Australian (ASX) Stock Market Forum

Neural network software

Jump to new
Thanks for your expert opinion sinner.

I'll have a tinker with that tonight. For binary style prediction (up/down next day), would the light or perf version be recommended?
 
No worries. Hardly an expert. Although, my macro-economics professor father trying to convince me to come back to Sydney and study finance for PhD. If I was to do that it'd definitely be in quant/financial math and then you can call me expert :p:

Personally, I am yet to find a vector space that caused trouble for Light, so no reason to use Perf really. With the way modern desktop computers are kitted out it's pretty much irrelevant anyway. I remember when we were studying the stuff (not that many years ago!!!) we had to run our training and classification on the University supercomputer!

These days I will write a script to translate data into vector space in the morning, start it training on the data-set around lunch and usually by the time the I get to work the next day it is all done training and ready to classify all on the desktop.

Binary prediction is a good learning exercise for any inference model, so it would definitely be a great place to start. Just don't get too hung up on the concept of binary predictions as the goal here. As mentioned in my first post on this thread, it is a good concept for maybe 1 or 2 neurons in a bigger Neural Network (Hurst exponent + buy the dips/sell the rips robots, anyone? ;) ).
 
All those SVM softwares are really geared towards ppl much more into programming than I am. Hard to believe there's no commercial versions aimed at traders. Even installing the SVM lite was beyond me. I need something much more point and click, so if anyone knows of something along those lines, pls let us know.
 
I just started experimenting with ANNs at the moment. Just a simple trading method trying to get the whole system operating correctly.

Even just using a couple of simple trading decisions the data that the 'learner' spits out is in the hundreds of megabytes.

Garbage in, garbage out? Or some kind of sensible statistical recognition? Only time and programming headaches will tell.

When I say I just started, I did - right before Christmas. I thought a good time to see how it went, given the poor trading conditions, likely substantially different from the data it was 'learning' from.

23rd Dec to 31st Dec:
DetailedStatement.gif

I am pleased given its very simple logic. At the moment I am keeping as much redundant code out of it because for it to learn it needs to run through its logic tr/b/millions of times so every line of code is important.

Great to hear you have had success Sinner :)

May I ask if you have any advice? Not so much specific advice, but general overall advice. I don't want any secrets or systems, but it would be good to know some pitfalls before I fall into them, or some thoughts you gained from your experience. It would be most appreciated.
 
Top