Machine learning in trading: theory, models, practice and algo-trading - page 586
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
So try to find it.)) Such an MLP would be optimal.
Yesterday I found a convolutional NS - usually used for image recognition. Naturally, there are all the utilities - training, etc. Made for use in Python.
There are also recurrence, etc., but that's not very interesting yet.
Since the convolutional network is not fully connected, we can greatly increase the number of neurons without loss of performance. But I have to understand all the details, I haven't got into it yet.
Popular description -https://geektimes.ru/post/74326/Take a look at HTM - I wrote to you earlier. In its structure the use of context is built in. And there is an implementation in python.
About non-stationarity from Haykin
About what I (and not only) and wrote, and here as it is already long ago proved :)
I happen to know that there is a new version of the gbm package that is not yet in the cran repositories.
https://github.com/gbm-developers/gbm3
installation:
After that the updated library in R is called gbm3:
I just changed the name of the library from gbm to gbm3 in the code of the r script, the rest of the code worked without changes.
When I was searching for parameters of gbm model it sometimes happened that the training consumed a lot of RAM and the whole computer hung for a dozen minutes. With this new version this has not happened yet, I recommend to try it.
About non-stationarity from Haykin
About what I (and not only) and wrote, and here as though everything is already long ago proved :)
I do not understand at all the text about non-stationarity: the arrival of new observations destroys the previously detected relations? It may be due to my ignorance of NS, but in trees exactly one line of observations is considered and it is impossible to destroy the previously constructed trees. They may not occur in the future, in the future there may be exactly the same trees, but they will belong to a different class, but the old everything will remain immutable.
PS.
There are trees that consider a few lines when building a tree... but it doesn't seem to change the point
I do not understand at all the text about non-stationarity: the arrival of new observations destroys the previously identified relationships? It may be due to my ignorance of NS, but the trees consider exactly one line of observations and it is impossible to destroy the previously constructed trees. They may not occur in the future, in the future there may be exactly the same trees, but they will belong to a different class, but the old everything will remain immutable.
PS.
There are trees that consider a few lines when building a tree... but it doesn't seem to change the point.
No, it boils down to building pseudo-stationary series consistently, retraining as often as possible... that's basically what I'm doing
or build linear/nonlinear filters... I understand that before this you need to consider the dynamics of change in the effect of predictors on the target, and try to adapt the output signal through the filter coefficients, depending on changes in the environment
Well, in general there is nothing special. At least in this chapter.
About non-stationarity from Haykin
That's what I (and not only) wrote about, and here it's like everything has been proved a long time ago :)
About non-stationarity from Haykin
That's what I (and not only) wrote about, and here it's like everything has been proved a long time ago :)
Have a look at HTM - I wrote to you earlier. Its structure makes use of context. And there is an implementation in python.
I couldn't find something in our correspondence. What is HTM and what is it?
Actually, I just started studying Python 2 or 3 days ago. I don't understand much about it yet). So, it will take me a while to get down to business.
I couldn't find something in our correspondence. What is HTM and what is it?
Actually, I just started studying Python 2 or 3 days ago. I don't understand much about it yet). So it will take me a while to get down to business.
This is a good place to start: https://numenta.org/implementations/
There is a book in Russian - the translation is almost adequate.