Machine learning in trading: theory, models, practice and algo-trading - page 2264
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Yes, I do not really remember, I remember it changed something, but not so much... But this is a completely different picture
because the exponential growth in the log scale looks like a linear one)
because the exponential growth in the log scale looks like a linear one)
and it should have a better effect on generalization, in theory
Regarding the generation approach itself, a criticism from me)
When you create data and go through models looking for a model that will work on "new data" , do you understand that it's a fitting ? do you realize that it's a fitting?
Since this "new data" is involved in the choice of the model, it is not"new data"... it's not very obvious but it's true !
on the 3rd one i checked, yeah.
Continuing the theme.... dug up the old code of those GMMs of yours.
found 4 good models that are on the plus side of "like new data"
GMM model was generated on 500 points test on 15k points
And here they are, on really new data (third sample)
What is interesting, if we reverse the position opening signals for the last two (sinking) models, then they start to earn very well with the commission taken into account
And I slowly continue to experiment with training neural networks with fitness function ....
I came up with this way of describing the utility function (fitness function) , instead of teaching the network to get maximum profit I tried to teach the network to "maximally beautiful graph of income"
What is "maximally beautiful income graph"? I put it as coefficient of correlation of linearly increasing line and yield graph
here is the balance on the triple data, the commission is taken into account
correlation coefficient 0.9947626 almost unity)) even on the graph it is clear that it is like a ruler))
Blue shows balance on a test sample of 800 points of 5 min euro
And this is how balance looks like on a test sample of 5k points
nice))
And I slowly continue to experiment with training neural networks with fitness function ....
I came up with this way of describing the utility function (fitness function) , instead of teaching the network to maximize profit, I tried to teach the network to "maximize beautiful graph of income"
What is "maximally beautiful income graph"? I put it as coefficient of correlation of a linearly increasing line and the graph of profitability
here is the balance on the triple data, the commission is taken into account
correlation coefficient 0.9947626 almost unity)) even on the graph it is clear that it is like a ruler))
Blue shows balance on a test sample of 800 points of 5 min euro
And this is how balance looks like on a test sample of 5k points
cool )))
did you expect something else? )
Well, it should have a better effect on generalization, in theory
it doesn't owe anyone anything.
You were expecting something else? )
I, on the contrary, am satisfied, teaching to max. profits is, for example, a lottery on the test, and here at least a small island of stability
800 points on a 5 minute is not too littleBut the reverse is more difficult. With MO on logic. Almost impossible, you can only approximate
IO can only be broken by another stronger IOReverse grid is an interesting topic.
Put noise on the inputs. Get the spectrum at the output. Build a filter on it.
Then it will turn out that it is possible to get the same result with a combination of bags.
Then we will come to random convolution packet (I forget its name).
And then fantasy will come to an end...
Fade. Darkness. The Factory.
And I slowly continue to experiment with training neural networks with fitness function ....
I came up with this way of describing the utility function (fitness function) , instead of teaching the network to get maximum profit I tried to teach the network to "maximally beautiful graph of income"
What is "maximally beautiful income graph"? I put it as coefficient of correlation of a linearly increasing line and the graph of profitability
here is the balance on the triple data, the commission is taken into account
correlation coefficient 0.9947626 almost unity)) even on the graph it is clear that it is like a ruler))
Blue shows balance on a test sample of 800 points of 5 min euro
And this is how balance looks like on a test sample of 5k points
funny ))
Same thing with pair trading, looking for nice spread and it runs away with AOS.
Continuing the theme.... dug up the old code of those GMMs of yours
found 4 good models that are on the plus side of "like new data"
GMM model was generated on 500 points test on 15k points
And here they are, on really new data (third sample)
What is interesting, if we reverse the position opening signals for the last two (draining) models, then they start to earn very well with the taken into account commission
There the pattern ends, no matter how you teach it. For this you need to semplicate other examples, what was talked about yesterday