Interpolation, approximation and the like (alglib package) - page 2
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
Exactly interpolation? Are you sure? Not approximation? And it's not redrawable?
You'll be interpolating every tick.
If you need interpolation over intermediate nodes (ZigZag nodes for example) without redrawing, then the whole point is where the next node will be.
You can only create a non redrawable clear ZigZag if you have a time machine. There is no way you can determine without a time machine that the current bar is an extremum.
There is someone on the forum periodically who I call a "ponytails redrawer".
The whole point is the ponytail.
It is a classic of this genre - to shift the SMA to the left by half a period and finish drawing those half-periods by a polynomial of some degree. Here's an example -https://www.mql5.com/ru/forum/224374. Surely this has been seen before.
One can use splines to make a very nice interpolation along zigzag extrema but one should clearly understand that between the last two or three knots there will be redrawing. There is no way without it!
If not redrawn, it's not interpolation, but what I call a trace from the approximating line (not interpolation!).
Apart from polynomials, I don't see anything comprehensible so far.
Here's a specially recorded a gif to demonstrate an example of polynomial of higher power (10), in order to understand how much less "beautiful" it is than I would like :))
And for the calculation of polynomials of higher powers the accuracy of double is not enough. It will be necessary to use special libraries using types of higher accuracy. But personally I don't see the use of polynomials of degree over 5.
Simply through interpolation, the graph can be changed beyond recognition, while approximation is just a cruder approximation. And there it says in the alglib package about the cubic spline, that you can get an interpolant value from the new data. An approximation is also possible, but it would be the usual regularisation or smoothing of the original data. I need a pretty good transformer of features for MO. There's also multidimensional interpolation inverse distance weighting, which works in multidimensional space and also looks tempting, at first glance... but until you get a feel for it, it's hard to say.
It's just that through interpolation you can change the graph beyond recognition, and approximation is just a coarser approximation. And it says in the alglib package about the cubic spline that you can get an interpolant value on the new data. An approximation is also possible, but it would be the usual regularisation or smoothing of the original data. I need a pretty good transformer of features for MO. There's also multidimensional interpolation inverse distance weighting, which works in multidimensional space and also looks tempting at first glance... but until you get a feel for it, it's hard to say.
Spline will still be redrawn between the last nodes.
Think about it:
We don't know where the next node will be.
Spline will still be redrawn between the last nodes.
Think about it:
We don't know where the next node will be.
As it turns out, yes... but if new value lies in already known interval on normalized data, then we can get the value of the spline. And it makes no difference where the curve goes next.
On the other hand if there will be spikes, it is desirable to have tails at both ends. In splines these are left and right bounds? I'll read some more articles.
It sort of turns out that yes... but if the new value lies in an already known interval on the normalized data, then you can get the value of the spline. And it makes no difference where the curve goes next.
On the other hand if there will be spikes, it is desirable to have tails at both ends. In splines these are left and right bounds? I will read some more articles.
It all works fine when no new points (nodes) are added. And for stock trading that's the point - where a new point will appear.
Of course, these are all great tools to charm the naive public.
But I believe that for a trader in this approximation-interpolation field, only what makes a quality extrapolation forecast can be of value.
This all works fine when no new points (nodes) are added. And for stock trading, that's the point - where a new point will appear.
Of course, these are all great tools to charm the naive public.
But I believe that for a trader in this approximation-interpolation field, only what makes a quality extrapolation forecast can be of value.
Perhaps the task itself is not set correctly, it's kind of creative stuff that needs to be done somehow. I'd be satisfied with multivariate kernel tricks instead of polynomials and splines, but I haven't found them anywhere and I can't write them myself.
A qualitative extrapolation forecast via polynomials on a single BP is of course nonsense too. If even neural networks on multiple features do not work for everyone.Perhaps the task itself is not set correctly, as a creative thing that needs to be done somehow. I would be satisfied with multivariate kernel tricks instead of polynomials and splines, but I haven't found them anywhere
I agree - this stuff deserves to be studied.
In fact I was lying when I said that only extrapolation is important.
The application of various approximation and interpolation methods (to a lesser extent), including multivariate ones, is the mathematical foundation in solving the pattern recognition problem, which is the foundation of AI.
And modern traders without AI will find it harder and harder in the future.
And modern traders without AI in the future will find it increasingly difficult.
It is a myth: traders cannot do without complex calculations
it's a reality.
this is reality
No one has solved it - wrong.
What hasn't anyone solved? The problem of interpolating a function? The problem of interpolating a function - no one has solved such a problem and no one ever will.