the present.
2. Ive introduced a seasonal set which goes back 10 years for the date 28 days either side of the current day so it has a. It more information to work with on any seasonal trends. It could help a bit for distinguishing what wins summer jump races, for example.
I re ran this model on 2 May as a day where we had 9 winners from 50 and it selected quite a few non runners but got 3 more winners from 5 less selections and turned an 8pt loss into an 8pt gain. The model takes a little longer to run (about 5 minutes) and it will still try and identify value bets too. The Spearmans and Pearsons were lower so it hopefully isn't just hooking itself to the market like glue now. I'd looked at Pearson 0.80 as ideal but it feels like its just copying the market whereas 0.6 is too inaccurate so around 0.7 may prove the point of accuracy vs disagreeing where we need to.
I am going to test this incrementally, post today's races but start re-testing it from previous days too and see if it improves on the current model. It will also be important to see if it is looking at different features. I will also keep a record of the RMSE as each day runs too.
I don't mind recording every race but we will just enjoy any success and try to tackle the failures head on. The main challenges are seeming to be bigger fields and the jumps handicaps and bigger fields > lower strike rates > bigger prices.
Here's hoping.
AustinDillon75
Colt
Today's selections, the spreadsheet may need some tidying up etc but I am going to backdate this now and test it from 27 April onwards, so I'll put today's results up and run the bets from those dates until such time that we catch up with the present.