Machine learning in trading: theory, models, practice and algo-trading - page 539
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 can't understand anything without the code.
I should make a simple Expert Advisor with a minimum set of functions, just to read the values of the indicator and write them in the log (or better - in csv file, for a better comparison later). The results of two runs should be different. Then this should be sent to Service Desk together with the Expert Advisor's code.
You can also attach the code there -https://www.mql5.com/ru/forum/1111/page2096 - knowledgeable people will check and confirm or explain what is wrong.
Okay, I will try to write something there. I am not testing an EA in Agent, but an indicator, which calls an indicator from another symbol. Maybe there is a difference in it. I don't know ... The time goes, the problem persists :-(
https://geektimes.ru/post/294617/
i.e. it's all a forward without any adaptation to the market :) a lot of noisy trades that i have to get rid of as well
MO is certainly very hard on the brain
F
I finally started to get something with the adaptive system that retrains itself... I really spent a lot of time on it and it is still a rough version, but it is already something and the equity graph is awkward, but "honest", i.e. it is all a forward without any adjustment to the market :) a lot of noisy deals, which I also have to get rid of
This is very hard for the brain.
F
I would like to thank you for your comments.On the whole it is similar, yes ) but the model is a little bit ahead of the market by returns
and used m15 timeframe, that's why so many trades. I tried to put it off because I was sick and tired of doing it already :)
I can't yet. I've been busy with household problems for a year now. I have another month left. Then I will start to post the results, the source material is ready.
More than a month has passed. We are waiting for the story about practical use of GARCH, with pictures...
More than a month has passed. We're waiting for a story about the practical application of GARCH, with pictures...
Yes, I did.
I came across an understandable description of the LSTM neuron, wrote a little code to test it. Article -http://datareview.info/article/znakomstvo-s-arhitekturoy-lstm-setey/
In the code I take a 100 bar of eurusd m5, count increments for each bar and train the lstm neuron to predict the next increment based on the last known one.
Learning was done without complex analytical equations, neuron weights are adjusted on discrete lbfgs optimization, so it is worse, but for a simple test it will do.
The prediction estimate (R2) turned out to be a bit more than zero, which is very low, but still better than a random guess. Taking into account that the lstm neuron takes not some indicators or array of increases, but only one single value from which it predicts the next one, and it repeats for every bar, and generally everything is very simple - the result is better than I expected. But if I take thousands of bars then R2 score turns out to be < 0, too bad. It seems to me that the result of such a model worsens much at Forex using new data, I have to invent some bicycle with crossvalidation, the profit will not be obtained in such a simple way as now.
Now I need to somehow make up a network of these neurons, but that was not mentioned in the article.
I came across an understandable description of the LSTM neuron, wrote a little code to test it. Article -http://datareview.info/article/znakomstvo-s-arhitekturoy-lstm-setey/
In the code I take a 100-bar eurusd m5, count increments for each bar and train the lstm neuron to predict the next increment using the last known one.
I have made training without complex analytical equations, neuron weights are fitted on discrete optimization lbfgs, so it is worse, but it will do for a simple test.
The prediction estimate (R2) turned out to be a bit more than zero, which is very low, but still better than a random guess. Taking into account that the lstm neuron takes not some indicators or array of increases, but only one value from which it predicts the next one, and it's repeated for every bar, and generally everything is very simple - the result is better than I expected. But if I take thousands of bars then R2 score turns out to be < 0, too bad. It seems to me that the result of such a model worsens much at Forex using new data, I have to invent some bicycle with crossvalidation, the profit will not be obtained in such a simple way as now.
Now I need to somehow compose a network out of these neurons, but that was not mentioned in the article.
Lstm network even predicts seasonal cycles worse than arima, but it takes much longer to learn... I haven't understood usefulness of these networks for anything yet :)
I have a friend who was always excited by them, learned keras, took a simple series at work with seasonal profit and taught the net for almost a day... After that he swore at it for a long time