"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 37
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
Meta EngiNeuro (MEN) (c)
so we are engineers :)
Meta EngiNeuro (MEN) (c)
so we are engineers :)
Shit. Meta is no longer the cathetus in such a name...
EngiNeuro is good (thumbs up)
EngiNeuroSolution ?
Enjin and Solution don't mix, here it's essentially the same thing
Vladimir I have a slightly sideways question,
how many (approximate) iterations of Learning Passes are needed to train the above networks?
ZY at least the order of magnitude.
how many (approximate) iterations of FF are needed to train the above networks?
What is a FF?
Fitness Function, I didn't express it quite correctly, FF is in GA, in network learning algorithm it is how many training passes.
If training examples (input output) do not exist, then the parallel with FF GA is direct, both there and there you have to do direct calculation of the grid and then probably by the results of the pass to do postprocessing to check the value through the trade function.
About the logo. Immediately ask for something like balls-nodes with connections, but this is in my opinion stupid. You can take for example puzzles, remove the short-range, change the colors on the metaquot, etc.
At least the puzzles have some kind of integration and interconnection.
Vladimir I have a slightly sideways question,
how many (approximate) iterations of Learning Passes are needed to train the above described networks?
ZY at least the order of magnitude.
The filters are trained without a teacher by presenting randomly selected 10000-40000 sections of history (the same number of iterations). Learning is very fast. Depending on the computer, 10,000 filters are trained on 20,000 sections of history in 1-2 minutes on 360 GPU CUDA processors, about 1 hour on 4 Intel processors with 16 tracks, 3-4 hours on my laptop with one processor and two tracks. Time doesn't matter here, though. Even if it takes me a day or two to train filters like this, it is only done once for each quote (EURUSD, USDJPY, etc.). Once the filters are trained, they do not change and are used to filter new prices. Filtering itself is very fast - we count the sum of products of price and filter coefficients.
By the way, you may ask why there are so many filters - 10 000? I cited data from my own image recognition project. For quotes, there will be much fewer filters, maybe 10-100, the less the better. Here is a rough analogy. Quotation is speech. The filters are the phonemes that make up words. There are 43 phonemes in Russian(https://ru.wikipedia.org/wiki/%D0%A4%D0%BE%D0%BD%D0%B5%D0%BC%D0%B0). Our task is to find a primer on the phonemes of quotes.