Machine learning in trading: theory, models, practice and algo-trading - page 2373

New comment
 
secret:

Well, it's on the S&P that there is demolition. And the fx doesn't have it on a regular basis.

Well, if you remove inflation, the trend there is not so constant either. On the forex, inflation is, in part,automatically removed.

secret:

The idea with the put-call ratio, etc. is good, but I have not seen anyone boasting about profits from it) After all, the market moves by participants with money, not with experience, while the liquidity of currency options is none. And their purpose is different.

Sure, all the fish have long been caught, but sometimes you have to check if there are any new ones).

secret:

p.s. it would be nice to have a branch on rumors, because they are everywhere off the subject)

Only for) But they will riddle a branch) For most random processes is some philosophical and chatterbox stuff from the category of those things about which no matter what you say - everything will be true in some sense) Again, there will always be a lot of practical practitioners, burning desire to explain the theoretical theorists as terribly far from the practical practice)

 
Aleksey Nikolayev:

Well, if you take away inflation, the trend there is not so constant either. On the forex market, inflation is, in part,automatically removed.

I'm sure all the fish have long been caught, but sometimes you have to check to see if there are any new ones.)

Only for) But they will riddle a branch) For most random processes is some philosophical and chatterbox crap from the category of those things about which no matter what you say - everything will be correct in some sense) Again, there will always be a lot of practical practitioners, burning desire to explain to the theoretical theorists how terribly far they are from the practical practice)

Right, it's better to argue about trees and forests
 
Can anyone help make a script in R that takes a sample and applies "Lasso regression" model to it? I heard an opinion that it works well for binary traits with high correlation - I want to try it. However, I need a classification, not a regression, but I think this is not very important here - the man was talking about classification.
 
Aleksey Vyazmikin:
Can someone help make a script in R that takes a sample and applies "Lasso regression" model on it? I heard an opinion that it works well for binary features with high correlation - I want to try it. But I need classification, not regression, but I think this is not very important here - the man was talking about classification.

Look towards ROCKET, the algorithm creates a bunch of decorrelated features

 
Aleksey Vyazmikin:
Can someone help make a script in R that takes a sample and applies "Lasso regression" model on it? I heard an opinion that it works well for binary features with high correlation - I want to try it. But I need classification, not regression, but I think that is not very important - the man was talking about classification.

https://www.pluralsight.com/guides/linear-lasso-and-ridge-regression-with-r

http://www.science.smith.edu/~jcrouser/SDS293/labs/lab10-r.html

https://rstatisticsblog.com/data-science-in-action/machine-learning/lasso-regression/

 
mytarmailS:


We need to create a model of the market for those --- A simplified feature space that has useful properties for us


Why simplify

1) Visibility, perceptibility.

2) Simpler space implies more repeatability, regularities are easier to find and do not repeat every two years

3) Minimization of the chance of combinatorial explosion in the search for patterns

4) Smart simplification removes noise


What are the useful properties (what you need from a model)

1) The model should be adequate to market movements

2) Repeatability of data within the model

3) simplicity


May someone else add something to it. I also invite you to discuss the variants of the model

A mathematical model of physical processes with reasonable assumptions is an achievable goal with sufficient accuracy.

Multifactor mathematical models of simple processes, Brownian motion is a good example, also reach the required accuracy.

But I have not encountered complex systems.

Alexei Nikolaev in the blogs did a model of minority games on R.

It turned out similar to tick movement. But adding factors of different influences to the model just does not work. And where to go, into complex game models or make a more complex model out of a simple one, it is difficult to decide more or less correctly. While the model initially simple does not turn out because of the number of classes / groups of acting variables more than 2. In Brownian motion there is 1 class, and the model is quite complex.

 
Maxim Dmitrievsky:

look towards ROCKET, the algorithm creates a bunch of decorrelated features

I need to experiment with binary ones.

 
mytarmailS:

https://www.pluralsight.com/guides/linear-lasso-and-ridge-regression-with-r

http://www.science.smith.edu/~jcrouser/SDS293/labs/lab10-r.html

https://rstatisticsblog.com/data-science-in-action/machine-learning/lasso-regression/

Thanks, but I'm a passenger in R, in all the variety of examples of code there I would drown - I would like a trivial working script, like there aren't many parameters...

 
Aleksey Vyazmikin:

Thanks, but I'm a passenger in R, in all the variety of examples of code there I would drown - I would like a trivial working script, like there are not many parameters...

Here you go, my lazy friend.

library(glmnet)
Y <- sample(c(-1,1),100,replace = T)   #  data
X <- matrix(rnorm(400),ncol = 4)     #  target
tr <- 1:80 #  train idx

lambda_seq <- 10^seq(2, -2, by = -.1)
#  identifying best lamda
best_lam <- cv.glmnet(x = X[tr,], y = Y[tr],alpha = 1, lambda = lambda_seq, nfolds = 5)$lambda.min

#  train model with best lamda value identified
lasso_best <- glmnet(x = X[tr,], y = Y[tr], alpha = 1, lambda = best_lam)
#  predict new data
pred <- predict(lasso_best, s = best_lam, newx = X[-tr,])


Don't ask me about the model, I haven't used it and don't intend to use it. I've grown out of that way of thinking, so I believe in miracle models.)

 
Valeriy Yastremskiy:

A mathematical model of physical processes with reasonable assumptions is an achievable goal with sufficient accuracy.

Multifactor mathematical models of simple processes, Brownian motion is a good example, also achieve the desired accuracy.

But I have not encountered complex systems.

Alexei Nikolaev in the blogs did a model of minority games on R.

It turned out similar to tick movement. But adding factors of different influences to the model just does not work. And where to go, into complex game models or make a more complex model out of a simple one, it is difficult to decide more or less correctly. While the model initially simple does not turn out because of the number of classes / groups of acting variables more than 2. In Brownian motion there is 1 class, and the model is quite complex.

complicated

1...236623672368236923702371237223732374237523762377237823792380...3656
New comment