NS + indicators. Experiment. - page 6

 

To Prival Thanks for your help!!! But, I don't intend to recognize tanks, thank God, the task is much easier...

alexx 22.12.2007 11:45

Understood, thanks. And your example looked at. It turns out the normalization to the form -1 +1. I'll try to experiment with your version.

I forgot to ask one more question. I understood that you are using NeuroShell DayTrader. Why are you not satisfied with NeuroShell2? I'm asking because I have NS2 version 4.0, and I'm not interested in other similar packages. May be I am mistaken? What do you personally like about DayTrader?

NeuroShell2 is a wonderful program, but it has to be implemented in MT4. It is not hard, but it is very expensive. In Neuroschell Day Trader all studies and hypothesis tests can be done in 2-3 days. And you can check everything in real time in visual testing mode - a unique feature! You can create any NS and immediately check it.

nen 22.12.2007 13:54

klot, here's a suggestion. You only use price differences, i.e. you only consider price value. Try to use timeframes as well. Most indicators only work with price. And not only indicators, but most traders use price changes in the market. But price is very much related to time. Available versions of indicators for pattern search (not only Gartley) also consider mainly only the price. In relation to ZZ, we can propose to use the parameter Number of bars, during which the ZZ beam was built.

It is possible to use the Fibo Time tool. I will try to show charts using Fibo Time on Onyx in the nearest future. Closer to the New Year or after the New Year.

I am testing all the options, I think I will find the best one by the next championship. Come and see me on the forum...

 
Mathemat:
Prival, I remember your first post. The problem is that the confidence interval (plus or minus 3 s.c.o.) only makes sense in the Gaussian approximation. Then the vast majority of possible outcomes will be within it (0.997). And if 0.7, there will be too many errors. And the most important problem is in the estimation of the m.o. at the current moment.

If the probability is 0.88(9), it works for any PZ and at k=3 for any PZ that has both PZ and PZ covers the value with probability 0.88(9). There is a proof below the branch.
 
klot:

To Prival Thanks for your help!!! But, I'm not going to recognise tanks, thank God, the task is much easier...


And I think it was easier with tanks :-). I have already coped with them, but the forex could not. It would be very interesting to find out how the NS will work if we apply the AMA derivatives.
 
Prival:
klot:

To Prival Thanks for your help!!! But, I'm not going to recognise tanks, thank God, the task is much easier...


I think it was easier with tanks :-). They have already been dealt with, but the forex is not working out. It would be very interesting to know how the NS will work if we submit derivatives of the AMA.
It may be several. Only on the output of what, also a derivative of AMA?...or several.....
 
njel:
Prival:
klot:

To Prival Thanks for your help!!! But, I'm not going to recognize tanks, thank God, the task is much easier...


I think it was easier with the tanks :-). I have already mastered them, but I cannot do forex. It would be very interesting to find out how the NS would work if we submitted the AMA derivatives.
It may be several. Only on the output of what, also a derivative of AMA?... or several....


The output of the algorithm I think the recognition should not be four action Short, Close Short,Long,Close Long. But something like this.

  1. Straight-line Up movement (angle of slope, speed, acceleration and possible time of end of this movement).
  2. Oscillations relative to horizon (its frequency, amplitude and phase,
  3. possible
  4. dwell time)
  5. Oscillations and its parameters relative to first two movements
  6. .

I.e. with respect to the trajectory being analysed, many alternative hypotheses are put forward for its possible motion. This is what I mean by the classes that need to be recognised, between these classes of motion there is a zone of uncertainty. I.e. until one of the hypotheses is chosen by some criterion, I do not know what to do. And when the decision has been made (the enemy's behaviour has been recognised). The algorithm of analysis comes into action - when to shoot, where to shoot and whether it is necessary to shoot now. (Replace Buy, Sell, TP where necessary.)

 
Prival: The output of the algorithm, I think, should not be the four actions Short, Close Short,Long, Close Long. And something like this
  1. Straight-line Upwards motion (inclination angle, speed, acceleration and possible time of termination of this motion)
  2. Straight-line Downwards motion (inclination angle, speed, acceleration and possible time of termination of this motion)
  3. Oscillation relative to the horizon (its frequency amplitude and phase, possible lifetime)
  4. Oscillation and its parameters relative to the first two motions
  5. .

I.e. with respect to the trajectory being analysed, many alternative hypotheses are put forward for its possible motion. This is what I mean by the classes that need to be recognised, between these classes of motion there is a zone of uncertainty. I.e. until one of the hypotheses is chosen according to some criterion, I do not know what to do. And when the decision is made (the enemy's behaviour is recognised), the algorithm for analysing when to shoot, where to shoot and whether it is necessary to shoot now enters into action. (Shoot - where necessary, replace with Buy, Sell, TP).

And how can a neural network know so much and predict so much?
 
LeoV:

And how can a neural network know so much and foresee so much?

If you create a forex-specific algorithm, you can achieve this. But if you use a ready-made software product (NS, etc.), I don't know. If you want the algorithm to have such outputs, then create it. While using an algorithm, even wonderfully advertised (even NS or any other) without knowing and not understanding how it works (black box) IHMO useless, a waste of time.

Many do it, they just look through indicators on the input. The main thing is the output, once you know it, it becomes more clear what to put on the input. But for many traders it's still a black box. For instance, I would never trust a black box to trade with my money. That's why I am so sceptical about NS.

 
Prival:

If you create a forex-specific algorithm, you can achieve this. But if you use a ready-made software product (NS, etc.), I don't know. If you want the algorithm to have such outputs, then create it. While using an algorithm, even wonderfully advertised (even NS or any other) without knowing and not understanding how it works (black box) IHMO useless, a waste of time.

Many do it, they just look through indicators on the input. The main thing is the output, once you know it, it becomes more clear what to put on the input. But for many traders it's still a black box. For instance, I would never trust a black box to trade with my money. That's why I am so sceptical about NS.


So the NS is essentially a black box. In the sense of making a decision. It is impossible to logically understand why a neural network made that particular decision. That is why it is a neural network. If it were possible to "predict" and understand a neuronet - then it would be possible to "predict" and understand a person - why he makes this or that decision. But, unfortunately, it is impossible to do it.......
 
Prival: That's why I'm so sceptical about NS
And I'm not doing neural networks at the behest of Batter, I've been doing it for a long time. And I think this direction is very promising.....
 
LeoV:
Prival:That's why I'm so sceptical about the NS
I don't do neural networks at the behest of Batter, I've been doing it for a long time now. And I think this direction is very promising.....

The fact that it is promising and interesting with this I agree. But you are wrong about the fact that it is impossible to logically understand "why a neuronet made exactly such a decision". The logic of decision making can and must be programmed (this is how it is done in normal algorithms).