Zigzag indicator and neural nets

 
Greetings romantics from the high road ! :)))))))
Interested in the zigzag indicator. I came up with the idea while thinking about neural nets. The thought process was as follows. I want to look deeply into the history without using too much data. In the article Predicting Prices with Neural Networks the weavelet decomposition was mentioned. What does the author thinks it is, I still do not understand it. If we take the standard definitions and Haar's wavelet, then wavelet decomposition is a set of bags (I didn't prove it strictly, but it seems so). Why it helps so much to look into history, kill me a herd of moose, I don't understand. The zigzag, on the other hand, is a much more interesting thing. It's a piecewise linear approximation of the curve. It takes only the most interesting things from history - the extremes and their times. I've written my zigzag. For the time being it's in Matlab, although I hope to port it to mql4 at weekend. My zigzag is looking for moves no smaller than a given step. I've done some experiments with it. In particular, I was interested in how many pieces the zigzag would be composed of, depending on a step, and at what step the largest number of new pieces appear. Alas, I now have to run. At weekend I hope to systematize my results and post them with sources in matlab and mql, and with pictures. In the meantime, please. As I understand there are a lot of different zigzags. I wonder if you know some links where all information about zigzags is systematized and abstracted? Share please! His Pomosh Purr-face will pray for you to his Master and Mentor, the Great Enlightened Meow ! :))))))
 
See nen' s corner on Onyx, what's there about zigzag and other related stuff. Do you need a link or do you know how to find it yourself?
 

The maximum accumulation of workable ZZs and their analyses are collected on nen's onix. It seems to me.

There are his articles here in CodeBase.

Added: greetings Mathemat, with synchronicity you/us :).

 
Hello, Bookkeper. Have you been seriously digging in that direction as well? The direction is indeed very promising, but also very time-consuming to automate. Anyway, nen himself on Onyx (and, I think, in an interview on the Champ website) mentioned that his own system is not a robot.
 

Have a look at the next thread: 'H-volatility FX' - maybe you'll find something useful...

 
Thank you guys ! Alas, I'm out on my mobile at the moment, but I'll be sure to check it out tonight !
 
Mathemat:
Hi Bookkeper. Have you dug in this direction seriously as well? The direction is indeed very promising, but also very time-consuming to automate. Anyway, nen himself on Onyx (and I think in an interview on the Champe site) mentioned that his own system is not a robot.

Nah, I dabbled at first (ZigAndZag) and now I'm doing a zigzag sewer very slowly, with long interruptions, on a very low-arithmetic level, likely to end up dabbling too :). Never think first what it's for later, I'm having so much fun.

For example, I've been thinking about ZZ incorrectly for a long time: with ZZ one should not try to predict price levels, but the bars of formed reversals where they will be (in the zone to the right of #0). And at what price level - who cares? But it's going to take me a long time to get to that point.

 
Well, I've already expressed this point on Onyx: it's not just the price that is predicted, but also the time. The accuracy of time prediction itself is much lower than for prices (that's what wave analysts say), so it's hard to rely on one time. Besides, the reversal bar itself, even if it is accurately predicted, may be very large; where to open within it? More precisely - when?
 

On the timeframe (H4, D1, ...) we look for the reversal bar as a time interval, and the "moment" of the reversal itself - on a small TF within this interval (if post factum - with a lag), a set of statistics, .... I have not got to the details yet, but I have not got to the details.

The reversal bar is small or large - it is monopoietic, it is not for pipsing.

 
Bookkeeper:
Mathemat:
Hello, Bookkeper. Have you dug in that direction as well? The direction is indeed very promising, but also very labor-intensive to automate. Anyway, nen himself on Onyx (and, I think, in an interview on the Champ site) mentioned that his own system is not a robot.

Nah, I dabbled at first (ZigAndZag) and now I'm doing a zigzag sewer very slowly, with long interruptions, on a very low-arithmetic level, likely to end up dabbling too :). Never think first what it will be for later, I amuse myself that way.

For example, I've been thinking about ZZ incorrectly for a long time: with ZZ one should not try to predict price levels, but the bars of formed reversals where they will be (in the zone to the right of #0). And at what price level - who cares? But it's going to take me a long time to get to that point.


Apparently thoughts are material and fly through the air. That's what I think too.

There is a "TournamentPoint" addon in NeuroshellDayTrader program. So, there is an indicator in it, which calculates the probability of occurrence of the next extremum of Zig-Zag.

Works pretty well. I.e. a neural network that has readings of this indicator on its inputs is practically sink-proof. But that is not the main thing. The main thing is to find out the algorithm to calculate the probability of Peak or Trough.

Do you have any idea how to calculate it?

 
klot:
.........
I have already written - no real ones yet. Except: Convinced of the asymmetric movement up and down, which means that Zigi and Zagi must be calculated separately, without connection. While I sit and think, and I have this process difficult, and most importantly long (see if someone has time to do something :).