Machine learning in trading: theory, models, practice and algo-trading - page 2320
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 tried to understand how ROCKET works. That is, they generate random kernels, while they do not overlap in frequency (do not correlate). Why not just take wavelets or Fourier, what is the trick?
The trick is that datasigners don't know about csos, that's why they create already a long time ago
I tried to understand how ROCKET works. That is, they generate random kernels, while they do not overlap in frequency (do not correlate). Why not just take wavelets or Fourier, what is the trick?
I don't know what wavelets are, but convolutions worked well in NS, so they were transferred to such an algorithm, which also works well.
there are also similar algorithms on shaplets, but this one seems to be better.
there are others and a comparison
http://timeseriesclassification.com/algorithm.php
the more right the better. It makes sense to port the response part to mql, for tests. I wanted to do it, but got busy with something else.
The point is that datascientists don't know about DSP, that's why they create things that have already been created for a long time
you are so smart, but you don't know anything about DSP or data science)
you're so smart, but you don't know anything about DSP or data scientist))
Yes, that's right))
Listen, is it possible to create an algorithm that will on the fly "remodulate" the price in a "stable" form, for example, the input price, and the output sum of sinusoids but sinusoids all with the same frequencies and phases (each with its own), we will get a series with stable characteristics!
I don't know what wavelets are, but convolutions worked well in NS, so they were transferred to this algorithm, which also works well
there are also similar algorithms on shaplets, but this one seems to be better.
there are others and a comparison
http://timeseriesclassification.com/algorithm.php
the more right the better. It makes sense to port the response part to mql, for tests. I wanted to do it, but I have to do something else.
Listen, can you create an algorithm that will on the fly "remodulate" the price in a "stable" form, such as the input price, and the output of the sum of sinusoids but sinusoids all with the same frequencies and phases (each has its own), we get a series with stable characteristics!
Like here, only on sine waves? That's a good idea.
Like here, only on sine waves? It's supposed to be possible.
No, not that at all...
The market is not stationary, algorithms are not trained on it, they die immediately at birth, what they learned in the past will never be repeated in the future...
What if we try to make it stationary.
1) choose "k" major harmonics on the fly and take them as the market model
2) but these harmonics will also "float" over time in frequency, phase, and amplitude
3) we have to find out how to tune them permanently, so each harmonic always has the same frequency, amplitude and phase
If we obtain it, we will get a "market model" made of the sum of sinusoids which is convenient for studying, and the patterns are always repeating because the harmonics are always in the same diapasons