Discussion of article "Fuzzy Logic in trading strategies"
I have been waiting for such an article for a long time, but somehow it is not fully disclosed.
How does the system behave in the future after optimisation on history, is there a chance to make profit, and is it treading around zero?
I started reading the article and on the first example I got a lot of errors.
It turned out that there was no ENUM_LINE_END in Curve.mqh. So I wrote it. A bunch of errors appeared in Canvas.mqh. For example, there is no Attach.
I don't understand anything at all. The other day I used Graphics.mqh, I run my codes and they don't work either.
The dates of all files are from 05.09.2017 - when I upgraded to 1653. Yes and exactly everything worked in 1653.
I don't understand anything, and I apologise to the Author for this almost off-topic.
In order not to be unsubstantiated, attached the mqh files. I don't understand at all what could happen.
ZЫ I've set 1643 and 1653 - it doesn't help.
I started reading the article and on the first example I got a lot of errors.
It turned out that there was no ENUM_LINE_END in Curve.mqh. So I wrote it. A bunch of errors appeared in Canvas.mqh. For example, there is no Attach.
I don't understand anything at all. The other day I used Graphics.mqh, I run my codes and they don't work either.
The dates of all files are from 05.09.2017 - when I upgraded to 1653. Yes and exactly everything worked in 1653.
I don't understand anything, and I apologise to the Author for this almost offtopic.
In order not to be unsubstantiated, attached the mqh files. I don't understand at all what could happen.
ZЫ I've set 1643 and 1653 - it doesn't help.
Good day. You are using an old version of the Canvas.mqh file.
The current version of the file is in the attachment.
I have been waiting for such an article for a long time, but somehow it is not fully disclosed.
How does the system behave in the future after optimisation on the history, is there a chance to make profit, and is it treading around zero?
Hi, after optimisation any system behaves randomly on a new sample, if there was not at least a walk-forward testing, this article was not about that, but about the fact that MT5 has such a library, which nobody uses, and which can be used in original ways :) You can even make a neural network by feeding the outputs of several fuzzy logics to the inputs of others, and screwing some optimiser to them to adjust the weights. But there are already Fuzzy neural networks, but they are not in this library.
P.S. more about optimisation - since fuzzy reduces the probability of overtraining, it should still be better on new data, the question is to what extent fuzzy logic inputs describe the market. It's clear that 3 oscillators that describe roughly the same thing do so poorly.
I started reading the article and on the first example I got a lot of errors.
It turned out that there was no ENUM_LINE_END in Curve.mqh. So I wrote it. A bunch of errors appeared in Canvas.mqh. For example, there is no Attach.
I don't understand anything at all. The other day I used Graphics.mqh, I run my codes and they don't work either.
The dates of all files are from 05.09.2017 - when I upgraded to 1653. Yes and exactly everything worked in 1653.
I don't understand anything, and I apologise to the Author for this almost off-topic.
In order not to be unsubstantiated, attached the mqh files. I don't understand at all what could happen.
ZЫ I have set 1643 and 1653 - it doesn't help.
I don't understand anything either, try the version you were given, I didn't change anything in this library, if it doesn't help I can send you my own.
Good afternoon. You are using an old version of the Canvas.mqh file.
The current version of the file is in the attachment.
Good afternoon, Thank you, I am. I just don't understand where the old one could have come from and why every reinstallation of the terminal changed its date but not its contents? It must be a bug.
I don't understand anything either, try the version you were given, I didn't change anything in this library, if it doesn't help I can send you mine.
Thank you, I understand. I will read your article...
You can even make a neural network by feeding the outputs of several fuzzy logics to the inputs of others, and screwing some optimiser to them to adjust the weights.
Or we can take a ready-made PNN and feed it directly with the described classes and samples without a fuzzy layer. We get probability estimation of all classes/outputs, we can analyse the resulting basis functions as an analogue of membership functions.
I am not very familiar with them, unfortunately, I have only used Bayesian classifier and it turned out that it is no better than other linear models like the same regression +-. That's why I'm not sure if I should use PNN instead of MLP or RDF, I'll probably describe Random forest in the next article, it's fast and the quality of models is good.
I did experiments in Microsoft Azure Studio, there you can quickly compare models on the same set.
- studio.azureml.net
Dividing the "shopping space" into zones of fuzzy logic. is not in itself a manifestation of "crisp" logic:)
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
New article Fuzzy Logic in trading strategies has been published:
The article considers an example of applying the fuzzy logic to build a simple trading system, using the Fuzzy library. Variants for improving the system by combining fuzzy logic, genetic algorithms and neural networks are proposed.
The resulting membership functions at the output, after optimization (the inputs remain unchanged since they were not optimized):
Before the changes:
Author: Maxim Dmitrievsky