Machine learning in trading: theory, models, practice and algo-trading - page 2313

[mytarmailS]

elibrarius:
In theory, yes.
But still, what's the point of the action?

You can't argue with that, there's no point...

elibrarius:
I need to get 10 out of 100. Is there a solution?

I don't know how it works in alglib, try to pull info from psa function and see how many components you need to describe 100 prizes.

Or just take the first n columns of the psa and steamroll...

But all this is useless... if you have 50k features you just need Rsa, and if you want to play around without understanding what you're doing you don't need Rsa at all, the result will be worse than without it with 99.999... % probability you got it...

[Maxim Dmitrievsky]

elibrarius:
In theory, yes.
But still, what is the point of the action? If there is no gain in speed, but rather a slowdown, for an extra operation.
It is necessary to get 10 out of 100. Is there a solution?

Take the 1st 10 components of the cov. matrix

[Forester]

Maxim Dmitrievsky:

take the 1st 10 components of the matrix

The matrix will not work - it is 100x100.

And we need to get 10x1000 from it, not 10x10 or 10x100, but by some calculations. That is, for each of the 1000 rows to apply 10 GC

[Maxim Dmitrievsky]

elibrarius:

The matrix will not work - it is 100x100.

And we need it not 10x10 or 10x100, but by some calculations get 10x1000. I.e. for each of 1000 rows apply 10 GC

Each component point is the sum of products of attribute values by their cov. values, as far as I remember

Do the math and compare it to sklearn.

The matrix may turn out to be inverted, then from the end. You should check it additionally there

[Forester]

Maxim Dmitrievsky:

Each component point is the sum of the products of feature values by their cov. values, as far as I remember

Do the math and compare it to sklearn.

The matrix may turn out to be inverted, then from the end. You should check it additionally there

In general, you need cycles like the ones above.

[Maxim Dmitrievsky]

elibrarius:
In general, you need cycles like the ones above.

There s (bestfeatures) is just the number of components which are selected for training, like

the value of each component is multiplied by the value of the attribute and they are added together. And so for each row of the original series.

I have to remember, I have no time. I must read the documentation.

[Forester]

Maxim Dmitrievsky:

There s (bestfeatures) is just the number of components that are selected for training, like

the value of each component is multiplied by the value of the attribute and they are added together. And so for each row of the original series.

I have to remember, I have no time. I need to read the documentation.

Yes, the result is the same as in sklearn. At first I was dealing with the first example code, and there it was at all 100 GCs.
Now everything is fine.

[mytarmailS]

I decided to see what's inside the network, on each layer... I decreased the dimensionality through umap to two components in each layer

A network with three internal layers, almost untrained, only 400 examples... but still fun to look at...

[Vladimir Perervenko]

mytarmailS:

I decided to see what's inside the network, on each layer... I decreased the dimensionality through umap to two components in each layer

A network with three internal layers, almost untrained, only 400 examples... but still fun to look at...

How did you manage that?

What's the output dimension?

[Vladimir Perervenko]

cemal:
Ludwig has deep learning models without having to write code, no programming skills required to teach the model:https://ludwig-ai.github.io/ludwig-docs/

Installed it recently. Haven't gotten around to checking it out yet. They promise miracles.