Hi,
Got a bot that I did some back testing on and all seemed good so I put it live. The only things different between live and practice is the stake size, a number of runners condition and an unmatched bet condition to make sure it only fires 1 bet.
8 or 9 times out of 10 the live bot and practice bots pick the same horses around the same time with roughly the same odds. But sometimes it picks different horses and/or fires at different times around 30 seconds apart. This has unfortunately led to 2 losses today on the live bot that were actual wins on the practice bot. There have been a couple of races the live bot has traded that the practice bot hasn't and vice versa.
Does anyone have any ideas please? I was under the impression that on small stakes the practice mode is virtually identical as the live environment and I wouldn't have thought that my additional rules on the live bot would make a difference to the otherwise normal bot.
Thanks everyone.
Bot chooses different horses in practice mode vs live
live will always be different in some way, depending a bit on what your bot is doing... Your practice mode is not actually participating in the market, so your fill rate wont be representative of live.mtrehearn wrote: ↑Mon Jul 04, 2022 8:40 pmHi,
Got a bot that I did some back testing on and all seemed good so I put it live. The only things different between live and practice is the stake size, a number of runners condition and an unmatched bet condition to make sure it only fires 1 bet.
8 or 9 times out of 10 the live bot and practice bots pick the same horses around the same time with roughly the same odds. But sometimes it picks different horses and/or fires at different times around 30 seconds apart. This has unfortunately led to 2 losses today on the live bot that were actual wins on the practice bot. There have been a couple of races the live bot has traded that the practice bot hasn't and vice versa.
Does anyone have any ideas please? I was under the impression that on small stakes the practice mode is virtually identical as the live environment and I wouldn't have thought that my additional rules on the live bot would make a difference to the otherwise normal bot.
Thanks everyone.
There are a number of aspects that will differ, eg.adverse selection is typical, ie.your losers match more than your winners compared to practice mode.
Thank you. The bot looks to lay at least the 4th horse subject to a bunch of conditions. I can handle the time and / or filling of bets being different as long as the same horse is picked in live vs practice as that's obviously what I've modelled my bot in. Is there any way around or at least to somewhat predict the "adverse selection" or at least get the bots matching horse selection?goat68 wrote: ↑Mon Jul 04, 2022 8:52 pmlive will always be different in some way, depending a bit on what your bot is doing... Your practice mode is not actually participating in the market, so your fill rate wont be representative of live.mtrehearn wrote: ↑Mon Jul 04, 2022 8:40 pmHi,
Got a bot that I did some back testing on and all seemed good so I put it live. The only things different between live and practice is the stake size, a number of runners condition and an unmatched bet condition to make sure it only fires 1 bet.
8 or 9 times out of 10 the live bot and practice bots pick the same horses around the same time with roughly the same odds. But sometimes it picks different horses and/or fires at different times around 30 seconds apart. This has unfortunately led to 2 losses today on the live bot that were actual wins on the practice bot. There have been a couple of races the live bot has traded that the practice bot hasn't and vice versa.
Does anyone have any ideas please? I was under the impression that on small stakes the practice mode is virtually identical as the live environment and I wouldn't have thought that my additional rules on the live bot would make a difference to the otherwise normal bot.
Thanks everyone.
There are a number of aspects that will differ, eg.adverse selection is typical, ie.your losers match more than your winners compared to practice mode.
Thanks
Run it live for a week or so, and see if the end result/performance really is different... don't aim to get it identical, you won't, just measure the live resultsmtrehearn wrote: ↑Mon Jul 04, 2022 8:59 pmThank you. The bot looks to lay at least the 4th horse subject to a bunch of conditions. I can handle the time and / or filling of bets being different as long as the same horse is picked in live vs practice as that's obviously what I've modelled my bot in. Is there any way around or at least to somewhat predict the "adverse selection" or at least get the bots matching horse selection?goat68 wrote: ↑Mon Jul 04, 2022 8:52 pmlive will always be different in some way, depending a bit on what your bot is doing... Your practice mode is not actually participating in the market, so your fill rate wont be representative of live.mtrehearn wrote: ↑Mon Jul 04, 2022 8:40 pmHi,
Got a bot that I did some back testing on and all seemed good so I put it live. The only things different between live and practice is the stake size, a number of runners condition and an unmatched bet condition to make sure it only fires 1 bet.
8 or 9 times out of 10 the live bot and practice bots pick the same horses around the same time with roughly the same odds. But sometimes it picks different horses and/or fires at different times around 30 seconds apart. This has unfortunately led to 2 losses today on the live bot that were actual wins on the practice bot. There have been a couple of races the live bot has traded that the practice bot hasn't and vice versa.
Does anyone have any ideas please? I was under the impression that on small stakes the practice mode is virtually identical as the live environment and I wouldn't have thought that my additional rules on the live bot would make a difference to the otherwise normal bot.
Thanks everyone.
There are a number of aspects that will differ, eg.adverse selection is typical, ie.your losers match more than your winners compared to practice mode.
Thanks
I've just copied the rule under a different name and changed the stake size, added the unmatched bet on market rule and removed the offsetting that was on the practice bot. That's all. Don't get the difference in behaviour and sadly it cost me today. Not much but still.
