Hi there,
Is it possible to run a bot (practice mode) on a race or a set of races that has already run, like a bot training area in BA?
The purpose of this would be to alter just one thing at a time so I can accurately compare results more effectively than if I ran them on different days.
Don't think my computer will run more than one application of BA software.
Is there a way to do this?
-
ShaunWhite
- Posts: 10527
- Joined: Sat Sep 03, 2016 3:42 am
No, and as every race market is different, I'm not sure what you would prove? All you end up with is a bot which would have worked on a race in the past. Like getting last week's lotto numbers.
https://www.betangel.com/forum/viewtopi ... 20&t=11983
Its known as backtesting, as long as you don't overfit against your training set (markets) it can give you a very good idea of what is profitable or not. A strategy that is profitable over the last 7 days has a high chance of being profitable for the next as unlike lotto numbers humans are moving markets and they tend not to be completely random
Its known as backtesting, as long as you don't overfit against your training set (markets) it can give you a very good idea of what is profitable or not. A strategy that is profitable over the last 7 days has a high chance of being profitable for the next as unlike lotto numbers humans are moving markets and they tend not to be completely random
I don't think you can replay data into BA (well, obviously it's possible, but faking BetFair servers and feeding data to BA is a very significant and complex task - not impossible, but almost certainly not practical... I'd guess BA can do this, as part of their testing/development, but that's something different!Zodiac108 wrote:Hi there,
Is it possible to run a bot (practice mode) on a race or a set of races that has already run, like a bot training area in BA?
But, if you're talking about feeding historic data to your bot (presumably written in VBA?), then the short answer is yes.
The longer answer is that it really depends on whether you have historic data and how your bot/system works.
If you have the data (what data you need, depends on what data your bot requires to make choices), then you can 'play' that data through your bot logic, over and over, amending things slightly so that you can see how it reacts to different factors.
But it really depends on your system. The more iterative the bot is in its interaction with the market, the less reliable it will be to replay scenarios over and over. That's because the more you interact with the market, the more you affect it.
So - simplistically (really, really simplistically) - say this was your system:
Track horses with odds < X
If horse odds increase by Y% and are greater than Z, then make a LAY bet.
If that was your entire system, then replaying works perfectly. You can change values X, Y and Z and your logic will make different decisions and you can then assess the outcomes.
But if your system keeps dipping into the market to make more trades, then the odds that you play back in (from real life), won't truly reflect what they would have been if you had actually made trades.
The more you trade (frequency, value) the more you'll affect the market. But to be honest, nothing is perfectly accurate and for most of us, we aren't trading with high enough values to shift the market significantly (I guess...).
Ian
Yup - obviously if this was a feature added to BA then it would make life much easier for a lot of people
My answer was assuming that the OP's bot was VBA and based on 'rules' - so no need to really have BA for that sort of what-if scenario testing. But BAMI would be a huge help for anyone doing interactive trading on BA.
Ian
Once you have a trading framework developed it is actually relatively easy to replace data from betfair to that of a csv or database feed. I have spent the weekend on optimising mine and on pre-play racing its very accurate, in-play is proving much more difficult although streaming certainly helps.EyePeaSea wrote:I don't think you can replay data into BA (well, obviously it's possible, but faking BetFair servers and feeding data to BA is a very significant and complex task - not impossible, but almost certainly not practical... I'd guess BA can do this, as part of their testing/development, but that's something different!
Betangel already has practise mode which uses a simple algorithm to predict if you will get matched, not 100% accurate due to xm, so the hard part has already been developed...
Hi Linus,
Yes, that's exactly what I do. I was just trying to make sure that the OP knew that I meant it could be done outside of BA.
My system has multiple variables (which make about 9k variations) and each variation will have a small effect on when/how much to trade. I run the past 4 months data through all those variations and then it's just a simple job of seeing which one would have had the best/most consistent results. Variations that seemed great for the first month mostly dropped off - but now I've got more data, I'm more confident that the top scoring variation is going to work long term. More confident - not very confident. Ask me in a couple of months
Ian
Yes, that's exactly what I do. I was just trying to make sure that the OP knew that I meant it could be done outside of BA.
My system has multiple variables (which make about 9k variations) and each variation will have a small effect on when/how much to trade. I run the past 4 months data through all those variations and then it's just a simple job of seeing which one would have had the best/most consistent results. Variations that seemed great for the first month mostly dropped off - but now I've got more data, I'm more confident that the top scoring variation is going to work long term. More confident - not very confident. Ask me in a couple of months
Ian
-
ShaunWhite
- Posts: 10527
- Joined: Sat Sep 03, 2016 3:42 am
You're right of course...but the guy asked about A race, singular. Hence the lotto reference.LinusP wrote: A strategy that is profitable over the last 7 days has a high chance of being profitable for the next as unlike lotto numbers humans are moving markets and they tend not to be completely random
My reply wasn't intended to be negative or unhelpful...Obviously if you have the data and the skill to manipulate it, then of course anything you can imagine is possible and I credit the OP with enough brains to know this. Is it possible in BA? ..no. Is using a small sample set helpful?...no.