Why is the normal distribution not normal? - page 46

New comment
 
Urain:

I've heard many times about thick tails of distribution, but I don't understand what the point is, I made an indicator that outputs bar size distribution (based on Close[i]-Close[i+1] difference) into separates, can someone explain why distribution is narrower than normal?

The benchmark is the red line distribution yellow histogram.

and the indicator that was used to build it. Original title (Distribution_Histogram_&_norm_test)


Can you modify it to fit +/- Close difference?

And really, parameters for normal distribution should be calculated based on histogram. And so you just adjusted the height? )

 
TVA_11:


Can you make a modification to account for the +/- difference Close?

And really, the parameters for a normal distribution need to be calculated from the histogram. And so you just adjusted the height? )

If you mean + bars and - bars allocation separately, then it is done so in the indicator. For the relative distribution it doesn't matter, but for the absolute one it's a problem, I have to change the code and shift indicator's buffer backwards to compensate the previous shift (the original shift cannot be removed since array indices cannot be negative).
 
Urain:
If you mean + bars and - bars separately then it's done so in the indicator. Since the MO tends to zero it is logical to display the distribution from -free to +free, but visualization of -free is not convenient therefore we shifted it to the left so that the significant part of the -distribution goes to the area greater than zero. It doesn't matter for relative distribution, but it is a problem for absolute distribution, I have to change the code and add shift of indicator's buffer backwards to compensate previous shift (initial shift cannot be removed since array indices cannot be negative).

In my indicator there is no such problem - there is no need to shift the bars. Histograms will always be at the end of the chart at any value from -infinity to +infinity.

By the way, I corrected the indicator. Both bar sizes and shifts were transformed by the same parameters. Now it is correct - by individual parameters as in indicator settings.


So any suggestions on my question, gentlemen mathematicians?

 

joo:

My indicator does not have this problem - there is no need to move the bars

...

So any suggestions on my question, gentlemen of mathematics?

Even a fool can move bars. Where's the money?
 
Reshetov:
Even a fool can move bars. Where's the money?
Do you need to be paid to help me?
 

joo, I experimented with this approach (only I used a hyperbolic tangent rather than a sigmoid).

Nothing interesting came out.

 
lea:

joo, I experimented with this approach (only I used a hyperbolic tangent rather than a sigmoid).

Nothing interesting came out.

Are you sure you know why I need it? If you know how to "straighten" the distribution, help me. How is the hyperbolic tangent (which, by the way, has 4 times the degree, while sigmoid has only one, which is more preferable in terms of saving system resources) better than sigmoid?
 
joo:
Are you sure you know why I need it? If you know how to "straighten" the distribution, help me. How the hyperbolic tangent (which, by the way, has 4 times its degree, and in sigmoid once, which is more preferable in terms of system resources) is better than sigmoid?

There is almost no difference if the limit is -1 to 1, and the tangent is slower if the limit is 0 to 1.

double sigma(double d)// от 0 до 1
{return( 1.0/(1.0+MathExp(-d)) );}

double tanh(double d)// от -1 до 1
{ double D=MathExp(-d); return( (1.0-D)/(1.0+D) );}

So if you cast hypertangent to the form [0;1] then it's two additional operations *0.5 and +1,

The same two operations *2 and -1 are required for sigma when converting it to the form [-1;1].


sigma has 3 operations a hypertangent has 5, so if you add one of the 2 operations to one of the functions you get either 5;5 or 3;7

 
joo:
If you know how to "straighten" the distribution - help me. How is a hyperbolic tangent (by the way, it has 4 times the powers, while in sigmoid one time, which is more preferable in terms of saving system resources) better than sigmoid?

My task involved sliding window processing (i.e. there were two parameters - window length and coefficient with tanh argument). If this is suitable for your task - I can send you a code snippet.

I used tanh, because it was more convenient for me (I needed result series with zero average). In general, you can use tables to calculate such functions.

 
lea:

If it suits your needs, I can send you a code snippet.

Yeah.
1...394041424344454647
New comment
Reason: