Machine learning in trading: theory, models, practice and algo-trading - page 3373

 
mytarmailS #:
What does it mean :runs faster

On the same data the result gives earlier, and significantly. Yandex advertisers are very fond of posting comparisons.

 
СанСаныч Фоменко #:

On the same data, the result is earlier, and significantly so. Advertisers from Yandex are very fond of posting comparisons.

What does it mean : On the same data the result gives earlier ??? it can be interpreted in any way you want, and this is the second time I've asked you.

You formulate your thoughts like athird-grader, you don't understand a thing.


Do you want the speed of learning?

Do you need speed of prediction?


If the former, then why bother with boosts at all?

 
mytarmailS #:

What does it mean : On the same data the result is earlier ??? it can be interpreted in any way, and this is the second time I have asked this question.

You formulate your thoughts like athird-grader, you don't understand a damn thing...


Do you want the speed of learning?

Do you need speed of prediction?


If the former, why bother with boosts at all?

To keep you in control, I'll roll back: you ask questions like a first-grader: you need to predict ONE value - who cares about speed? But training is another matter, especially if it is retrained at every step.

The working day is over. That's it.

 
СанСаныч Фоменко #:

To keep you in control, I'll roll back: you ask questions like a first grader: you have to predict ONE value - who cares about speed? But learning is another matter, especially if it is retrained at every step.

The working day is over. That's it.

Well, it's only in your head everyone needs what you need and no one needs what you don't need, in reality it's different.

If you need to learn quickly, what are the boosts? You could start training neural nets with minibatches.

Like this, two pages of light boosts, but in fact he needs to learn quickly, and all because he can not formulate the task, even for himself.

 
СанСаныч Фоменко #:

On the same data, the result is earlier, and significantly so. Advertisers from Yandex are very fond of posting comparisons.

Trust the Flow, enter a resource state, and everything will work out.

Lgbt is a bit faster to learn, catbust is faster to execute. Catbust is the strongest out of the box. One can be converted to the other. The learning speed depends on the depth of the trees, the number of trees, and the gradient step. And some other things you can't get into or it will kill you. You can trim and retrain, not necessarily from scratch.
 
I wonder, does anyone read this endless stream of articles called "neural networks are easy"?
It seems to me that if you calculate the average reading time of this nonsense, it won't exceed 10-15 seconds.


Psst. The guy's just trying to make money to buy a car.
 
СанСаныч Фоменко #:

To keep you in control, I'll roll back: you ask questions like a first grader: you have to predict ONE value - who cares about speed? But learning is another matter, especially if it is retrained at every step.

The working day is over. That's it.

I would, of course, carve every one of those lines in stone.

For posterity.

But I have a vague doubt.

What's the one value?

Like on an X scale? Or a Y scale?

Or Buy, Sell?

P.Z.

No matter how you spin it, all of these values are not constant.

They're all time variables.

P.Z..

Or does Proffesore know better?

The workday is over. That's it.

 
mytarmailS #:
I wonder, does anyone read this endless stream of articles called "neural networks are easy"?
It seems to me that if you calculate the average reading time of this nonsense, it won't exceed 10-15 seconds.


Psst. The guy's just trying to make money to buy a car.

Some time ago there was an article in which it was argued that NS are not good for tabular data, and we are dealing with them, and more suitable for images, parsing text ......

In addition, NS due to their layers, perseptrons are extremely difficult to use.

 
Read then what are tabular data and sequences, and what is the difference. Already in Torah, after Perervenko, who can't tell the difference. But this is no longer surprising.
 
СанСаныч Фоменко #:

Some time ago there was an article in which it was argued that NS are not good enough for tabular data, and we deal with them, and are more suitable for images and text parsing.


It always seemed obvious, though hardly provable formally.

Recently I saw this statement in a SHAD textbook on MO, where it was also said (without confirmation) that modern deep networks work on tabular data no worse than tree bousting.