Machine learning in trading: theory, models, practice and algo-trading - page 2698

[Maxim Dmitrievsky]

Maxim Kuznetsov #:

don't forget to add the real time...or you'll end up like everyone else :-)

a la 2 pcs: y=abs(sin(x))*sin(x) ; with a frequency of 1 day and 1 week ; the phase shift is better calculated in advance

because the probabilities of indicators and line crossings depend on them.

it was by the way about the harmful, hated here Fourier :-)

Bad way, it is better to use Van Hot encoding or radial functions, and it doesn't give much when this sign is one of several.

It doesn't add or remove anything.

at least that's how it worked for me.

and this is because any oscillating signs fluctuate differently at different times due to heteroscedasticity (volatility), that's why they are already taken into account.

https://developer.nvidia.com/blog/three-approaches-to-encoding-time-information-as-features-for-ml-models/

[Maxim Kuznetsov]

elibrarius #:
Sine together with cosine must be supplied as 2 fics. Otherwise 0,5 and dr will occur 2 times per revolution, like 2 identical times...
Or you can just have day number and hour number. No difference. They are equally well memorised.

Day number/hour number doesn't look good either - there will be a big 23-0 "gap" periodically.

then, to avoid repetitions, add another sign a la "before/after noon" (the sign of the derivative of the sine wave) and leave sin^2 to clock the time (and at the same time scale the signals).

Or as namesake advises. In my opinion excessive.

(cycles are fakapy on large TFs, but on small day/week they are just there, they cannot be thrown out and not taken into account, they are "carrier").

[Uladzimir Izerski]

mytarmailS #:
0) yes I am...)

1) I haven't deployed the whole thing yet,

1. there are problems with the curse of dimensionality and combinatorial explosion, but this is solvable in theory, in favour of accuracy....

2. There is a problem with the fact that the search algorithm is slow, a lot of things need to be written in C or C++, and I don't know how to do it.

3. Even an optimised algorithm will not be able to search for patterns in a large date, we need to search for patterns locally.....

But in general, if it doesn't work, nothing works...

2) Yes.

By the way, you can replace the word "event" with the word "rule".

There is no precision in the markets.

There is only probability with error).

[Forester]

Maxim Kuznetsov #:

The day number/hour number doesn't look good either - periodically there will be a big "gap" 23-0

then, to avoid repetitions, add another sign a la "before/after noon" (the sign of the derivative of the sine wave) and leave sin^2 to clock the time (and at the same time scale the signals).

Or as namesake advises. In my opinion excessive.

(cycles are fakapy on large TFs, but on small day/week they are just there, they cannot be thrown out and not taken into account, they are "carrier").

With the square of the sine you will get 4 times per turn 0.5.

[Maxim Kuznetsov]

elibrarius #:
With the square of the sine, you get 4 times 0.5 per revolution.

see above (all), I said "taking into account the sign" - sin(x)*abs(sin(x)).

[Forester]

Maxim Kuznetsov #:

see above (all), I said "taking into account the sign" - sin(x)*abs(sin(x))

"It's a great feature.)

Draw a graph of your invention.

[mytarmailS]

Uladzimir Izerski #:

There is no precision in the markets.

There is only probability with a margin of error.)

You don't understand what I'm talking about...

1. there are problems with the curse of dimensionality and combinatorial explosion, but this is solvable in theory, in favour of accuracy ...

read what the curse of dimensionality and combinatorial explosion are, wiki will help...

It's solvable in favour of accuracy. - It means that you can deal with the above problems, but the accuracy will suffer, i.e. it will be an approximation of the solution, not a solution.

To make it even simpler, let's say you have 10,000 features that you look at, it takes a long time to find patterns for all of them, there are a lot of combinations (curse of dimension ality).

You can reduce the dimensionality of these 10 000 features to 2-5 features, but with a loss of accuracy, but you can work with it.

Now I hope it's clear what kind of accuracy we're talking about?

[Maxim Kuznetsov]

elibrarius #:

"great" feature)

Draw a graph of your invention.

and what ? it is like that... if you don't use NN, DL, that's how it's traded.

Do you see anything familiar ?

[Forester]

Maxim Kuznetsov #:

And what ? It is so and so in fact...if you don't go into the NN, DL, that's what it's traded at.

Do you see anything familiar?

I see something familiar, already 3-4 times in your posts.
2 times at 0.5 per turn.))))))

[Aleksey Nikolayev]

Cyclic time (hour number, etc.) is easy to use, for example, in KNN, if the metric is written correctly. Or in some developments of this method like local regression.