Machine learning in trading: theory, models, practice and algo-trading - page 392
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
This is the work of the NS out of sample by the signal price
And this is entry by 0.00050 pips better than the signal.
The deals are smaller, but the id of the curve is much better and here the output is on the signal, but not 50 pips better. That is, in this case, the curve will be even more beautiful. Here's the stat advantage for you....
What do you mean? A link to JPrediction?
Literally, yes.
Literally, yes.
I sent it to you in a personal message.
Bottom line. Help to run JPrediction on GPU??????
I should rewrite it on mql first, no need in a separate program, it is not convenient
I copied it to the personal message.
Send it to
Well, you need to rewrite it in mql first, no need a separate program, not convenient
See
The result of the program is the code in MQL5. This will take a lot of time to rewrite. Now you just start it on your virtual machine and the result will be a code for MQL5. The main thing is to learn how to run it on the video card. Even a 20X boost will be a success. IMHO
So, the result of the program is the code in MQL5. It would take some time to rewrite it, but this way you can run it on a video card and the result will be the code for MQL5. The main thing is to learn how to run it on the video board. Even a 20X boost will be a success. IMHO
Strangely, why the author no longer supports the site, I understand the development from a certain moment gone into hiding
And paralleling on gpu is the first thing, what a person working with neural networks should think about)
It's strange, why the author no longer supports the site, I understand the development from a certain moment went into hiding
And paralleling on gpu is the first thing, what a person working with neural networks should think about)
Well, after version 3 the paralleling appeared, I just at that time wrote about it that it would be nice to parallele. Just as I understand the program twists the data and back and forth. Dividing randomly for 100 times and at each division training for 1000 epochs. In general, it goes something like this. In other words, the point was ON, OFF. He supported it, but on another site, and then as it disappeared. I think maybe sick or something. But because he disappeared so suddenly, I think that the return of it is not planned. Given that he once said that he would soon go, but where. Remains only to guess and use what we have, unfortunately :-(
But the latest version of the predictor is also quite workable. Here is the result of work from 05.22 to the present day with limits on 0.00050 pips better than the signal price. In my opinion, it is quite workable.
Well, after version 3 and appeared paralleling, I just at the time I wrote him about it, that it would be good to parallele. Just as I understand the program twists the data and up and down. Dividing randomly for 100 times and at each division training for 1000 epochs. In general, it goes something like this. In other words, the point was ON, OFF. He supported it, but on another site, and then as it disappeared. I think he may have been sick or something. But because he disappeared so suddenly, I think that the return of it is not planned. Given that he once said that he would soon go, but where. Remains only to guess and use what is, unfortunately :-(
Yes, in general it takes a long time to count by gesture, in the studio the decision trees were counted instantly, but let's see what kind of error, it's even interesting to compare models... does it make sense at all...