Machine learning in trading: theory, models, practice and algo-trading - page 2628

[mytarmailS]

elibrarius #:

The cross validation (valuing forward) is not explained, what is wrong with it?

Because the signs are floating, it's all adjustment to the track/test, validation will be different

[Forester]

mytarmailS #:
The signs are floating, it's all a fit for the track/test, validation will be different

What kind of fitting is that? It's the opposite. It's a way of weeding out the randomly successful parts of the test.
What kind of validation are you suggesting?

[Maxim Dmitrievsky]

mytarmailS #:

a recommendation came up on Medium on your topic, it may come in handy, I didn't get into it

I was interested in this approach because the trained models can be easily transferred to the terminal (I think)

https://medium.com/@james_laidler/generating-a-rules-based-system-using-iguanas-762843dd1418

[Forester]

Maxim Dmitrievsky #:

a recommendation came up on Medium on your topic, it may come in handy, I didn't get into it

I was interested in this approach because the trained models can be easily transferred to the terminal (I think)

https://medium.com/@james_laidler/generating-a-rules-based-system-using-iguanas-762843dd1418

As I understand it, it looks for good tree branches (or paths to some leaves) using cross validation optimisation. For example, a path like this

(X['Age']>-0.125)&(X['Embarked_C']==True)

And then all these paths are fed as features into new model. But it is not a ready model to be transferred to the terminal, it is a ready-made features for training.

[СанСаныч Фоменко]

mytarmailS #:

The importance of signs in the moving window (indicators and prices)

At one moment the indicator may be 10% important and at another moment it may be 0,05% important, such is the truth of life)

If you think it solves everything, you should be proud of it.

That's what the four signs of Fisher's Iris look like.

Or if you zoom in on the sliding window.

There, finally.

I don't care about the models, take the fastest and easiest, a perfect RF model.

The main problem is in the signs-predictors. The ones you showed are just junk. You should look for ones with predictive power that changes little as the window moves, at least a channel of 10%. If you can't find signs with weak variability, then the target is junk. You should look for another target. And so for years.

PS.

Once posted a similar result on the variability of my traits but in tabular form. There are often those with sd > 100%. But you can find ones with sd < 20% quite quickly.

[Aleksey Nikolayev]

СанСаныч Фоменко #:

There, finally.

Don't care about models, take the fastest and easiest ones, a perfect RF model.

The main problem is in the signs-predictors. The ones you show are just junk. You should look for ones with predictive power that changes little as the window moves, at least a channel of 10%. If you can't find signs with weak variability, then the target is junk. You should look for another target. And so for years.

PS.

Once posted a similar result on the variability of my traits but in tabular form. There are often those with sd > 100%. But one can pretty quickly find ones with sd < 20%.

It is possible to understand the word 'search' in this context in different ways. At least in two ways. One can either try to manually find signs which catch the "physics" of the market, or one can try to construct them by means of MO from the raw prices.

[Maxim Dmitrievsky]

elibrarius #:
As I understand it, good tree branches (or paths to some leaves) are searched for (using optimization with cross validation). For example, a path like this

(X['Age']>-0.125)&(X['Embarked_C']==True)

And then all these paths are fed as features into a new model. But it's not a ready-made model to be transferred to the terminal - it's a ready-made features for training.

It seems to be written as a binary classifier. I'll look at it later, I have an excuse to get into the subject for a while.

[Forester]

СанСаныч Фоменко #:
You have to look for ones that have a weak variability in their predictive power when the window moves, at least a channel of 10%. If you can't find signs with weak variability, then the target is junk. You should look for another target. And so on for years.

Have you found something? Or is it all for naught?

[Flavio Javier Jarabeck]

WHY do we now have THOUSANDS of pages translated into crude Portuguese on our Forum?

It doesn't make sense...

Whose brilliant idea was it?

For those who think I'm exaggerating, just read the thread title, it's bizarre...

[secret]

mytarmailS #:

The importance of signs in the moving window (indicators and prices)

Take Twitter sanction, it's the most useful sign right now.

And forget about forex, it's too efficient. Switch to crypto, for example.