Machine learning in trading: theory, models, practice and algo-trading - page 1006
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
First of all, you have to define what do you mean by fractals: an indicator, like in the picture, or some kind of mathematical model?
If the indicator, then on mcl4 made at least 100 bars on the left and right, it will count.
I've written before examples, of course the model. What you have in the picture is a smoker's fractal.
I wrote earlier examples, of course the model. What you have in your picture are smoker's fractals.
The fractals you see in the picture are smoker's fractals - well, there are whole communities that use them to draw trend lines and trade on them.) But I've tried it in my tester and I know what's what.)
Ah. Yes, I remembered something about Almazov).
What is the problem with automating, complex algorithm like arim?
Ah. Yes, I remembered something about Almazov.)
And what is the problem with automating a complex algorithm like arim?
It's not clear what to automate... whether it's patterns or correlations
i decided to try one of the traits of fractals - self-similarity of sorts, but i think nothing will come out of it
I don't really know what to automate... whether it's patterns or correlations
I decided to try one of the features of fractals - self-similarity of sorts, but I don't think it will work, I'll give it a shot
I've been wanting to ask a question for a long time. Suppose I have a 150-bar head and shoulders pattern. I need to find similar patterns, but they will be found if the number of bars is almost the same in the pattern itself and in the pattern found. How to get away from the exact number of bars to look for a frequent one and derive the average for it or something else?
I've been wanting to ask a question for a long time. Suppose I have a head and shoulders pattern 150 bars long. I need to find similar patterns by history, but they will be found if the number of bars is almost the same in the pattern itself and in the pattern found. How to get away from the exact number of bars look for some frequent one and output the average for it or something else?
I don't know, just take a bigger pattern at random and repeatedly adjust it to the necessary size.
ah, you can also do it via convolution, but I don't know howI don't know, I could thin a bigger pattern randomly, and then take it all several times to the right size, which time the correlation will be better and take it as a copy (if they do not differ much by the number of bars).
I can do it with a convolution, but I don't know how.I thought it would be possible to additionally gather a collection of history and then compare every pattern, it will be long, but it can be quickly calculated using hourly data.
I don't get it, who is that Gramazeka1 is hiding behind my name? What's the deal at......????
Just kidding, I'm all that... Я!!!!
I'm sure many are familiar with Reshetov's work, but no one has fully understood his concept. One of the points of his work is sticking to the Shepley axiom. Frankly speaking in simple language, the sum of weight coefficients of polynomial network is equal to one, or minus one. Thereby optimizer's task is to find coefficients of network, which are distributed in limits from 0 to 1. It is not important which length is polynomial, it is important that sum of coefficients equals unity and by including in training of any non-informative sign we select coefficient resource for it (resource is limited from 0 to 1) due to coefficients of informative predictor and make it less significant. That is why this algorithm is demanding for data preprocessing. The better we clean them out, the better will be the training result and operation of the model on AOS in general. As far as I know, none of the packages and network sets in R use Shepley's axiomatics. Hence the result....
"You should learn to read books first, instead of burning them" I mean that none of the representatives of this thread bothered to go into the essence of the work. Looked superficially and successfully threw in a drawer ...... IMHO!!!!!
that's it... and how are the conversions done there, any examples? There are kernel tricks in my opinion... and it dawned on me that my magic algorithm is unbearably lacking there
Now I was thinking that in addition you could build a collection of history and then compare each pattern, it will be long, but it will quickly count on the hour markers.
The funny thing is that the pattern can drift so much that no methods are able to determine its similarity... you will have to do affine transformations for each of them.
And if the pattern is floating, everything after it is floating too.
and it's pretty easy to find similarities by eye... :) that's why I can't automate it
Misha, could I have a look at your trades? To hell with the signal, but show me a couple or three real trades with comments.
The trick is that the pattern can float so much that no methods can determine the similarity... you have to do affine transformations for each
and if the pattern is floating, everything after it is floating too.
I have such a case.