Today's Greyhounds

[Fugazi]

Seeing we are talking results and modelling. First 6.5 days live results on my latest model, fully automated, using a combination of history stats and pre off volume, price data. Still early days but I'll take 100 points a week happily. No trading, straight backs and lays.
At the moment it's ranking the dogs in each race by win probability, backing the highest and laying the lowest. So far it's performing as the backtesting showed
Sunday, which was hot, was the only poor day. Will see if it keeps up.

I think being successful in both back and lay is a very promising sign.

Suspect we have similar models, only a couple of parameters I can really add to mine from whats available out there.

I used timeform stats for mine. Trainer and owner had to be excluded from mine as it needed a lot of computing power, maybe thats the additional edge it needs. Also maybe one other stat I think Timeform doesn't use time to first bend or something that other sites do

Had a top data scientist work on mine so its certainly the missing stats hindering me rather than the way it was made. May get in contact with him and see if we can find a way to get the last few stats in

[sniffer66]

I created mine myself, but its using proprietary data that I've sourced, but includes weather, temp, rain, trainer , and a few other stats I'm keeping to myself
Tested back to the start of 2024 and results on unseen data were very decent. Be great if it stays this way, as it will scale

[Fugazi]

I created mine myself, but its using proprietary data that I've sourced, but includes weather, temp, rain, trainer , and a few other stats I'm keeping to myself
Tested back to the start of 2024 and results on unseen data were very decent. Be great if it stays this way, as it will scale

I was on a discord that used to put weather on their data. Maybe it was you on that discord.

I knew it would have an impact but thought it would be too much trouble for the edge it would provide. I mean theres only a finite number of stats you can be using ignoring counting how many shits the dog takes and who is controlling the hare

[sniffer66]

I created mine myself, but its using proprietary data that I've sourced, but includes weather, temp, rain, trainer , and a few other stats I'm keeping to myself
Tested back to the start of 2024 and results on unseen data were very decent. Be great if it stays this way, as it will scale

I was on a discord that used to put weather on their data. Maybe it was you on that discord.

I knew it would have an impact but thought it would be too much trouble for the edge it would provide. I mean theres only a finite number of stats you can be using ignoring counting how many shits the dog takes and who is controlling the hare

To be honest, when I analyse feature importance, weather is low on the scale. Certain trainers\tracks are up high, so worth looking at. As is last second data

[Fugazi]

I created mine myself, but its using proprietary data that I've sourced, but includes weather, temp, rain, trainer , and a few other stats I'm keeping to myself
Tested back to the start of 2024 and results on unseen data were very decent. Be great if it stays this way, as it will scale

I was on a discord that used to put weather on their data. Maybe it was you on that discord.

I knew it would have an impact but thought it would be too much trouble for the edge it would provide. I mean theres only a finite number of stats you can be using ignoring counting how many shits the dog takes and who is controlling the hare

To be honest, when I analyse feature importance, weather is low on the scale. Certain trainers\tracks are up high, so worth looking at. As is last second data

Last second?

[sniffer66]

I was on a discord that used to put weather on their data. Maybe it was you on that discord.

I knew it would have an impact but thought it would be too much trouble for the edge it would provide. I mean theres only a finite number of stats you can be using ignoring counting how many shits the dog takes and who is controlling the hare

To be honest, when I analyse feature importance, weather is low on the scale. Certain trainers\tracks are up high, so worth looking at. As is last second data

Last second?

Volume, price, book%

[Fugazi]

To be honest, when I analyse feature importance, weather is low on the scale. Certain trainers\tracks are up high, so worth looking at. As is last second data

Last second?

