"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 72
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 two inputs are the values of some indicator or will someone external feed the data ?
external data, as well as the output, you can display can exclude, as desired.
ZY this is a drawing from an XOR problem, classic MLP
Now I just had a stray thought:
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 complex, it requires a lot of checks, but can be solved.
What do you think, brothers, is not too high sweep? cluster certainly sacred cow, but we have to work hard.
It would be very cool! But the question would be how agents would be remunerated in this case?
How about this?
Otherwise the meaning of neurons 1, 2 and 8 is unclear
How about this?
Otherwise the meaning of neurons 1 and 2 is incomprehensible
First xml-file: network before preprocessing -UrainTaskPre.xml
When you have a lot of neurons of the same type, it's easy to make a mistake - here you define a template and preprocessor should unwrap it into a ready-made network file
Then the preprocessor handles substitutions <REPEAT .../>, <INCLUDE .../>, "expr:..." and we get the network
UrainTaskPost.xml
Below is a script that opens an XML file, parses it into an XML tree, creates a neural network from it, initializes the inputs and here it Run()
Note:
The class CNeuronService handles neurons of type "EXTERN" - a little inconsistency left over from the old days
It's easier to get used to it than to correct it )))
Results of the script:
UrainTaskPost.xml must be placed in a common terminal folder, for example, C:\Documents and Settings\All Users\Application Data\MetaQuotes\Terminal\Common\Files
Files from the archive to MQL5\Include
First xml-file: network before preprocessing -UrainTaskPre.xml
When you have a lot of neurons of the same type it's easy to make a mistake - here we set a template and preprocessor should unwrap it into a ready-made network file
Then the preprocessor handles the substitution of <REPEAT .../> and we get the network
UrainTaskPost.xml
Below is a script that opens an XML file, parses it into an XML tree, creates a neural network from it, initializes the inputs and there it Run()
Already warmer, now explain how to interpret the file.
Where to write what to create links, or poke where it says they are set.
Already warmer, now explain how to interpret the file.
where to write what to create the connections, or point where it says that they are set.
1) Each neuron must have a unique name, for example NEURON:1.HID:0
The name tells you that it is the 1st neuron of layer HID:0
2) The links use for addressing the neuron names + weights
3) every neuron class can (and must) override the virtual bool Init (CXmlElement *aXmlElement, CNet *aNet) and load from XML
P.S.
XML files are better to edit with Notepad than with Excel ))
1) Each neuron must have a unique name, for example NEURON:1.HID:0
The name implies that it is the 1st neuron of the HID:0 layer
2) the links use the neuron names + weights for addressing
3) every neuron class can (and must) override the virtual bool Init (CXmlElement *aXmlElement, CNet *aNet) and load from XML
P.S.
It's better to edit XML files with Notepad than with Excel ))
Yes, I've edited it with both Excel and Word and Notepad, but it's not obvious, it's just two neurons, and it will be 200-2000, and each will have 100-200 links.
If I have to fiddle with it, I want it to be graphically useful or display, or, even better, graphically editable.
If you're going to do something, it should be more convenient than this: columns where we get it from, rows where we distribute it to
Yes, I've got it in both Excel and Word and notepad, well, not obvious, it will be a wreck, while two neurons and will be 200-2000 and each for 100-200 connections.
If you have to fiddle with it, it would be easy to graphically display, or even better to edit graphically.
That's what PreProcessor is for.
Whether you have 2 neurons or 100, the file size is the same.
Well, if you want to edit graphically, no one is stopping you - just write a GUI and use
PreProcessor is exactly for this
Whether you have 2 or 100 neurons, the file looks the same
And if you want to use graphics, you're free to write a GUI and use it.
Well, let's not hack off the shoulder, let's lay it all out in clear lines. I do not know that for others the same, but for me the way of loading that I posted above is obvious in terms of the algorithm of loading:
read a string, initialize 3 layer type objects,
read the first line, then call the first layer object, construct layer type input, initialize array of 2 neurons.
read the second line, then call the second object layer, construct the layer type mlp, initialize the array of 2 neurons.
read the third line, then call the second object layer, construct the mlp layer type, initialize the neuron array of 1.
Since the number of layers is 3, we finish the construction on the third line and assign connections.
for which we go through the table and call the necessary neurons, telling them the connections if the table contains 1.
And how would the algorithm work in xml ?
And the main advantage is that since the connection table is binary, it's perfectly shrinkable. The table for 8 neurons fits in one ulong. True, the farther into the woods the thicker the partisans in general the size with a large number of neurons is calculated as
X=N*N/8
where N is the number of neurons,
X is the number of bytes.