Anyone up for collaberating on replicating Chan scripts in Python?
- Thread starter dtrader98
- Start date
Well, it's pretty easy to get the code, but I don't want to be in any copyright infringement. If you are really interested to port it, I'll give a simple example I'm working on right now.
cadf(seq1,seq2,0,1); is the matlab code
to generate augmented dickey-fuller test on two sequences order 0, lag 1.
If you could translate to an equivalent function with arguments in Python, that would be great!
(P.S. There are many other languages to do it, that is not the purpose of this thread/experiment).
those are generic functions. look at...
http://numpy.scipy.org/
http://www.scipy.org/
http://code.google.com/p/econpy/
http://rpy.sourceforge.net/rpy2.html
http://numpy.scipy.org/
http://www.scipy.org/
http://code.google.com/p/econpy/
http://rpy.sourceforge.net/rpy2.html
Hi Rosy,
Thanks for sharing some generic links, although I already got that far. I was specifically looking for the CADF.py (or equivalent to matlab.cadf.m) function to be readily instantiated in the form mentioned. If you have a link to that; that would be great.
Also, I don't want to mix languages (i.e. no calls to R, etc.). The goal is to keep it all py.
Even better,
a = ([1,2,3,4,5,6])
b=([1.2,2.2,3.2,4.2,5.2,6.2])
could you show an explicit example of these two series being run through a
cadf.py function with equivalent statistical output to matlab?
Back in 96 and 97 when I was even more of a newb than I am today, I developed a couple of projects in a functional programming language called Miranda (which is probably now dead)....seems these things are back in vogue.
Could be fun to pick that stuff up again and see how it can be applied in a trading context.
Quote from dtrader98:
I have found python to be faster at implementing concepts in my head, than java... It's that simple, I think.
However, like any language, I have to familiarize myself with the syntax (like slicing quirks not using 1st element);
and the best way to do so is practice on examples, just like back in school.
I doubt the execution time would be faster, just less complex. Cointegration would be a good idea; I didn't want to limit the topic list. Not sure why you would see it as chaotic? If someone like yourself were on board, you could architect from a top level and break down the tasks. I mean the algorithms are already there; how hard can it be?
You're right about many of the advantage of python.
Still, to make it easy to quickly "implement" a trading idea with python (I know python) that will serve a group of people (not just yourself) requires enormous work.
It's because you need basic plumbing to load and process the data, synchronize it for multi symbol portfolio trading, handle generating bars in multiple time frames, handle processing of various types of orders (simulated unless you're using tick data). If using tick data , you have many months of work to get the performance down to anything acceptable.... the list goes on and on and on.
I want to propose to you a possibly simpler and quicker option to use python.
Have you heard of IronPython? It's a free python implementation which runs on .NET instead of the python parser. That means it's fully compatible with C#, VB, etc.
That also means it can interact with already existing C# objects and so on.
It would be so cool to interactively write a couple trading ideas, run them and 10 different symbols with some portfolio % allocation and have them run and a chart pops up that you can dynamically examine with performance stats, etc.
Well, some people have suggested adding python to the tickzoom project.
That's because it's open source. Has the platform separate from the GUI as an API and the data engine also separate again as an API.
One hedge fund, for example is bypassing the tz GUI and using the platform as a black box programmatically with their own GUI and server systems.
The best way to integrate tickzoom into Python would be to do it using IronPython which runs on Windows and Linux.
Tickzoom already has all those facilities and plumbing described above (and many more) for historical testing, optimizing, replay and easy switch to live trading.
Of course, no actual paying users of tickzoom as requested python but it's on my list only because I find the idea very interesting and geeky personally.
If you're at all interested in doing this and thing integrating with TickZoom would suit your purposes, I would be interested in contributing to your effort.
If you think we can do it, I'll give you the exclusive opportunity to using the free trial as long as we need to get python working on tickzoom. If you get it working, I'll give you a free license so you can run it live.
I think the market for this financially is very, very small--almost non-existent. I'm only offering this because it's very interesting technically and "geeky" so I'll be happy to support you however I can to do this.
If you want to discuss further, just sign up on the waiting list (on my signature) and reply to the emails--that's my real personal email used for the list. I don't get PMs or email notification on elitetrader.
Wayne