"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 25
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
If you need an example code for the differential evolution method, I can send you one. The method is quite simple. It looks for global extrema.
No, that's not what I said.
The period of testing and optimization is as important a parameter of the system as the value of take profit if it is used for example. And it is silly to remove some data or add some data just because the network is retrained. It is better to change something in the conservatory (for example, change the topology of the network).
TheXpert, thanks, I'll look into it.
Create, gentlemen programmers, such a NS and we will start testing it.
You seem to program a little bit.
Let me give you a dll with a grid and show how to use it (it is not difficult), and you provide a model for tests and discussions. After all, it's you who create the inputs and outputs, the grid doesn't care what to learn.
True alas, I can only provide an example for 4k. Grid echo grid.
The source code of the dllkas I do not want to lay out yet. If you are afraid, I can give the sources to a trusted person to compile.
I can put the assembly (dllka + EA) also for all to see.
The testing and optimization period is as important a parameter of the system as the take profit value if it is used for example. And it is silly to remove some data or add some data just because the network is retrained. It is better to change something in the conservatory (for example, change the topology of the network).
If you were involved in networking about 10 years ago, you should first refresh your knowledge in the relevant field before you say anything.
Come on. MLP is as good as it gets.
Let's not fight, gentlemen, that's my point.
Let's not fight, gentlemen, that's what I'm saying.
No one has yet.
But it is necessary to understand, why they reduce number of neurons in layers and increase number of samples in Sample (the essence is the same in both cases - decrease of degrees of freedom of neurons) - naturally, up to certain limits - until the network stops learning.
Who needs this Sisyphean task of creating a neural engine, if no one will understand how to work with it elementary. Looks like you'll need to write a manual for dummies on how to work effectively with the engine - like useful tips. Otherwise we will face a lot of accusations from ordinary users, traders: "I use neuro-engine, but my deposit is melting before my eyes - what's wrong with it? I'll complain!".
Well, that's what it's desirable to have someone competent, but not a programmer.
There is no need to make a spaceship now, no matter how much experience and baggage each of you already has under your belt.
You don't need a car or a bicycle, just make a scooter. Simple, dumb, understandable and reliable. It will save you at this stage from arguments about the formula for the best fuel for a starship.
The mere fact of having a working scooter will attract attention. It won't solve the super problem? It doesn't have to. It is enough that it will attract attention and it will be easier for you to come to a compromise on a simple one.
I will formulate what I need for trading. The other traders will add.
...
This is what I would like to get.
Create, gentlemen programmers, such a NS and we will start testing it. The test results will be posted on the forum and we will discuss them. With this approach, traders and programmers for a short time will start to talk in a language that is understandable to both sides.
From the foregoing I conclude that:
It is necessary to have three blocks
1 block of preprocessing
2 NS
3 block post-processing
preprocessing will prepare and distribute the necessary data to the inputs.
The NS is dangling between them.
Post processing includes both samples of outputs and interpretation of outputs for Expert Advisor.
Preprocessing and postprocessing should be in the same class, otherwise there may be difficulties with the correlation of inputs and output patterns for training, this class exports two output arrays (inputs and outputs) to the NS, the output array is exported through mutually exclusive methods (in training mode only read, in working mode only write).
Thus, the user saturates the virtual methods of pre and post processing with his own, connects to the class of pre_post_processing of NS and calls the class from the EA. The Expert Advisor class only gives post-processed data (receives data directly from the market environment).
Something like this.
Maybe I can do it. I don't qualify as a programmer, I don't have two degrees. But I'm good at something.
Victor, any help is welcome.
No need to make a spaceship now, no matter how much experience and baggage each of you already have under his belt.
You don't need a car, you don't need a bicycle, just a scooter. Simple, dumb, understandable and reliable. It will save you at this stage from arguments about the formula for the best fuel for a starship.
The mere fact of having a working scooter will attract attention. It won't solve the super problem? It doesn't have to. It's enough to draw attention and it will make it easier for you to compromise on the simple.
I agree. Let's make a scooter/bicycle (I mean, you don't have to invent it, you just need to implement it in MQL).
Let's set a specific task and think about how to maximize its effectiveness. For example, imagine that I don't know what MQL5 is (and I know OOP only at dummies level), I've never dealt with NS (not that with NS, I have a poor understanding of Expert Advisor systems, etc.). But I can trade hands and have a certain TS.
So, let's imagine that all my knowledge was enough to formulate the following:
1. I need a neuronet, which based on certain data and certain rules (not known in advance) will produce a certain result in the form of a certain signal. At output we will deal with the following variants: buy, sell, wait.
At that signals must come from various sources. For example, it should be: simple candlestick patterns, standard indicators (CCI, MA, RSI, Stochastic, etc.), the price feed and other things that can come to a trader's mind.
The logic and functionality of the network will be realized in what form, and how exactly the network will be trained is not important (let's assume that for me the NS is a BLACK/BLACK box). But in terms of training it is important to have an opportunity to train using the data from a pre-prepared file (or several files), for the actions of a trader / expert on a trading account (maybe some other options will be available).
4. The set of rules should be able to be saved in a file and downloaded from it if necessary.
Proceeding from all mentioned above, we need to get a NS which will consist of "two" layers of neurons (remember, I'm a dummie!):
a) The first layer consists of neurons analyzing a primary signal (indicators, candlestick patterns, stream of quotes, etc.). In a simplified version, each neuron processes only one input source. At output we get three variants - buy, sell, wait.
b) Receives coded signal (buy, sell, wait) from one or more "initial" level neurons. Generates a single signal based on the rules.