Machine learning in trading: theory, practice, trading and more - page 1332

elibrarius  
Maxim Dmitrievsky:

What do you mean something better? that it takes 10 hours to train because mikhailo trained it on datasets of 100 points?

And where are the proven results?

I'm just saying that if standard means and approaches fail to achieve the goal, these M.O. perversions will certainly not lead to anything good.

That's exactly why I didn't study it in detail.

And about the fact that someone uses it - you said (whether about toxic, or about visard).

And about using only standard tools, I still don't agree. Otherwise science would have stood still and we would still be sitting in caves and studying location of embers in the fire, predicting fatness of mammoth tomorrow.)

Alexei's results look good. We just need to check on the account/signal for a year. Unfortunately he has such a time frame... with drawdowns for half a year. If I were him, I'd look into something less long term. Even a one week drawdown would make me want to fix something.

elibrarius  

And so the argument is about nothing. Someone likes to invent something, he invents. Who doesn't like it, doesn't invent.

It's like in science - you can set up 1000 experiments of which 999 will be unsuccessful and only 1 will lead to the discovery of something new and useful. But 999 unsuccessful experiments will of course take a lot of your time. So it is better to do it when you are well provided for (salary/grant/sponsor/own funds). In our field would also be a supercomputer, so that not 10 hours, but 10 minutes of calculation took.

Maxim Dmitrievsky  
elibrarius:

That's exactly why I didn't study it in detail.

And about the fact that someone uses it - you said (either about toxic, or about wizard).

And about using only standard tools, I still do not agree. Otherwise science would have stood still and we would still be sitting in caves and studying the location of embers in the fire, predicting the fatness of tomorrow's mammoth)))

Alexei's results look good. We just need to check on the account/signal for a year. Unfortunately he has such a time frame... with drawdowns for half a year. If I were him, I'd look into something less long term. Even a one week drawdown would make me want to fix something.

What I'm saying is that the results are predictable and based on theory.

you don't have to do a lot of unnecessary work, although it may be useful

We are just talking about different things and examples are beside the point.

Aleksey Vyazmikin  
Maxim, I swapped the sample places - for training and validation, the test left - what will be the result based on scientific dogma? I myself do not know yet, the processing is not yet complete.
Maxim Dmitrievsky  
Aleksey Vyazmikin:
Maxim, I swapped the sample places - for training and validation, the test left - what will be the result based on scientific dogma? I myself do not know yet, the processing is not yet complete.

None, it won't do anything.

And what was expected in general?
Maxim Dmitrievsky  
Aleksey Vyazmikin:

I expect that at least the models will not be similar to those before the swap of the sampling places.

So they won't look alike, so what then? )

Aleksey Vyazmikin  
Maxim Dmitrievsky:

So they won't look alike, so what then? )

This will be good, because then you can use them in parallel.

Maxim Dmitrievsky  
Aleksey Vyazmikin:

This will be good, because they can be used in parallel then.

Yes, we are shooting in the sky, i.e. the optimal solution has not been found.

More seed change and a bunch of garbage
Vladimir Perervenko  
Maxim Dmitrievsky:

it looks like we need to pump R instead of python, Renat wrote that soon there will be a direct link without crutches

i.e. catbust will be able to run in 1 line from mt5

I must have missed that message. Do you know when it was or the link?

Good luck