Machine learning in trading: theory, models, practice and algo-trading - page 2683

 
Maxim Dmitrievsky #:
It's a lot of surprises, to be honest. It's better to transfer everything.

1) I agree, if you want to do production you have to port it.

2) agree that at the stage of mcl research it's a dead end to spend a week/month writing something that is already written in rk or python and can be applied in 2 minutes.

2 minutes vs week/month

 
mytarmailS #:

1) I agree, if you do production you have to move it.

2) agree that at the stage of mcl research it's a dead end to spend a week/month writing something that is already written in rk or python and can be used for 2 minutes.

2 minutes vs week/month.

It depends on experience. At the stage of learning MO, python or R is better to familiarise yourself with the area. When you've found what you need, you can write right away in the platform, especially if the ml engine will be.
 
Maxim Dmitrievsky #:
It depends on experience. At the stage of learning MO python or R is better to get acquainted with the field.

Not from experience, but from common sense, Rca is a language for data analysis, that's why it's the fastest and most convenient to do analysis there (nobody writes websites in C++, not because you can't, but because there is lavascript for that)

Maxim Dmitrievsky #:
When you find what you need, you can write in the platform.

YES

Maxim Dmit rievsky #:
especially if ml engine will be.

NO

will make an engine,

if you want feature selection

feature selection

ifyou want feature engenering

feature engenerating

♪ youwant autoML ♪

make autoML

you want.........

And it will be endless, it will always be behind the trends....

 
mytarmailS #:

Not from experience, but from common sense, Rca is a language for data analysis, that's why it's the fastest and most convenient to do analysis there (nobody writes sites in C++, not because you can't, but because there's lavascript for that).

YES

NO

they're gonna make an engine,

you want feature selection

feature selection

feature engenering

feature engenerating

want autoML

autoML

want.........

And it will be endless, it will always be behind the trends.....

I definitely won't want autoML, it's for nerds 😀
 
Maxim Dmitrievsky #:
I definitely don't want an auto ml, it's for nerds 😀

I mean you had to explain that it was an example, right? )))

 
mytarmailS #:

I mean, you had to explain that it was an example, right? )))

I'm saying that when you have something, it's better to rewrite it for the terminal, without layering.

there are enough tricks with execution and peculiarities of account types, and you have to keep an eye on the correctness of the layering work.
 
Maxim Dmitrievsky #:

I'm saying that when you have something, it's better to rewrite it for the terminal, without layering.

there are enough tricks with execution and peculiarities of account types, and you have to keep track of the correctness of the layering work.

ok ... Never mind...

Listen, you used to use tranformers, how did they show themselves?

 
mytarmailS #:

Okay. let's move on.

Listen, you've been using tranformers, how did they perform?

No, I haven't. Not without good preprocessing. I've used the recurrence ones. Takes a long time to learn, no output. Transformers take even longer to learn
 
Maxim Dmitrievsky #:
I haven't used them... they won't show without good preprocessing. I've used recurrent ones. Takes a long time to learn, no output. Transformers take even longer to learn
Okay.
So CNN should be the first place to get dragoned.
At least for the sake of features.
 
mytarmailS #:
I see.
So CNN should be the first place to get dragoned.
If only for the sake of features.
There is TCN for time series.
Or better yet, just spam chips through convolutional kernels and learn, like, for example, in ROCKET.