Volume, price, book%

Ahh. Thought that was it but then doubted myself and wondered if you meant the last time it placed second

[ForFolksSake]

Seeing we are talking results and modelling. First 6.5 days live results on my latest model, fully automated, using a combination of history stats and pre off volume, price data. Still early days but I'll take 100 points a week happily. No trading, straight backs and lays.
At the moment it's ranking the dogs in each race by win probability, backing the highest and laying the lowest. So far it's performing as the backtesting showed
Sunday, which was hot, was the only poor day. Will see if it keeps up.

laying the lowest

What odds are you 'laying' - they are the ones that are potentially going to knock you out

[sniffer66]

Seeing we are talking results and modelling. First 6.5 days live results on my latest model, fully automated, using a combination of history stats and pre off volume, price data. Still early days but I'll take 100 points a week happily. No trading, straight backs and lays.
At the moment it's ranking the dogs in each race by win probability, backing the highest and laying the lowest. So far it's performing as the backtesting showed
Sunday, which was hot, was the only poor day. Will see if it keeps up.

laying the lowest

What odds are you 'laying' - they are the ones that are potentially going to knock you out

Under 12.0 currently. Backtesting showed higher returns over, but under that shows a reasonable return with less liability. Over and I don't enter. Around a 92% strike rate

[Fugazi]

Seeing we are talking results and modelling. First 6.5 days live results on my latest model, fully automated, using a combination of history stats and pre off volume, price data. Still early days but I'll take 100 points a week happily. No trading, straight backs and lays.
At the moment it's ranking the dogs in each race by win probability, backing the highest and laying the lowest. So far it's performing as the backtesting showed
Sunday, which was hot, was the only poor day. Will see if it keeps up.

laying the lowest

What odds are you 'laying' - they are the ones that are potentially going to knock you out

Under 12.0 currently. Backtesting showed higher returns over, but under that shows a reasonable return with less liability. Over and I don't enter. Around a 92% strike rate

You're not taking BSP then?

[sniffer66]

What odds are you 'laying' - they are the ones that are potentially going to knock you out

Under 12.0 currently. Backtesting showed higher returns over, but under that shows a reasonable return with less liability. Over and I don't enter. Around a 92% strike rate

You're not taking BSP then?

I will, at least on the backs, but I'm on test stakes atm so not hitting min stake. 32 points up yesterday

[sniffer66]

Still going strong....

Dogs1.JPG

[Fugazi]

Still going strong....

Dogs1.JPG

Honestly think greyhound modelling is one of the easiest (I said easiest, not easy ). Its possible to include every available statistic. Of course there are the things you can't control like cheating / condition of traps / hare controller / other inside knowledge. And factors such as when a dog is in/off season would be a bit too time consuming.

[sniffer66]

Still going strong....

Dogs1.JPG

Honestly think greyhound modelling is one of the easiest (I said easiest, not easy ). Its possible to include every available statistic. Of course there are the things you can't control like cheating / condition of traps / hare controller / other inside knowledge. And factors such as when a dog is in/off season would be a bit too time consuming.

It's interesting, I grabbed every (and I mean EVERY) in-play stat from the Sofascore API and used as features to ML model in real time for the footy - momentum, indiv. player stats, all the other usual data, and I couldn't get a decent prediction rate for goals\result. I haven't married pre-KO data to that yet but that's next on the list. Hopefully that might tip the balance

[Fugazi]

Still going strong....

Dogs1.JPG

Honestly think greyhound modelling is one of the easiest (I said easiest, not easy ). Its possible to include every available statistic. Of course there are the things you can't control like cheating / condition of traps / hare controller / other inside knowledge. And factors such as when a dog is in/off season would be a bit too time consuming.

It's interesting, I grabbed every (and I mean EVERY) in-play stat from the Sofascore API and used as features to ML model in real time for the footy - momentum, indiv. player stats, all the other usual data, and I couldn't get a decent prediction rate for goals\result. I haven't married pre-KO data to that yet but that's next on the list. Hopefully that might tip the balance

Too many other factors.

Weather (some teams play better in rain, cold, etc) and may change during game
Distance travelled to game, and last few games
Line up
Time between last few games
Motivation
Ground condition
Size of crowd
Time of day
Formation changes

Just to name a few

The computing power alone just to consider play lineup combinations properly would be huge. I think for the top events you need a team of people.

And at lower levels real in depth knowledge and following of teams.

Modelling can work for football for sure. But significant effort. Would be 9/5 to create and keep updated in my opinion.