Trader Action Weights - Gink3/econ_sim GitHub Wiki

Problem Description

Currently I need to create a set of weights to determine trader actions

Decision Trees

Looks like a decent option for selecting different actions. It would allow me to explicitly check for certain behavior, but might limit the adaptability of the model.

Pros Cons
Easy to setup Can limit adaptability
If shallow can be inexpensive to run Can be challenging to add nodes at large depths

Neural Nets

Neural Networks are my go to currently. I am looking at using the Neuroflow crate for rust. This could be used with a genetic algorithm to increase the fitness in traders and would be more adaptable to events, which are a planned feature of the simulation.

Currently looking at a few different crates to use implementing the neural nets in rust.

Input Space

  1. Current stock for sale
  2. Stock Price
  3. Current stock holdings
  4. Recent Price change (Daily or weekly?)

Output

2 output nodes

  1. Action choice
  2. Optional sell/buy amounts