Machine learning in trading: theory, models, practice and algo-trading - page 2169

[iwelimorn]

Maxim Dmitrievsky:

You do genius things ) I'll start watching on Monday, to my favorites

Thank you! At least something to dilute the "theoretical" reasoning...

[Maxim Dmitrievsky]

welimorn:

If you need it, here is a small script that calculates the relative area under the intersection of the curves of the distribution of the value of a feature for two clusters.

I think this is a good metric for choosing a thinning method, target analysis, features, and who knows what else.

area_overlap is yellow in the picture.

the topic is getting serious )

here's a more complicated (but promising) way to read

https://ml4trading.io/chapter/19

it's about densities and how you can sample features and labels from the posterior using autoencoder. You can also transpose feature spaces, to your liking.

in particular, CVAE (conditinal variational autoencoders) are interesting

[Maxim Dmitrievsky]

+ total decorrelation of features through random convolution kernels.

I want to transfer the response part to mql, so that I can get new features from the trained model, in the terminal

https://pyts.readthedocs.io/en/latest/auto_examples/transformation/plot_rocket.html

front end, in general

[Maxim Dmitrievsky]

After a few experiments so far, the conclusion is:

sampling from generative model and then training on those samples - very good (but so far I used only GMM (gaussian mixture model), not autoencoder)

feature decorrelation via ROCKET is ok, there are improvements in generalizability

[Renat Akhtyamov]

Maxim Dmitrievsky:

What do you want to see? Here is the real TC on the MO from my article. The training is only 1 month, then generalization for 2 years. And it's real and it works

show the intersection of MAs and transactions as well as on my screenshot

and why would you show me a tester?

I don't know why I should use a tester to show my trades, show me your trades //demo or real, it doesn't matter.

[Maxim Dmitrievsky]

Renat Akhtyamov:

show the intersection of MAs and deals also, as on my screenshot

and why should I show the tester?

you show your trade //demo or real, it does not matter

first show me yours (demo or real), for a month, a live signal on this service

or don't quack

[kapelmann]

kaban_:
Is there any use of neural networks at all, does it work for trading?

No, it doesn't, the MO only gives more or less "scientific" understanding that in purely price(s) there is almost no information about the future from the change. There are some "traces" of course, crumbs, but they are not predictable, the spread / commission eats everything up and the reason is not in the algorithms but in the data.

In pawn hedge funds there is thousands of times more data (terabytes per day) and they trade HFT, so you can build a bun out of crumbs, and we ***simply, all kinds of new frameworks and ***different from social networks.

[Renat Akhtyamov]

Maxim Dmitrievsky:

yours first show (demo or real), for a month, a live signal on this service

or don't quack

you can't do it on the mash, that's why you don't show it

If you fail, why are you advising people here?

[Maxim Dmitrievsky]

Renat Akhtyamov:

You can't do it on the mash, that's why you don't show it.

If you fail, why are you advising people here?

Are you going to show me? Or are you going to keep on being a sissy?

I advise no one.

You'd better learn Russian.

[Renat Akhtyamov]

Maxim Dmitrievsky:

Are you going to monitor? Or are you going to keep on sneezing?

I don't recommend anything to anyone.

gobble