How to form the input values for the NS correctly. - page 19

 
TheXpert писал (а) >>
My IMHO, zigzag as inputs to NS is useless, and as information compression too. It shows peaks, but in no way reflects the dynamics in between. Especially since it reacts to almost any spike, so, again, IMHO, screw it.

I might add that by learning on the zig-zag, the network is more likely to simply learn to repeat it with a 1 bar lag. The well-known "today will be like yesterday and tomorrow will be like today" effect. This way it (the network) will achieve the minimum possible error during the learning period. But it will not work in the future.

 
LeoV писал (а) >>

I might add that by learning on the zig-zag, the network is more likely to just learn to repeat it with a 1 bar delay. The well-known "today will be like yesterday and tomorrow will be like today" effect. In this way it will achieve the lowest possible error.

I think that if you get the sampling right with a ZZ, it can be mixed, then there won't be this effect.

I still don't think that RZ is a teacher for NS.

 
StatBars писал (а) >>

I think that if you do the sampling properly with RQ, you can mix it up, then it won't have this effect.

I still don't think it's a teacher for NS.

In theory, time series are not advised to shuffle. So stir it or not, you'll get it all the same.....))))))))))))

 
LeoV писал (а) >>

In theory, it is not advisable to mix time series. So mix it up or not, you'll get it all the same.....))))))))))))

There is set A - vectors Sell, let's say.

There is set B - vectors Buy, and set C - vectors Hold. Values of vectors should be relative.

So I suggested to mix them up.

There is an idea (that was advised by one good man), and LeoV could help me with it.

Let's take the most infuriatingly over-drawn indicator and teach it to produce correct signals. Naturally we should get a result

so that our inputs will be just perfect, or nearly so... These values should be taken as a teacher for the NS. The advantage here is that we feed vectors BUY/SELL which the network itself has chosen as optimal ones. But a set of vectors Hold must be trimmed manually. Just to make sure the sample does not consist of 90 % of Hold vectors and only 5 % on Buy/Sell...

 
StatBars писал (а) >> We take the most relentlessly redrawing indicator in the NS, teach the NS to give the right signals.
There was a similar idea: input - hard real data, output - even the future. I do not see any contradiction in it.
 
StatBars писал (а) >>

There is set A - vectors Sell let's say.

There is set B - vectors Buy, and set C - vectors Hold. The values of vectors must be relative.

So I suggested to mix them up.

There is an idea (that was advised by one good man), and LeoV could help me with it.

Let's take the most infuriatingly over-drawn indicator and teach it to produce correct signals. Naturally, we should get a result

so that our inputs will be just perfect, or nearly so... These values should be taken as a teacher for the NS. The advantage here is that we feed vectors BUY/SELL which the network itself has chosen as optimal ones. But a set of vectors Hold must be trimmed manually. Just to make sure that the sample does not consist of 90 percent of Hold vectors and only 5 percent of Buy/Sell vectors...

Excuse me? We use the input indicator as the output one? Or what? The essence is not clear.

 
LeoV писал (а) >>

Excuse me? Do we then use the input turkey as the output? Or what? The point is not clear.

We feed an indicator (re-drawable) to the input of GS. We obtain entry points (Optimal Buy/Hold/Sell in the Outputs of the MS). And then we describe these points using other indices, normal ones, or something else. But before that we need to trim Hold manually... Is it clear?

 
Mathemat писал (а) >>
There was a similar idea: input - hard real data, output - even the future. I do not see any contradictions in it.

It makes sense, but only if the noise is removed.

 
StatBars писал (а) >>

Input to the NS input is an indicator (redrawable).

What do you mean already redrawn? And what are the reasons to expect that the NS itself (without a teacher) will find good entry points? In other words, how do you imagine the target function of such a NS?

 
lna01 писал (а) >>

You mean already redrawn? And what are the reasons to expect that the NS itself (without a teacher) will find good entry points? In other words, how do you represent the target function of such a NS?

Yes, already redrawn.(SSA, Hodrick(or something like that))

The target function in NS is Optimal Buy/Hold/Sell based on Close. NS will find it without a teacher. Maximize Correct Signals is an error function that is used for learning. Of course I may be confused but I think the sense is clear.