Yeah that's the plan at least. I just keep thinking there has to be a way around it. I hate things I can't explain and this is one of those things right now.goat68 wrote: ↑Mon Jul 04, 2022 9:13 pmRun it live for a week or so, and see if the end result/performance really is different... don't aim to get it identical, you won't, just measure the live resultsmtrehearn wrote: ↑Mon Jul 04, 2022 8:59 pmThank you. The bot looks to lay at least the 4th horse subject to a bunch of conditions. I can handle the time and / or filling of bets being different as long as the same horse is picked in live vs practice as that's obviously what I've modelled my bot in. Is there any way around or at least to somewhat predict the "adverse selection" or at least get the bots matching horse selection?goat68 wrote: ↑Mon Jul 04, 2022 8:52 pm
live will always be different in some way, depending a bit on what your bot is doing... Your practice mode is not actually participating in the market, so your fill rate wont be representative of live.
There are a number of aspects that will differ, eg.adverse selection is typical, ie.your losers match more than your winners compared to practice mode.
Thanks
Could it be related to stake size? Betfair's minimum is now £1 so if I am doing say 10p/20p/50p bets instead of the £2 bets the practice bot was at will that make a difference to selection do you think?goat68 wrote: ↑Mon Jul 04, 2022 9:13 pmRun it live for a week or so, and see if the end result/performance really is different... don't aim to get it identical, you won't, just measure the live resultsmtrehearn wrote: ↑Mon Jul 04, 2022 8:59 pmThank you. The bot looks to lay at least the 4th horse subject to a bunch of conditions. I can handle the time and / or filling of bets being different as long as the same horse is picked in live vs practice as that's obviously what I've modelled my bot in. Is there any way around or at least to somewhat predict the "adverse selection" or at least get the bots matching horse selection?goat68 wrote: ↑Mon Jul 04, 2022 8:52 pm
live will always be different in some way, depending a bit on what your bot is doing... Your practice mode is not actually participating in the market, so your fill rate wont be representative of live.
There are a number of aspects that will differ, eg.adverse selection is typical, ie.your losers match more than your winners compared to practice mode.
Thanks
Yes. Depends what your rules are doing, but stake size difference can make a huge difference.mtrehearn wrote: ↑Mon Jul 04, 2022 9:24 pmCould it be related to stake size? Betfair's minimum is now £1 so if I am doing say 10p/20p/50p bets instead of the £2 bets the practice bot was at will that make a difference to selection do you think?goat68 wrote: ↑Mon Jul 04, 2022 9:13 pmRun it live for a week or so, and see if the end result/performance really is different... don't aim to get it identical, you won't, just measure the live resultsmtrehearn wrote: ↑Mon Jul 04, 2022 8:59 pm
Thank you. The bot looks to lay at least the 4th horse subject to a bunch of conditions. I can handle the time and / or filling of bets being different as long as the same horse is picked in live vs practice as that's obviously what I've modelled my bot in. Is there any way around or at least to somewhat predict the "adverse selection" or at least get the bots matching horse selection?
Thanks
Would stake size affect horse selection like that though for such low stakes, if so what is a good size stake to start with? I ran the practice bot with £2 stakes and £20 stakes and it was choosing the same horses every time. Do you think its more to do with stake size than live vs practice environments? Thank you.goat68 wrote: ↑Mon Jul 04, 2022 9:35 pmYes. Depends what your rules are doing, but stake size difference can make a huge difference.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
Is it in-running? Small bets take double the usual 1s delay.
Yeah it's in play. Well I've got a 5 second fill or kill on it so not sure how or if the delay for smaller bets effect it....? I just want it to work like it did in practice mode lol. If I ran the bot using varying stakes all on practice mode (for arguments sake say 10p / 50p / £1 - do you think I'd get similar results to what I'm getting here with smaller stakes (I. E. my current live mode) having more lay bets matched that actually win vs higher £2 stakes (my practice bot)? Or is it really going to mean just going live with higher stakes? If any of that makes sense lolShaunWhite wrote: ↑Mon Jul 04, 2022 10:05 pmIs it in-running? Small bets take double the usual 1s delay.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
Sub-min bets take 2s instead of 1s to appear in the market, but I don't know if you're placing them beforehand or in response to something. Generally though as has been said, Practice Mode is a test of your functionality and not how a bot will actually perform. Timing, position in queue and your bet's effect on the market will all play a part.mtrehearn wrote: ↑Mon Jul 04, 2022 10:12 pmYeah it's in play. Well I've got a 5 second fill or kill on it so not sure how or if the delay for smaller bets effect it....?ShaunWhite wrote: ↑Mon Jul 04, 2022 10:05 pmIs it in-running? Small bets take double the usual 1s delay.
Its in response to something. So if XYZ happens lay the 4th down. When you mentioned appear in the market what does that actually mean? Is that before it is being placed? So does it wait an extra second before placing compared to regular above min bets? If my bot waits for XYZ to be true for say 20 seconds before placing the bet - will it help to change the 20 seconds?ShaunWhite wrote: ↑Tue Jul 05, 2022 12:36 amSub-min bets take 2s instead of 1s to appear in the market, but I don't know if you're placing them beforehand or in response to something. Generally though as has been said, Practice Mode is a test of your functionality and not how a bot will actually perform. Timing, position in queue and your bet's effect on the market will all play a part.mtrehearn wrote: ↑Mon Jul 04, 2022 10:12 pmYeah it's in play. Well I've got a 5 second fill or kill on it so not sure how or if the delay for smaller bets effect it....?ShaunWhite wrote: ↑Mon Jul 04, 2022 10:05 pmIs it in-running? Small bets take double the usual 1s delay.