Machine learning in trading: theory, models, practice and algo-trading - page 3664

New comment
 
fxsaber #:

I had a question like this. If a Python chart had this information, you could keep your mind off dumb questions.

It would be better to be distracted by code
 
It looks like something not corresponding to what is written is drawn vertically. 1.5-2 pips for 2000 trades - something wrong...
I agree - the charts are incomprehensible. That's why there are a lot of questions not about the code, but just to understand the situation.
 
Forester #:
It looks like something not corresponding to what is written is drawn vertically. 1.5-2 pips for 2000 trades - something wrong...
I agree - the charts are incomprehensible. That's why there are a lot of questions not about the code, but just to understand the situation.
2500 or 25000, depending on what a pip is for you.
I have 5 decimal places.

This is really some bullshit we're discussing.

I'd better go to bed.

 

More fun pictures. The method itself can give, even without additional tricks.

Learn the matrix (I will discuss only it).


 

And 15 minutes in the same period


 
Maxim Dmitrievsky #:

More fun pictures. The method itself can give, even without additional tricks.

Learn the matrix (I will discuss only it).


The most interesting thing is the target. As I understood the chips many cope, even about MAshki wrote, but the specifics are also interesting.
 
Forester #:
The most interesting thing is the targeting. As I understood, many fiches can cope, even about MAshki wrote, but the specifics are also interesting.
If the targets somehow miraculously work, then there is no need to fix anything.

F-I divides the dataset into traded and not traded, traditionally 2 models are trained afterwards.
 
Maxim Dmitrievsky #:
If the targets are somehow miraculously working, there is no need to fix anything.

F-I divides the dataset into tradable and non-tradable, traditionally 2 models are trained afterwards.
Have you never tried to trade these clusters directly? Would the statistics be better or worse? If these clusters already separate classes well, I don't see what additional training by other models will do?
Additional cleaning up of errors, for example from 30 % to 20 %? Well, it's like in a tree just to increase the depth of splitting by 1-2 levels. But this often leads to retraining.
 
Forester #:
You still haven't tried just trading these clusters directly? Would the statistics be better or worse? If these clusters already separate classes well, I don't see what additional training with other models will do?
Additional cleaning up of errors, for example from 30 % to 20 %? Well, it's like in a tree just to increase the depth of splitting by 1-2 levels. But this often leads to retraining.
This is a f-ya for data preprocessing. Clusters do not participate in trading.
 

Your last idea with separating the clusters by class is original. I just got back to MO a few days ago. As I check out the ideas that have been piling up for up to 10 months, I'll check this one out too.

In general, your clusters are like individual leaves in a tree.

Alexey Vyazmikin and I have studied them, a little differently. I had drawdowns or no trading there for up to 2 years. Your charts look more interesting.

1...365736583659366036613662366336643665366636673668366936703671...3699
New comment