Machine learning in trading: theory, models, practice and algo-trading - page 2435
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
I don't weigh the grafts because I don't have them in my "knowledge base". Not yet, I have everything done in the form of images, like a person...
As for the width of the zone, you need to reveal the whole concept here, I'll tell the essence without revealing details, just because it's long...
We have a current pattern (the last n prices or whatever).
In the "knowledge base" I'm looking for the same patterns and see how they end up.
Of course I need to translate the patterns into a normalized coordinate domain.
We get a kind of chaos in the trajectories of the patterns, nothing interesting...
Just a remark: this is how you try to predict your trends, ZZ and other stuff (blue elipsis).
You're messing with the purest chaos.
Okay, let's move on, but is it chaos? If we look at it with a human eye, we see that there is a certain consistency in all three trajectories, namely, we see the zone from which price bounces down in all three cases
The algorithm does this search automatically.
The answer to the question :
This is the width of the found zone , it is constantly different for objective reasons.
Further we transfer the found zone back from the normalized coordinates to the absolute coordinates and get the expected rebound zone
(That's it, everything is simple))
ps The smarter the algorithm of searching for patterns in the BR and the richer the BR is, the smarter the robot is
I myself did a similar thing on a MySQL database. I was storing all the quotes of bars 50 bars back and 20 bars forward. Then I was looking for the similar ones to the current one. For the 10-20 most similar in the past, found the average in the future. Usually this average was a horizontal line with small fluctuations. About 3 years ago, I posted these graphs here.
Terrible brakes...
After that, I got into MO.
MO (NS/forest) is also a database, but orders of magnitude faster, with the possibility of averaging, sometimes averaging is a generalization.
1) Where do the percentages of probability come from then? Or purely, since more than half of the samples moved up and reversed, the percentage is 50+?
2) As I understand it, just looking for the maximum and minimum of peak ZZ essentially with a certain tolerance, which is the boundary.
3)From what metric do you do scale conversions to scale historical developments?
4)Are these probabilities stored with two coordinates describing the final pivot point, or do you store the entire path of price movement?
5) I used something similar for determining the take profit - I took a dozen different levels from which the bounce occurred on the history and averaged the value - there were different approaches.
6) I think your method is also suitable for take profit or MM switch, and you may get a successful nettler using these levels.
1) I always take the closest 20 patterns, if 15 of them have bounce out of 20, then 15/20 = 0,75, i get a high probability zone
2) Sort of yes, but the tolerance is not fixed but more adaptive, more like a human view. A clever clustering algorithm does this
3) The pattern is normalized and then everything is normalized to the pattern.
4) I don't know what you mean.
5) I doubt that it is similar, the secret is in the normalization of the space around the pattern, and the pattern itself is a reduced space, not the usual price
6) Looking for rebound zones to know where to buy and where to put a stop to make a nettler?
Waiting for the signal. The pictures drawn in the graphic editor are not impressive.
Yeah I drew them all day in Pint to impress you, sorry it didn't work...
I did a similar thing on a MySQL database. I kept all the quotes of bars 50 bars back and 20 bars forward. Then searched for similar ones to the current one. For 10-20 of the most similar ones in the past, I would find the average for the future. Usually this average was a horizontal line with small fluctuations.
It's the same as the net to train forecasts of the current trend, it's unworkable, I've been doing it for years... until I found a way to visualize what I was studying and realized a lot, I wish I hadn't.
Yeah, I spent all day drawing in Paint to impress you, too bad it didn't work...
This is the same as the network training to predict the current trend, it's unworkable crap, I killed many years on this... until I thought to visualize what I learn and realized a lot, it's a pity that only I
Visualized too. That's why I posted screenshots, because I had something to use.
It didn't work. Even with Picasso, it wouldn't work... I'm interested in what is, not in what I want to show as if it were. The signal best shows reality.
How many "knowledge bases" have you run in the tester?
I bet you don't even have a clear idea what it is, to say such nonsense.
Visualized it. That's why I posted screenshots, because there was something to it.
Did you normalize the data with respect to the pathotherms?
Do you even know how to do that?
Do you know why you should do that?
Do you know what the invariance of prices, volatility, and why it's necessary when trading?
...
I have 100 more questions that you don't have the answers to...
What's the point of proving something to me if you only want to show off your "eh"? like - "my opinion isn't worse than yours" WAY worse!
You recently wrote that you were testing a system with fixed stops and takeovers...
I don't even think it occurs to you that something is wrong...
Oh, well...
It's like telling the Indians how a TV set works, and every Indian proves something to you, argues, shows examples on palm trees,my opinion is as good as yours...
It's funny and annoying...
Everybody wants a ready-made robot for 30 fucking bucks a month...
How many "knowledge bases" have you run in the tester?
I bet you do not even have a clear idea of what it is, to say such nonsense.
Have you normalized the data relative to the patterns?
Do you even know how to do that?
Do you know why you do it?
Do you know what invariance to prices, volatility, and why it's necessary when trading?
...
I have 100 more questions that you don't have the answers to...
What's the point of proving something to me if you only want to show off your "eh"? like - "my opinion isn't worse than yours" WAY worse!
You recently wrote that you were testing a system with fixed stops and takeovers...
I don't know what's going on here.. .
Oh, well...
It's like telling the Indians how a TV set works, and every Indian proves something to you, argues, shows examples on palm trees,my opinion is as good as yours...
It's funny and annoying...
Everyone wants a ready-made robot for 30 fucking bucks a month.
The signal will be enough.)
Patterns are good. The problem is that the more you put a lot of pressure on them, the less often they will work. And the machine will have to be constantly retrained for new twists and turns.
It's all a matter of not understanding...
How does machine vision work? Do they retrain the tesla every second? There are more market patterns...
Or don't they have all that?
It's all a matter of misunderstanding...
How does machine vision work? They probably retrain the tesla every second? There are more market patterns...
Or is that all missing?
Tesla, as far as I know, needs constant recharging) without a battery sufficiently charged - it will not go, and the patterns will not work)
I'm clear with you...