Welcome to the AI and ML classes "almuni" group

[Maxime]

Hi,

First of all, thanks a lot for your interest in our AI-ML group. We
have over 20 people on this mailing list since its creation two days
ago.

Just a quick note, you can choose how often you want to receive emails
(every messages, one email per day, only web access with no emails)
using this link:

http://groups.google.com/group/ai-ml-class/subscribe

If you’d like to introduce yourself to the group that would be great.
I’m sure some of us have similar interests/experiences/etc.

Many of you had great project ideas that I’d definitely be interested
in working on with you. One of the most often suggested idea was to
start by working together on a kaggle contest. Most of you sent me
their kaggle username and are now on our “Concrete Narwhal” kaggle
team.

I submitted our first entry and we’re currently in 9th place with an
error of 0.95498

http://www.kaggle.com/c/AlgorithmicTradingChallenge/Leaderboard

This first entry simply assumes that the price at time 100 will be the
price at time 49. Since we also know the price at time 50, we can draw
a straight line from time 50 to time 100 to form the slope on the
right-angled trapezium (apparently that's the name for this kind of
shape) of prices from time 50 to 100. In other words, it gives us the
slope of the triangle on top of the rectangle formed when we keep the
price at time 50 constant.

http://www.kaggle.com/c/AlgorithmicTradingChallenge/Details/Background

http://en.wikipedia.org/wiki/Quadrilateral#Taxonomy

So the next step I’d like to try is to use perceptron’s/neural
networks to see how they do with the data:

https://www.ai-class.com/course/video/videolecture/53 (perceptron)

http://www.ml-class.org/course/video/list (neural networks)

http://en.wikipedia.org/wiki/Perceptron

I think I’ll use Matlab for that attempt. We can share our code here:

http://code.google.com/p/ai-ml-class/

Finally, we had a pretty enjoyable chat session with a few of you last
weekend. Let me know if you’d be interested to have another chat
session (maybe on skype)?

Thanks again for your interest. ;) Maxime

[Dhruvkaran Mehta]

Hey Maxime,
Just to clarify the approach, you are assuming a constant rate of
recovery between t=50 and 100 right?
Will setup svn today.
--Dhruv

--
Dhruv

[Maxime Leclerc]

Hi Dhruv,

First off, did you guys see this cool video about kaggle?

http://www.abc.net.au/catalyst/stories/3296837.htm

But back to this competition, you're right, our first submission was
more of a starting point that did not used AI or ML but just put our
team on the leaderboard.

It assumes as you said "a constant rate of recovery between t=50 and
100". It also assumes that the price at t = 100 will be the price at t
= 49.

In other words, it assumes that when someone buys or sells a lot of
shares and makes the share price move with this large transaction, it
will take 50 time units for the share price to get back at its
original value (i.e. the share price before the large buy or sell).

I think an intermediary step before looking at more complex models
like neural networks might be to try some form of regression like the
ones covered in ml-class (linear, polynomial, logistic, etc.).

What do you think?

Take care ;) Maxime

[Paul Tan]

Could you post the code you used to generate the data?

I'm trying to get started, but wanted to get a framework to input the
data and generate the output data that the competition requested.

I am thinking of just putting the data into a neural network with a
large number of hidden units and see what happens. I will post my
Octave code here for others to see if they have a better idea.

Thanks.

Paul Tan.

[anky]

whats ur skype id maxime?? ...i'll start working on the project
probably abt this Wednesday !!

[James]

Hello Maxime,

I am interested in joining the group.

My email is james...@gmail.com and my kaggle userId is jamesclyeh

Can you add me to the group?

Thanks,
James