Machine learning in trading: theory, models, practice and algo-trading - page 390
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
The experiment is good, but how to use its results in MT? Will I be able to unload the model and put it into the indicator to see how it works in the area outside the sample????
But the free version has a limit of 1000 transactions a month, i.e. it's not an option to use it in the tester, because there will be more than 1000 transactions during running in the tester. That's why there's a little hitch, I was just dealing with it
To quickly estimate the model and check predictors - yes, to use then this model in realtime - yes, to use the model in the Strategy Tester with calling through requesters - expensive, 30 rubles for 1000 calls to the server, for example for a month on 5 minutes in the tester it will be 8000 calls, i.e. 240 rubles for one run in the tester for a month on 5 minutes
Was the set originally prepared for the Reshetov classifier? By the way, it should be rewritten in mql5 for OpenCl, then it will be quick to calculate, a cool thing
Is it possible to process so much data with RNN? There are ready examples for 3-4 inputs, they can be expanded to 5-10. And it will be unrealistic to calculate something more, since the number of parameters to be optimized = 2N
Realistically, we need to rewrite it on gpu, can you do it?
Realistically, we need to rewrite it on the gpu, do you know how?
no
I'll rewrite it next week.)
I will rewrite it next week)
And how will you optimize it? Your own counter and your own genetic algorithm? Without using a tester optimizer?
Through regular optimizer, but NS will be trained on gbu, it has another version in java, I need to rewrite it on mql5
https://sites.google.com/site/libvmr/home/theory/method-brown-robinson-resetov
Through the in-house optimizer, but the NS will be trained on gbu
How will you select the coefficients? For 10 inputs you need 1024 coefficients, it's unrealistic. And dimensionality is too small for real tasks.
Read the link, you don't need to pick coefficients in the optimizer in this version.
the nuclear machine increases the dimensionality, it's a tricky system
So, first the model is trained on the graphics card, the k-values are saved, and the optimizer already picks stops and other stuff
How do you select coefficients? For 10 inputs you need 1024 coefficients, it's unrealistic. And dimensionality is too small for real tasks.
This is in Reshetov's RNN, a probabilistic model.
And then there's jPredictor, which Mikhail uses. Reshetov's neuron, it has a lot of inputs, and some kind of training instead of gradient descent.