Bayesian regression - Has anyone made an EA using this algorithm? - page 52
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Question: does SVM take into account interactions between variables or is it just the sum of the weighted individual components?
As far as I understand it, no.
If you are interested in the interaction of variables, there is a scaffolding - CORElearn, by the way, it has very interesting functions for evaluating importance based on its Relief evaluation criteria - a large number of options. There is a very interesting function of probability calibration, on the basis of which the nominal classes are then constructed.
Used it at one time....
As far as I understand it, no.
If you are interested in the interaction of variables, there is a scaffolding - CORElearn, by the way, it has very interesting functions for evaluating importance based on its Relief evaluation criteria - a large number of options. There is a very interesting function of probability calibration, on the basis of which the nominal classes are then constructed.
Been using it for a while....
Ok, thanks for the tip. Heard about this package.
I asked about SVM why, I read that this model, when taken with Gaussian kernel, classifies "almost as well" as scaffolding, boosting. But they learn much faster. I'm already exhausted from learning scaffolding, especially 5 at a time with crossvalidation.
OK, thanks for the tip. Heard about this package.
Why I asked about SVM, I read that this model, when taken with Gaussian kernel, classifies "almost as well" as scaffolding, boosting. But they learn much faster. I'm already exhausted from learning scaffolding, especially 5 at a time with crossvalidation.
I haven't compared speed, error rate is about the same. Much worse results for NS and completely unacceptable for linear regressions.
PS.
Saw somewhere that forests using crossvalidation can use all cores. Can't remember.
I did not compare the speed, the magnitude of the error is about the same. Much worse results for NS and completely unacceptable for linear regressions.
PS.
Saw somewhere that forests using crossvalidation can use all cores. Can't remember.
I use all kernels for scaffolding. In caret module there is parallelization etc. But even in this mode it takes a long time.
However, it's a fact that the speed of reference vectors should be higher. Besides, there is a library that uses the graphics core for this machine. It increases the speed by another 20 times.
The main thing is that the accuracy should be comparable. I will have to try it.
Hello, everyone!
Today, 255 years since the death of Thomas Bayes (1702-07.04.1761), English clergyman and mathematician.
https://ru.wikipedia.org/wiki/%D0%91%D0%B0%D0%B9%D0%B5%D1%81,_%D0%A2%D0%BE%D0%BC%D0%B0%D1%81
According to Wikipedia, Thomas Bayes published only two works during his lifetime - one on theology and one on mathematics.
Interesting fact. The opposite record of prolificacy belongs to our compatriot.
Thus, in 1992, the Schnobel Prize in Literature was awarded to Yuri Struchkov, of the Institute of Organic Compounds in Moscow, for the fruitfulness that surpasses all imaginable limits. From 1981 to 1990 (only nine years) he published 948 scientific works. On average, Struchkov published one paper every 3.9 days.
Thus, in 1992, the Schnobel Prize in Literature was awarded to Yuri Struchkov of the Institute of Organic Compounds in Moscow for his output beyond all imaginable limits. From 1981 to 1990 (only nine years) he published 948 scientific works. It turns out that on average Struchkov "gave away" one paper every 3.9 days.
So, a prize in literature. It's a little easier than the one in organic chemistry.
We could do that, too. No problem. It's just a question of who's going to publish it. :)
I conceived a regression EA (probably not Baisov's), externally, with the help of SanSanych's favourite R. But a lot of problems. Let's call them organizational. :)
So, a prize in literature. It's a little easier than the one in organic chemistry.
We could do that, too. No problem. It's just a question of who's going to publish it. :)
Thought about regression expert (probably not Baisovsky), external, with the help of a favorite SanSanych R. But a lot of problems. Let's call them organizational. :)
What are the problems? C R?
C R's problem is that I don't know him. :) It's a matter of time, slowly getting the hang of it.
The more difficult one is to provide real-time data exchange between R - software - MT5. I can't think of anything clever except files. I suppose that they will do for starters and then we will see.
But the exchange protocol (interface) itself is not visible.