Probabilistic neural networks, packages and algorithms for MT4 - page 13

 
I agree it depends on the type of neuron maybe the network should have different neurons, some give 0 1 and others a range... If a network is to solve a multiplication table for example in the range of 1 ... 100 at the output there should be a range at the input. Actually for Forex 5 values at the output are enough. I already wrote, but there is no way it is possible to send just 0 1 at the input a probable signal always lies in the range if we talk about a probability network
 
YuraZ:
But there is no way you can feed a simple 0 1 to the input, the probabilistic signal always lies in the range as far as the probabilistic network is concerned.
Analog to digital converters are designed for 0 1.
 
njel:
what to give as input and what to give as output depends on the activation function
often if the function is a hyperbolic tangent, then the inputs are normalized to -1...1 or 0...1
but who in neurosolutions compiled the dll?


In principle, what to give as input and what to get as output does not depend on the activation function. The only rule to be respected is that the input is dependent on the scope of the activation function. I.e. if you use a hypertangent, it is {+1;-1} if it is an arctangent, it is {-1.57,+1,57}, etc. ..... Be sure to scale (normalize) the data in the range of visibility of the activation function. Otherwise it will turn out to be nonsense!

Maths for a fifth grader!

 
<br / translate="no">
klot:
renegate:
Gentlemen!
So what shall we feed to the input of the neural network? What error functionality shall we choose?


Judging by the content, not many people are interested. Many people think it's about the software....


There is this persistent feeling that if one clearly and deeply understands exactly what needs to be fed to the NS input, the NS itself
The NS itself might not turn out to be very necessary. - Imho of course.
 
klot:
njel:
what to give as input and what to give as output depends on the activation function
often if the function is a hyperbolic tangent, then the inputs are normalized to -1...1 or 0...1
but who in neurosolutions compiled the dll?


In principle, what to give as input and what to get as output does not depend on the activation function. The only rule to be respected is that the input is dependent on the scope of the activation function. I.e. if you use a hypertangent, it is {+1;-1} if it is an arctangent, it is {-1.57,+1,57}, etc. ..... Be sure to scale (normalize) the data in the range of visibility of the activation function. Otherwise it will turn out to be nonsense!

Maths for a fifth grader!

)) Thanks for the clarification.
 
Aleku:
There is a strong feeling that if you understand clearly and deeply what exactly needs to be fed into the NS and for what purpose, the NS itself
the NS itself might not be very necessary. - Imho, of course.
Exactly what he said. It helps a lot (thinking over the essence of a task and preparation of data for NS inputs) to "order" first of all my thoughts and my delirious ideas, and I can part with many of them.
In general, there are benefits.
 
VBAG:
Aleku:
There is this persistent feeling that if you clearly and deeply understand what exactly needs to be fed to the NS input, the
NS itself may turn out not to be very necessary. - Imho, of course.
Exactly what he said. It helps a lot (thinking about the essence of the task and preparing data for the NS inputs) to "order" your thoughts and your delusional ideas, and it also helps you get rid of many of them.
In general, there are benefits.

It is rather important to decide on the way out. I am so far lame with this.
 
KimIV:
YuraZ:
You can't just give 0.1 at the input, the probability signal always lies in the range of the probability network
Analog to digital converters are just for 0 1.


Igor, I understand that the output should be simpler, but not in all situations!

I say again, if you teach a network... a multiplication table in the range of 0 - 10 on the input - then the output is no longer 0 and 1

The output is a specific number in the range of 0 - 100. The network can't just say yes or no - it can say I don't know.

If you apply it to forex, the output of the network should not be 0 or 1

the output should be at least something like

1-sell

2-buy

3 do nothing

3 variants are simplified and self-sufficient and again only for swing TS

another option on the output

1-sell

2-hold the sale - trend DOWN

3-close sale

4-buy

5-hold to buy

6-Close Buy

in forex, the network has two exits - can be ... but i guess it's hard to have only two options - the third one is very necessary = do nothing!

 
Aleku:

There's a strong feeling that if you clearly and deeply understand what exactly needs to be fed to the NS input, the
The NS itself might not be very necessary. - Imho, of course.


There is probably only one benefit here - a well-designed network is capable of adapting to a changing market

The creator understands what he's giving the input... but it must be difficult to perform day and night + psychology

And in the version that we see at the championship it's just not realistic to perform by hand!

 
YuraZ:

There's probably only one benefit - a well-designed network is able to adapt to a changing market

the creator understands what they're feeding in... but to execute day and night is probably difficult + psychology

And with the version that we see at the championship, it's impossible to hand-execute!


"Native" tasks for NS are tasks of classification, recognition. Accordingly, it is logical to use NS to
to classify the current situation in the market at the time horizon of interest ( with or without,
But situations can also be classified using large arrays of conditional operators having previously selected and analyzed those or other typical situations on a chart using our own natural NS.

The Expert Advisor has no psychological worry, it trades around the clock, it can perform extensive analytical work in all TFs and on all pairs simultaneously in a fraction of a second, it can trade on all pairs at once - well, these are standard advantages of the automated trading.

The Championship leader has the system of inertial type, or it may be called weather vane - where the wind blows, I trade.
I go wherever the wind blows. I.e. trades are opened in the direction of the current trend on a certain TF and held until
until the trend begins to change. It is natural that in such a system it is convenient to use NS for early identification of the current trend change, especially if one is familiar with this tool. Judging by the time of trade holding the trends stand out on H1, H4. By the way, the period from the middle of September, 07 to the end of November, 07 is surprisingly propitious for such systems on TF H1, there was not such a propitious period for them for several years - therefore one should not be blinded by the results of the Championship leader.