Discussion of article "Thomas DeMark's Sequential (TD SEQUENTIAL) using artificial intelligence" - page 2
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
You send me the file for training, I'll send you the model, and then test it and post the results. OK?
That's right!
Only the model is desirable in "humane" form, executable file (exe, jar) that one click to load the test dataset in the same format as was training, and the other to select the path to save the result where you need in csv.
The training dataset will be csv with delimiter "," (comma), 15 features and target last in total 16 approximately so:
..................
-1.129204,-1.129167,0.282294,-2.280221,-13.081081,1.855654,0.163391,6.384071,0,0,0,-1.434602,1.165473,6.521727,0,1
0.141149,0.141149,1.411552,0,4.704501,0.001642,-1.564355,0,0,0,0,-36.086637,-0.156291,-10.859675,0.200637,1
0.282292,0.141146,0.423445,0,0.613683,-0.355847,-0.328989,-1.063462,3.190799,3.191211,0,4.071769,3.565043,20.779214,-2.33066,1
1.411447,0.282292,0.141146,0,-0.49115,1.463979,2.700361,-1.063508,-1.063462,3.190799,0,15.394189,0.511692,-7.217575,1.160668,1
1.975945,1.411447,0.282292,-1.14011,3.160372,2.471691,-5.125828,-2.127108,-1.063508,-1.063462,0,-10.655282,1.961731,9.919539,-0.581819,1
0.282257,0.564523,1.975945,2.280221,-7.463255,-0.013203,-3.919166,0,0,-2.127108,0,-1.451862,-0.3113,-5.870295,-1.00175,-1
-0.705637,0.282257,0.564523,0,3.541205,0.261354,2.416635,0,0,0,0,3.18908,0.275705,13.998395,1.789844,-1
0,-0.282271,-1.270172,0,15.3989,4.815902,-1.18679,0,0,2.127291,0,-2.530371,-0.423919,6.862249,-0.031438,1
-0.141131,1.27023,0,0,2.443412,-0.05624,-8.284345,0,0,0,0,-3.995578,0.231936,4.123271,0.620976,-1
...............
In the attached file both lern and test(for your convenience), but I will check the model on a control test, from the next trading day, which you have not seen. If you train on lern(train.csv) and achieve on test(test.csv) accuracy >=~65% and it will be about the same (-1%) on my control test, well.... so Reshetov's classifier is a cool thing and he published it for nothing))))))
That's right!
Only the model is desirable in a "humane" form, executable file (exe, jar) that one click to load the test dataset in the same format as the training one, and the other to select the path to save the result where needed in csv.
The training dataset will be csv with delimiter "," (comma), 15 features and target last in total 16 approximately so:
..................
-1.129204,-1.129167,0.282294,-2.280221,-13.081081,1.855654,0.163391,6.384071,0,0,0,-1.434602,1.165473,6.521727,0,1
0.141149,0.141149,1.411552,0,4.704501,0.001642,-1.564355,0,0,0,0,-36.086637,-0.156291,-10.859675,0.200637,1
0.282292,0.141146,0.423445,0,0.613683,-0.355847,-0.328989,-1.063462,3.190799,3.191211,0,4.071769,3.565043,20.779214,-2.33066,1
1.411447,0.282292,0.141146,0,-0.49115,1.463979,2.700361,-1.063508,-1.063462,3.190799,0,15.394189,0.511692,-7.217575,1.160668,1
1.975945,1.411447,0.282292,-1.14011,3.160372,2.471691,-5.125828,-2.127108,-1.063508,-1.063462,0,-10.655282,1.961731,9.919539,-0.581819,1
0.282257,0.564523,1.975945,2.280221,-7.463255,-0.013203,-3.919166,0,0,-2.127108,0,-1.451862,-0.3113,-5.870295,-1.00175,-1
-0.705637,0.282257,0.564523,0,3.541205,0.261354,2.416635,0,0,0,0,3.18908,0.275705,13.998395,1.789844,-1
0,-0.282271,-1.270172,0,15.3989,4.815902,-1.18679,0,0,2.127291,0,-2.530371,-0.423919,6.862249,-0.031438,1
-0.141131,1.27023,0,0,2.443412,-0.05624,-8.284345,0,0,0,0,-3.995578,0.231936,4.123271,0.620976,-1
...............
