Machine learning in trading: theory, models, practice and algo-trading - page 1265
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
So I understand that you can try to change something here
What about the error has improved/had improved? In idea - should not, because up to the last point there is a division.
There's no need to take it out on Max.
He, in fact, is pulling this branch all by himself. Whether it's good or bad is another matter. But, thanks to him, this branch still has at least some relevance.
Take away this branch now, and there would be nothing to read on the forum at all, only to sort out other people's embarrassing program codes - where the "+" and "-" are mixed up.
So what reduced the size of the trees and the forest as a whole?
What with the error has improved / improved? In idea - should not, because to the last point of separation goes.
Not done yet, just "took a look", I'm not sitting around all day poking around too :)
the rebels have again revolted )) the question is - why bother, when you have as much brains as a loaf of bread
Yes! The main thing is that the "caravan" is coming.)
But it's going slowly - that's why people are indignant.
It has been said a billion times - it's all about the input data. Simply the price, because of its irreducible non-stationarity, doesn't work.
A billion times I've been asked - systematize your research.
What kind of data can your forests/networks make money on? Are there such clusters in real BP and its first differences?
Once again, I'll give you a hint - Doc has done a lot of research on tick series, well just VERY hard work. With thinning, without thinning. Just investigated artificial random sequences, etc. Putting it all into tables, charts. Ended up disappearing from the forum. I think he found the Grail. Judging by his posts on my PM, it's "almost probably" that way.
It is going very fast, in less than a year NS, the newest most promising approaches like RL have been studied, a couple of articles have been written to outline the approach
now studying the most rapidly developing Bayesian approach in machine learning, i.e. Bayes+RL. And yes, it's all in English.
everything I do is generally the coolest thing in MO right now, it's complicated, so I don't even read any retarded kindergarten stuff that was known 50 years ago. MO has come a long way since then.
Max, I get it, but still...
Here's Doc's post again, after which his results really started to improve:
"Thereare two experiment files in the atacha archive. Both contain values in normal distribution, the histograms are the same and almost symmetrical with respect to zero.
But these files have one very big difference: Markovness.
One file has memory (a non-markovian process), you can try to predict "the next value is greater than or less than zero" based on past values. You can apply neuronics and other machine learning to predict.
The other file has no memory (Markov process), any prediction will fail. Machine learning is powerless, but maybe Alexander can predict something with physics.
Who will learn to identify which file has memory and which does not - well done, and applying the same method to forex will finally prove that the pricing process is indeed Markovian.
It is also worth checking if the normal distribution is a sufficient condition for profitability of the model. Make a cumulative cum() random walk graph, and try to trade on it."
It's just obvious that he first learned how to work on artificial data with "memory" and then stupidly learned how to allocate it to real BPs.
That's all.
I gave examples on artificial data with memory, everything works fine. The non-stationarity of the market, all that, distributions, all that is clear
Moreover, my TS has also been working for quite a long time (with its own nuances), I'm just not trading because it's not interesting now. And just improving, it is interesting to learn new things
You can download the bot from the last article and earn as the Hindu does. Who prevents you from doing that?!!!
Pardon. Maybe I didn't read everything carefully... I'll see.
!!!
Pardon me. Maybe I didn't read the whole thing carefully... Let's see.
I have already attached a bunch of tests and screenshots, I do not know what else you need. I had already put a lot of tests and screenshots in it, and I don't know what else to do.
I am a guest in this thread. I just came by to share an article
Max, I understand, but still...
I'll give you Doc's message again, after which his results really began to improve:
"Thereare two test files in the atacha archive. Both contain values in normal distribution, the histograms are the same and almost symmetrical with respect to zero.
But these files have one very big difference: Markovness.
One file has memory (a non-markovian process), you can try to predict "the next value is greater than or less than zero" based on past values. You can apply neuronics and other machine learning to predict.
The other file has no memory (Markov process), any prediction will fail. Machine learning is powerless, but maybe Alexander can predict something with physics.
Who will learn to identify which file has memory and which does not - well done, and applying the same method to forex will finally prove that the pricing process is indeed Markovian.
It is also worth checking if the normal distribution is a sufficient condition for profitability of the model. Make a cumulative cum() random walk graph, and try to trade on it."
It's just that he obviously first learned how to work on artificial data with "memory" and then stupidly learned how to allocate it to real BPs.
That's all.
Well that's not the point, how come...don't you see, non-stationarity is inherent in such processes, clinging to the normal distribution as basically most, once there is a limit in the increments, everything is normal distribution, from the reverse, it will always be non-normal, even be non-existent, as such representation of the process, no distribution?
I have already attached a bunch of tests and screenshots, I don't know what else you need. There is no eternal TS, it is not difficult to make money situationally
I've already attached a couple of different tests and screenshots.
I am a guest in this thread. I just came by to share an article.
You're too late, not HFT)))) I had a lot of fun with this company 2 years ago, now it's not so good, but who knows, maybe they'll come up with something else.