How to form the input values for the NS correctly. - page 15
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
May I ask where this excerpt came from? I once tried to make a useful signal extraction from noise, but the work was never completed.
It's from Heikin's Neural Networks.
Mutual information is proposed as a target function. So it is a variant of learning without a teacher.
And this is what the end result will be? Some kind of sliding vector, i.e. multidimensional muving?
That is the intrigue. What can we expect from the input of such a trained network of prices of daily bar closes?
Probably get a vector of next possible values... is it like that.
P.S. Candid, what's going on with the unbeknownst internet resource? - Can't access the site since this morning.
That's the intrigue after all. What can we expect from the input of such a trained network of daily bar close prices?
Probably get a vector of next possible values... is it like that.
P.S. Candid, what's going on with the unbeknownst internet resource? - Can't access the site since this morning.
I have the same problem. No additional information available.
I suspect a redrawing muving :). It would certainly be interesting to see.
I have the same problem. No additional information available.
And what prevents you from working on opening prices on a fully formed candle?
I suspect a redrawing muving :). It would certainly be interesting to see.
Are you serious? After all, the muving has a noticeable positive autocorrelation between samples in the first difference series (it is a smooth function), i.e. we are talking about correlation of NS outputs and thus NOT entropy maximisation, and this contradicts the basic concept of NS training.
Or have I... ... read it?
And what prevents you from working at opening prices on a fully formed candle?
Well, nothing.
Why?
Are you serious? After all, the muving has a noticeable positive autocorrelation between samples in the first difference series (it is a smooth function), i.e. we are talking about correlation of NS outputs and thus NOT entropy maximisation, and this contradicts the basic concept of NS training.
Or have I... read it already?
Yes, nothing as a matter of fact.
Ah, why?
So that there is no redrawing muving :)
Are you serious? After all, a muving has a noticeable positive autocorrelation between samples in the first difference series (it is a smooth function), i.e. we are talking about correlation of NS outputs and therefore NOT entropy maximization, and this contradicts the basic concept of NS training
Muvings in my understanding - any characteristic calculated by a sliding window and associated with the last bar of this window :). Positive autocorrelation does not automatically follow from this.
By the way, is maximising mutual information equivalent to maximising entropy?
Good afternoon everyone.
The discussion seems to have shifted from how to form the inputs correctly to what to supply to the inputs.
Such topics on forums usually have a hard and short-lived fate. ((
Even if there are many suggestions for inputs, putting them together will not get a good result imho.
The easiest way to get a result is to take a trading idea and implement it using a neural network.
I.e. in this case it is not a difficult task of inventing TS by a neuronet (and from those inputs that came to mind or someone advised me),
but a simpler task to implement/improve a ready-made system/idea.
__________
For example, if we want to trade on market fluctuations, the inputs and the teacher can be based on МА, Fourier and other indicators that see the market as an oscillatory system.
If we want to create a trend system, our inputs should be indicators that give maximum information about the trend, its emergence, decay, strength, etc.
And the teacher for it should be the indicator which marks the flat, trend, correction and reversal.
If we want to make a system that will trade on breakout/breakout of important levels, it will probably include some zigzags, maybe a market profile, etc.
The teacher for it will probably be a zigzag and not the MA.
Same goes for channel trading systems, systems that use regression, etc. etc.
___________
Therefore I suggest that in order not to invent a vinaigrette of entries and teachers,
take some known good idea/system, match it with inputs and teachers, and build a network.
And see what happens.
___________
ZZZ. I see another approach to the problem of what to feed as solving the problem of selecting meaningful inputs (but again, for a particular teacher) from all the possible ones that come to mind.
Good afternoon everyone.
The discussion seems to have shifted from how to form the inputs correctly to what to supply to the inputs.
Such topics on forums usually have a hard and short-lived fate. ((
Even if there are many suggestions for inputs, putting them together will not get a good result imho.
The easiest way to get a result is to take a trading idea and implement it using a neural network.
I.e. in this case it is not a difficult task of inventing TS by a neuronet (and from those inputs that came to mind or someone advised me),
but a simpler task to implement/improve a ready system/idea.
I agree.
And what is more, not every common TS, crossing TS with NS may have really good results.
Just selecting market entries without conventional TS may go very far... and come to nothing...
Show me...
HERE'S A NICE INDICATOR.
this indicator is a beauty!
and based on ZZ principles - takes H and L
It perfectly shows the pivot points in the past
Nah, zigzag is not good, you got anything else?
---
>> Question:
what's wrong with feeding it as a teaching point?
what is better to feed in for learning - not a pivot point ?
i understand of course i can feed anything as long as the output is interesting to receive
---
p.s.
I'm talking about the points that are fed into the input as a teach
What is the best input for training - not a pivot point?
I understand, of course, that you can feed anything as long as the output is something interesting to get
You could try an entry point for your strategy.