How to form the input values for the NS correctly. - page 6
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
I meant this situation (A), but it's a bit different, I think it's better to look for the smallest difference between the entry point and the extremum and also the biggest difference, because there will be points where the price has not fallen below the entry point and vice versa. But the idea is to look for the nearest extremum, something like a zigzag...
Tomorrow I'll post the opening inductor with a glimpse into the future. It clearly shows that trades with TP=80...100 pt last about 1500 minutes, from that we can draw appropriate conclusions for different TFs. But as for finding two extrema for X pips up and X pips down, I don't think so. If we go down and reach X points, we may not reach them upwards. Did I understand you correctly?
Are you going to form the input data at every bar? Or only under some conditions, such as when the muwings cross, it means BUY, and the network at this moment receives the input vector and decides itself whether to enter or not, right?
Yes it is certainly an interesting question but probably not so.
I would like not to miss a single bar, to have a large number of samples (especially as the output is continuous, not discrete, and as I already said hopefully there will be a normal ratio of repeatability and inconsistency), but with the option of approximation with cutting the number of samples by some indicator I never thought about.
Yes it is certainly an interesting question but probably not so.
I would like not to miss a single bar, to have a large number of samples (especially as the output is continuous, not discrete and as I have already said hopefully there will be a normal repeatability and inconsistency ratio), but with the option of approximation with cutting down the number of samples by some indicator I have never thought about.
Now keep in mind, could be very useful! I've heard from many people that there should be a sane TS, and NS should only improve it a little... But how much these words are true ?????? f*ck knows...
2 StatBars Thank you very much for the articles.
What about non-normalized inputs? Can sigmoid be used or are other functions required?
I tried to find a universal non-linear function.
The disadvantage of sigmoid is that it has a limited range of values.
That's what I managed to get.
I am posting it here.
Comparison with sigmoid:
1. Convergence. Depends on problem and input data.
With input data |x| < 1 the sigmoid one naturally wins due to higher nonlinearity.
But it can be problematic to keep track of this with large layer sizes.
At |x| > 1 the sigmoid blows by convergence. True, not always.
Also, it removes the "freeze" effect where the output of the sigmoid tends to the boundary of the value area, as there is no boundary of the value area.
2. In most tasks, the need for data preprocessing disappears for the same reason. However, some value limits remain. It is not a good idea to feed knowingly large values of order 2 and higher.
3. It becomes possible to use a function on the output layer for the same reason
4. (-) There is no possibility to express f'(x) = g(f(x)) nicely, but this is immaterial. All the more so, this is not a speed-critical part of the program.
The function can be modified to get more non-linearity, but I am happy with this form.
Yes, by the way, it's good that you brought this up. I keep questioning whether it would be more correct (in your experience) to ration one sample by itself or in general on all samples?
Decided to rename the branch.
All at once, of course.
If normalised with respect to the first value, the first value will always be zero. Another option is to normalise with respect to the middle of the sample range. If different samples have different elements equal to zero, then there is no problem, zero is also a value.
EE, you are messing with something.
We have an input sample. For simplicity 1 input.
Looking for maximum and minimum on the ALL sample.
According to the values found, reduce the entire sample to an interval of -1 to 1
Eh, you're making a mess of things.
We have an input sample. For simplicity, we have 1 input.
We are looking for maximum and minimum on ALL of the sample.
According to values found we reduce the whole sample to an interval from -1 to 1
Why to [1;-1] ?
Eh, you're making a mess of things.
We have an input sample. For simplicity, we have 1 input.
We are looking for maximum and minimum on ALL of the sample.
According to the values found we bring the entire sample to an interval of -1 to 1
And if in the future, highs and lows are updated?
Why to [1;-1] ?
Others may be, it is not the benchmark, the most commonly used interval.