Machine learning in trading: theory, models, practice and algo-trading - page 639

 
Maxim Dmitrievsky:

It's just an advertising button, it's time to be more experienced on pirate sites))

You can be lucky .... )) Relaxed).

 
Yuriy Asaulenko:

You can be old sometimes, too.... ))

But we understand that this version is for review ... and will definitely buy a paper version, after we get acquainted :)

I'm already putting books under the wheels in the garage when I change them.

 
Maxim Dmitrievsky:

But we understand that this version is for familiarization... and we will definitely buy a paper version after we get acquainted with it :)

I'm already putting books under the tires in the garage when I'm changing them.

If I like it, I will.

 
Dr. Trader:

The principal components are calculated without targeting analysis. You can find the principal components, but whether they are useful for predicting the desired target is not known in advance.

And cross-entropy can be calculated with respect to a specific target, and the result will tell you which predictors should be removed because they interfere.
I want to try package EMCV, I've never noticed it before. If it works, I'll post samples here later.

Principal components are calculated without target analysis. It is possible to find principal components, but whether they will be useful for predicting the desired target is not known in advance.

Not necessarily without a target, for example with a target, plsRglm {plsRglm} and these are not the only principal components with respect to the target.

But I think you're right about the general line.

Could you run a window on the merge result and give graphs:

  • entropy values
  • adfTest results
  • ArchTest results
 

So far I found fileALL_cod.RData for the test, it was you once posted an example of predictors and triggers. As far as I remember good predictors are mixed with bad ones there, and if correctly sifted predictors and trained model on Rat_DF1 and tested on RAT_DF2, then errors should be the same. I'll test EMCV on this, at least you can quickly see if this package will give results or not.


SanSanych Fomenko:

Could you run a window on the merge result and give graphs:

  • entropy values
  • adfTest results
  • results of ArchTest

If the package can handle the first task, I'll give it a try.

 
Well, this is a dickens comrade!!!! In general, I calculated a lot of things and it seems to work, at least some estimate similar to the VI received... Although in the correctness of the calculations I have great doubts.... But with the help of this metric selected inputs and threw them to the training, let's see what happens.... From the first run, it has chosen two inputs out of 25 five and has shown good optimization results. I will try to test it to increase the number of entries. So that he built a more or less complex model, which is not embarrassing to look at.... That's it and throw it on the chart or a number of similar models I will put in MT to evaluate the work on the OOS well and of course to calculate the VI of these models, maybe something reasonable and will turn out....
 
Dr. Trader:

So far I found fileALL_cod.RData for the test, it was you once posted an example of predictors and triggers. As far as I remember good predictors are mixed with bad ones there, and if correctly sifted predictors and trained model on Rat_DF1 and tested on RAT_DF2, then errors should be the same. I will test EMCV on this, at least you can quickly see if this package will give results or not.


If the package can handle the first task, I'll try it.

I am waiting, it's very interesting.

 
Old Fa stood in an orchard. An orange tree spread out before him. Birds were flying everywhere, chaotically.
The birds did not land on the orange tree and he thought the fruit was poisonous. He would take a step to the side and see
the apple tree behind the orange tree, where on the fruit and a couple or even more of the birds flying chaotically around
would surely sit down... But he continued to stand in one place... Hungry, exhausted, depressed...
 

Here's the seven of me!!!! And mark this day on your calendar with a red pencil, because today is the day I downloaded R and will be spinning it little by little. I know that there are craftsmen on it, so I will not refuse any help to achieve their goals in the universe of R.

Well, what is there to hide. I need a reliable calculation of entropy and everything related to it. I understand that the package can be downloaded thematic toolkits. If there is one and you know about it, don't keep it to yourself... Talking out of turn!!!! Now you will definitely be listened to here. Thanks!!!!

 
Mihail Marchukajtes:

Here's the seven of me!!!! And mark this day on your calendar with a red pencil, because today is the day I downloaded R and will be spinning it little by little. I know that there are craftsmen on it, so I will not refuse any help to achieve their goals in the universe of R.

Well, what is there to hide. I need a reliable calculation of entropy and everything associated with it. I understand that the package can be downloaded thematic toolkits. If there is one and you know about it, don't keep it to yourself... Talking out of turn!!!! Now you will definitely be listened to here. Thanks!!!!

There's nothing to say. It's as clear as day in a crystal palace.

I count nonentropy. I instantly see the difference of the current distribution from the normal distribution (i.e. one parameter replaces all moments of NE at the same time!), i.e. I see the non-markness of the process with its "tails" entirely. All that's left is to make a matching table for making deals and it's done.