Discussing the article: "Developing a robot in Python and MQL5 (Part 1): Data preprocessing" - page 2

New comment
 
Maxim Dmitrievsky #:

all signs should be pseudo-stationary, like increments. Raw prices should be removed from the training.

New features are automatically generated in the article.

Let the initial feature be increments. Then one of the generated features will be the cumulative sum of the original data, which is equal to the price. In this case, we will not know that this feature is the price. It will just turn out to be remarkable and will be added to the training set.

 
Maxim Dmitrievsky #:

Raw prices need to be taken out of training.

That's if a price is predicted. And if an increment is predicted, then remove raw increments from training? Take the increments of increments?

 
fxsaber #:

New features are automatically generated in the article.

Let the original attribute be increments. Then one of the generated attributes will be the cumulative sum of the original data, which is equal to the price. We will not know that this feature is the price. It will just turn out to be remarkable and will be added to the training set.

Well, don't do that.

 
Maxim Dmitrievsky #:

Well, you don't have to do that.

Don't use automatic feature generation?

 
fxsaber #:

Don't use automatic feature generation?

Do not use the cumulative sum of increments as a feature

 
Maxim Dmitrievsky #:

not to use the cumulative sum of increments as an indicator

So this is one of the automatically generated attributes!

 
fxsaber #:

So this is one of the automatically generated signs!

I don't see this in the feature generation function

 
Maxim Dmitrievsky #:

all signs should be pseudo-stationary, like increments.

Suppose that the price is increments of some meta-price. Then it turns out that the price is good for predicting the meta-price. But if we can predict the meta-price, it automatically means that we can also predict the simple price, because there is an unambiguous relationship between price and meta-price.

 
fxsaber #:

Suppose that the price is increments of some meta-price. Then it turns out that the price is suitable for predicting the meta-price. But if we can predict the meta-price, it automatically means that we can predict the simple price as well, because there is an unambiguous relationship between price and meta-price.

Then the price must be pseudo-stationary. This is not observed in trending markets.

 
Maxim Dmitrievsky #:

I don't see this in the feature generation function

I am zero in MO, so I am relying on the article.

There are manual approach (human selects features) and automatic approach (with the help of algorithms).

If I understood correctly, the automatic is a wider field of human choices. If a human can choose a cumulative amount, then an automaton is even more so.

1234567
New comment