Machine learning in trading: theory, models, practice and algo-trading - page 3546

 
mytarmailS #:

pen test

You made a bot already? You're a monster.

 

Turns out Tinkov Bank has its own lib for BP :)

I didn't find anything special inside

https://etna.tinkoff.ru/

ETNA - Time Series Library
ETNA - Time Series Library
  • etna.tinkoff.ru
ETNA is the first python open source framework of T‑Bank.ru AI Center. It is designed to make working with time series simple, productive, and fun.
 
Maxim Dmitrievsky #:

You made a bot already? Monster.

More like an underbot, just enters and sets stop, but it's enough for me for now.

It is interesting to observe how the signal is formed, etc...

 
mytarmailS #:

It's more of an under-robot, just enters and stops, but that's good enough for me for now.

It is interesting to observe how the signal is formed, etc...

You can also try splines, but it is only for marking on the history.

Your way is more flexible in this respect, although I would still compare errors on the track/test. It should be obviously worse on the test.

Later I will try your method for marking trades and compare it with splines, for example.

In your case it is probably better to use some multiple Bayesian regression. To equalise the errors on the trayne/test.


 
Maxim Dmitrievsky #:

You can also try splines, but that's only for markup on history.

Your way is more flexible in this respect, though I would still compare errors on trace/test. It should be obviously worse on the test.

Later I will try your method for marking trades and compare it with splines, for example.

Try to do exactly as I did, build a candlestick price pattern...

The point here is not in the error and not in the curve fitted to the price, the point is in the study of the model itself and it does not matter whether it has diverged from the price or not.

 
mytarmailS #:

try doing exactly what I did, build a candlestick price pattern...

The point here is not in the error and not in the curve fitted to the price, the point is in the study of the model itself and it does not matter whether it has moved with the price or not.

I'll try it later. For now I'm expanding my arsenal of label markers. Your way works :)

But you should also try to build an unbiased model.
 
Maxim Dmitrievsky #:
But you should also try to build an unbiased model.

I don't understand why, because my signals are based on the model, so you don't have to look at the price at all...


Morning trading with the robot, he opens a position and puts a take, stop, I manage it further...

I only show the beginning of the day, because I'm ashamed of it... I started to cheat...


 
mytarmailS #:

I don't understand why, because my signals are based on the model, so you don't have to look at the price at all.


Morning trading with the robot, he opens a position and puts a take and stop, I manage it further....

I show only the beginning of the day, because I'm ashamed of it... I started to cheat...


Well, because you have a model dependence on prices (signs), over time it floats/shifts.
Maybe it's not you who started gesturing, but the model itself.
 
Maxim Dmitrievsky #:
Well, because you have a dependence of the model on prices (signs), and over time it floats/shifts.

Well, let it float, it doesn't change dramatically...

And my trading rules are inside the model and are not connected with the real price.

 
mytarmailS #:

Well, let it float away, it's not like it's going away...

And my trading rules are inside the model and are not connected with the real price.

You trade the real price. It turns out that the rules will also float.