System Specifications for the
Resident Artificial Intelligence

As of 2007:


>  10 processors: 4 AMD MP 2600, 4 AMD MP 2800, 1 dual core AMD 64 X2 3800+
>  average processing speed ~ 2.1 GHz per cpu
>  estimated total processing power: ~ 21 GHz

>  RAM requirements: ~ 5 Mbytes per single A.I. process



Specifications listed are for a single analysis, that is, for a single run of the program to analyze a single symbol, and produce a prediction.

>  time required for a single analysis: ~ 12 hours
>  each data point: closing price of the day for a given symbol
>  time-series data length: ~ 3 years (~250 points per year, 750 points total)
>  neural net architecture: 3-layer, feed forward
>  neural net population per run: 50 neural nets
>  back propagation training loops: 300
>  total evolution generations: 50
>  type of noise used: various distributions
>  neuron transfer funtion: sigmoid


Neural Network:

A graphic representation of one of the neural networks.

This graphical model was initially created to visually debug the program. The 3d view of the neural network's components provided visual feedback to the changes made to the code during its development.

Input neurons are the ones lined up on the left. The hidden layer neurons, spiral along the center, and the output neuron is at the far right.

The connections with values less than zero are colored red,
and the ones greater than zero are green. The color saturation for each neuron represents its input magnitude.

graphic of one neural network
  click to enlarge


back home