Hi Richard,
What I did is pretty basic for those into ML.
The only "smart" thing in this process is utilizing unconventionally the optimizer/backtest engine of AB.
It is possible to detect when the optimizer is running in-sample and when it is running out-of-sample,
therefore during the IS phase you can run some code, but in OS you can run a totally different code so that learning ant testing is done on the different set of data.
The good thing about this is that it makes possible implementation of ensemble learning.
This model is a classification one, the input is a zigzag where each feature is some metric defining each of the "n" legs previous to the target leg, expressed as up or down ( y[n] > y [n-1] ), the output are the coefficients for the hypothesis function that give the estimated probability on the direction of the target leg. This process happens during IS phase, at the end of which the coefficients are stored in static variables.
During the OS phase the coefficients are plugged in the hypothesis function along with the current "n" legs metrics and a buy signal is generated whenever the result > 0.5 ( estimated probability that the next leg will be up ) and a sell signal when it is < 0.5.
Hi Aron,
Wow! I don't know many people who can do this in AFL. Looking forward to your tutorial.
In the meantime can you please explain the meaning of Target : y5 > y 4?
Regards
Richard
Posted by: Aron Pipa <aron@myafl.com>
| Reply via web post | • | Reply to sender | • | Reply to group | • | Start a New Topic | • | Messages in this topic (7) |
This group is for the discussion between users only.
This is *NOT* technical support channel.
TO GET TECHNICAL SUPPORT send an e-mail directly to
SUPPORT {at} amibroker.com
TO SUBMIT SUGGESTIONS please use FEEDBACK CENTER at
http://www.amibroker.com/feedback/
(submissions sent via other channels won't be considered)
For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
http://www.amibroker.com/devlog/
EmoticonEmoticon