In the attached file both lern and test(for your convenience), but I will check the model on a control test, from the next trading day, which you have not seen. If you train on lern(train.csv) and achieve on test(test.csv) accuracy >=~65% and it will be about the same (-1%) on my control test, well.... so Reshetov's classifier is cool and he should have published it for nothing.)
OK. Let's continue the conversation tomorrow, because I admit my computer is counting now. And the time is already late. Tomorrow!!!!
Well, I looked at the training file, it has 16 inputs but 65 thousand lines. I will not even make it, because you have not understood the main theses while reading the article. One of them says that it is utopia to analyse the market every bar on minutes for 5 years of history. In other words, you are trying to build a model for the whole market, so to get a good level of generalisation will not work, as there are no grails. I build models within 30 lines, which covers the market for the last 2-3 months on 15 minutes. So I get a model of much better quality, but it does not work for a long time, maximum two weeks and that is enough for me. And you are trying to build a model for the whole market. This is definitely utopia. Read the article again, but carefully!!!!!.
P.sy I will not count your file, because it will take a week. It is a useless waste of machine resources and time.
I took the first 50 lines of the traine file, built the model and you can check for yourself how well it worked.
Unfortunately I cannot attach the archive. MQL forum is as always at its best. The most popular formats are not supported :-( I can send it to mail.
It is not minutes but seconds, training set from 2 weeks of trading, test set from one day and it is a very simplified version, date for one instrument, the simplest forecast for a second in front, one target, I have a model training for such data takes about a minute.
30 lines)))))))))))))))))))))))))))))))))))))))))))))))))) Uh-.... what can I tell you... good luck! You absolutely need it))))
Seconds for a fortnight with 65000 lines. I think it's you who needs luck. No, I'm serious. This approach has failed miserably. Although on 50 records of your file I got 70% of generalisation. In principle, the data is good enough for output. But unfortunately you do not understand the essence of the game, if you are trying to build a TS on seconds :-) All right, then here's a suggestion. Build a TS according to my recommendations, train it on your own data and on your own AI and you will see that the result will be much better than your approach. I'm not serious, choose a TS, (crossing of mashes) select the necessary number of inputs. Save the data in the "context of the day" and train your AI on it and you will be pleasantly surprised. Well, I will try to install the zip, I will throw the archive, although I think it will be useless for you, because the models are in the form of code for MQL4. and certainly not an exe file. Can you manage???
And in addition, the pound for today, both sell signals hit the same area and we can see that the first signal brought a minus. But since both signals fell into the same area, in order to profit on the second signal, it had to be reversed. That is to go against the signal, in this case we got an anti-model. The result is on the face of it!!!! And my 30 signals cover the market for about two months, so judge for yourself.
and green dots warn the trader that a signal is about to appear, this has its own advantage, that the signal does not appear as a snowball, but we expect it in advance.
Here's the model I built this morning, you can check it out.
Read.
Well done for deciding to post your TS and explain roughly how it works, plus explain a bit of theory in terms of your thoughts on the work of neural networks directly with the Forex market.
However, the article is too much "trader can guess", "trader should make a decision based on his own experience", etc. All very vague for such a loud title of the article.
As I understand, you are not a programmer, otherwise the article would be more informative, systematic, and you would refine your TS so that you would not reverse positions.
Then you would have posted the results of trading in the tester for a year or two, pre-training the network every fortnight.
Reversal is always at your own risk, there is no clear algorithm - this is a big disadvantage.
It's not even an article about neural networks, it's just a description of your TC.
I expected more specifics, more words about detailed work of the proposed neural network, examples of test samples, training examples, examples of work after training.
Metadological articles are not needed here, beginners will never get it right anyway, and knowledgeable people are not interested in it. People who have been in forex for a long time, who know programming, often look for an interesting idea with a detailed explanation of why it is a good idea, how it works, and how to use it. Then they adapt it to their requirements, build it into their Expert Advisors, indicators, etc.
For example, I have a data clusteriser based on Kohonen Neural Network in C++:
In it, the picture on the left is the original data, the picture on the right is after clustering, with the class name signed in each class of two digits, the minimum distance in the class, and the 5 lines that go to the neuron that defines the class . The network consists of 7*7 neurons. There are 49 classes in total.
You don't have a single example, what exactly you give input data and in what format, what you get as output, description of the learning algorithm, etc.