"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 71
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
Can I have a little example of how to use it for MT5?
Modern GPUs have 1GB of RAM or more
I can hardly imagine a training sample of larger size
If I'm not mistaken, modern graphics cards have two kinds of RAM - shared, which can be several GBytes and which is quite slow, and the one that each individual GPU processor has - fast, but a small amount of a few hundred KB. I reasoned that this is where the problem might arise.
There is nothing wrong with the macro level - tested )
But, if you say so, I take your word for it and now I am calm. :)
For GA it is reasonable to use the following architecture: GA itself on CPU, and heavy FF computations on GPU
Well, this is, as they say, a matter of technique. You can also do it this way. As far as I know, OpenCL allows you to select in the code exactly on which cores the computations will be run - on CPU or GPU.
An example of what? A neural network indicator, a neural network strategy, ...? ?
If I am not mistaken, modern video cards have two kinds of operating memory - common, which can be several GBytes and which is quite slow, and the one that each individual GPU processor has - fast, but a small amount of a few hundred KB. I reasoned that this is where the problem might arise.
General "slow" is the full analogue of DDR for CPU (~ 0.75 - 1.00 GHz - not very slow)
Fast is the analog of the CPU's cache memory.
GPUs allow to manage this very on-chip (cache) memory, unlike CPUs (I could be wrong, but somehow I have not come across such a question)
But all these are additional optimization issues, you can live without them
her.human:
An example of what? Neural network indicator, neural network strategy, ... ?
Yes yes, we would like such a push example and then we'll figure it out.
No way, tester GA is limited to 64 bits of search, and for training weights for each weight you need 16-64 bits (depending on the accuracy of the search), and weights can be up to..... several thousand, and can be tens and hundreds of thousands. UGA can cope with this quite well, and the tester can not do it.
So, either MQ should be given an order to upgrade GA (what I've been doing for two years unsuccessfully) or UGA should be built in, but it will be much slower than in cluster.
General "slow" is the full analogue of DDR for CPU (~0.75 - 1.00 GHz - not very slow)
Fast - this is the analog of the CPU cache memory.
GPUs allow to manage this very on-chip (cache) memory, unlike CPUs (I could be wrong, but somehow I have not come across such a question)
But all these questions are additional optimization issues, you can live without them
Not the point, memory, and on the GPU will be difficult to implement object logic twists, I have a pretty good idea how to implement the NS, as a complex interaction of objects within which there are functions of GPU calculations (this is about the micro level), but I do not see how this complex object code those whole NS to slip the GPU as several FF with different sets of weights (macrolevel) ???
The question is rhetorical, although if anyone imagines not mind to listen.
I have a funny idea:
Maybe to ask the MQ API to set tasks for cluster?
to distribute the task without passing the tester, directly from MQL5. The task is certainly difficult, there are a lot of checks to set, but it is manageable.
I've already started working on this question but I've never understood it.
Yes yes, we would like such a push example and then we'll figure it out.
Example 1:
The problem statement:
- You need to restore the algorithm of the indicator, the sources of which are not available (or it is a "re-drawing" indicator - by the way, it is not a bad teacher for NS), but the values on each bar are known
- indicator values are in the range (-1;+1)
Inputs:
- we know (suspect?) that this indicator uses the last 10 closing prices
Example #2:
Problem Statement:
- We need to select a neural network trading strategy that trades best (maximize Profit/MaxDD) on a stretch of history
Input data:
- Normalized MACD (13,35), MACD (50,100), RSI(8), RSI(13)
Example 3:
Problem Statement:
- To teach a neural network a multiplication table
Input data:
- x,y multipliers in the range (-1;+1)
You can offer your own example and, if it's possible, I will show you the network architecture in XML
You can offer your own example and, if possible, I will show you the network architecture in XML
and most importantly the source code of how this architecture was created, how the XML file is generated.
and most importantly the source code of how this architecture was created.
Are the two inputs the values of some indicator or will someone external feed the data?