Machine learning in trading: theory, models, practice and algo-trading - page 815

New comment
 
Maxim Dmitrievsky:

Now you take each predictor's historical byssell/hold estimate, convert it to a probability estimate.

Take several predictors and do the same for each of them

find conditional probabilities of profit

and then you feed it into NS or fuzzy sets, like in this example

the average estimation will fluctuate around 0.5 for each predictor, but the wonders of the Bayesian approach will bring totals to an acceptable level

is in theory :)

In all the classification models I know, the result can be ordered as a class, or it can be ordered as a class probability. Usually this probability is divided in half for the two classes. But there is a program that divides this probability not by half, but by some other considerations.

 
Vizard_:

)))

Wizard_, I read your posts carefully

Explain the pictures, what's up?

 
SanSanych Fomenko:

In all the classification models I know, the result can be ordered as a class, or it can be ordered as a class probability. Usually this probability is divided in half for the two classes. But there is a program that divides this probability not by half, but by some other considerations.

Yeah, logistic regression is called ))

 
Maxim Dmitrievsky:

Yeah, logistic regression is called ))

No, I meant

CORElearn::calibrate()

Given probability scores predictedProb as provided for example by a call to predict.CoreModel

and using one of available methods given by methods the function calibrates predicted probabilities

so that they match the actual probabilities of a binary class 1 provided by correctClass.

calibrate(correctClass, predictedProb, class1=1,
method = c("isoReg","binIsoReg","binning","mdlMerge"),
weight=NULL, noBins=10, assumeProbabilities=FALSE)


PS.

There are plenty of regressions that have a class in their output.

The best known and relatively simple one is glm().


SEE .

Actually it is highly desirable that the posts were more specific, with an indication of the original source, and preferably specific functions.

 
Vizard_:

Fa, you've been fucking around for years. glm(.~...,family = "binomial")
logistic))) Drop the fuck all. Only Doc and Toxic are adequate in this thread...

What did Toxic say once in his life that he suddenly became sane?

He doesn't write anything.

Coconut's totally inadequate and lost, and so are you.

 
Vizard_:

Only Doc and Toxic are adequate in this thread...

only Toxic

 
Don't attribute me here at all. I don't know anything......
 
Vizard_:

Fa, you've been bullshitting for years. glm(.~...,family = "binomial") is
logistic)) Drop the fuck all. Only Doc and Toxic are adequate in this thread...

Citizen in the mask, get under the bench and before you post bullshit:

  • read that my post is about calibration, for which I have the tool I mentioned, not logistic regression
  • Drop what the fuck you're doing and read the attachment, maybe you'll shut up for a few years in the joys of a variety of logistic regressions. After you read it, enlighten yourself here on how to use logistic regressions to calibrate classes.
There's more to this idea in the attachment

Files:
kk0hw_lvr9elz_GLM.zip  199 kb
 

The message of this topic is meaningless, because everyone has his own model. The only thing that unites those who took part in this mess is integration of external tools with MQL5. I have a converter from Spark Random Forest into Alglib (MQL5) format. If you want to make a community dedicated to integration, it will be of use to everyone.

P.s. I prefer Git

 
It makes a lot of sense. ) There are interesting things here. Only to read the whole topic is no longer real.
1...808809810811812813814815816817818819820821822...3629
New comment