Machine learning in trading: theory, models, practice and algo-trading - page 2137
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've lost interest in talking to you.)
no problem)
No. I don't deny the usefulness of indicators, even oscillators. But you can't build a machine model on them that will take into account the real price change in pips. This is important in making a decision to buy or sell. The signal will be there. But the longevity is under a big question.
A market model must be built at specific prices and times. Machine recognition implies a characteristic in the form of price, not a time blurred derivative of it.
My goal is not to build a market model, only to capitalize on price movements.
I haven't tried the models just on the oscillators from the article, and I don't know what would come out of it, if you had read the article you would know that it's not just oscillators that are used there. However, their usefulness turned out to be extremely high.
The lifetime of the model depends on the recurrence of events and oscillators can perfectly aggregate the events regardless of the price scale in absolute values, they are mainly sharpened on the relative changes of the indicators, so even a significant price change in pips allows them to stay in their probability zone, which facilitates learning. As a result, the very nature of oscillators favors the creation of stable models.
In fact, all that needs to be predicted is the speed of the upper and lower trend lines drawn from the last two extrema,
from the velocity it is possible to construct (extrapolate) the predicted rays of the trend
So we need regression models. It's more complicated with the upper (limit line).
I do not set a goal in the form of creating a market model.....
What are you stressing ))
He's like a robot, he follows an algorithm... which was repeated for the third or fourth time today...
First he says he's busy with bullshit, then he says he doesn't know anything about MO, and then, when you ask "What do you do and how", he shoots this deep philosophical thought like "water must flow / fire must burn" and disappears, then everything repeats the same ... ))
So we need regression models. With the top (the limit line) is more difficult.
Yes, I tried to do it many years ago, nothing interesting came out, with horizontal levels is better, but I was not experienced then, maybe I preprocessed the data badly
=====
I've been thinking for a while, now I definitely see the whole data preparation quite differently than before.... well, you can try
===========
By the way, a fantasy riddle
Does anybody know how to preprocess the data so that AMO understands that it is the same pattern, how to make the invariance of AMO to the size of the pattern.
I don't countconvolutional nets...
I wonder, who knows?
What are you stressing ))
He is like a robot, goes by the algorithm ... which was repeated today for the third or fourth time ...
First he says we're just bullshitting, then he says he doesn't know anything about MO, and then when you ask - "What do you do and how," he says some profound philosophical idea like "water flows / fire ignites" and disappears, then everything repeats in the same way ... ))
I do not strain :) Sometimes, such unexpected questions and statements, as in the opponent, allow you to structure your thoughts to answer, which is useful.
Yes, I tried that many years ago, nothing came out interesting, with horizontal levels is better, but I was not experienced then, maybe I preprocessed the data badly
=====
A little thought, now I definitely see the whole data preparation quite differently than before.... well, it's worth a try
It's worth a try.
Gotta decide on a metric to look for, speed won't work because of different volatility. You need something like SCO, and something else that can characterize the segment.
I wonder if anyone knows.
In my predictors, I describe the relative position of neighboring vertices. I get a vector like this at the output of N values.
It is worth a try.
We need to decide with the metric to look for, the speed will not work because of the different volatility. We need something like SCO, and something else that can characterize the segment.
It will work if the problem of invariance is solved.
So, should I thin the tics?
Maybe with the help of NS we can figure out how to thin the ticks.
Load a tick series, and let the neural network thin the flow, until it satisfies some conditions.
Let the experts tell you whether it is possible or not...
Perhaps, the ticks are in perspective, I'm working on M1.