The eigen memory trees (EMT) paper can be found at [1].
EMTs were developed to possess three qualities:
- iterative growth (i.e., the tree's memory bank grows one memory at a time)
- iterative learning (i.e., the tree's search function improves one example at a time)
- sublinear complexity (i.e., the tree's insertion and query complexity is O(log n))
This repository contains two experiments using 206 datasets to evaluate EMT on contextual bandit problems.
There are two experiments provided for EMT:
- Unbounded -- the tree keeps all the memories it is given (
python run_unbounded.py) - Bounded -- the tree must begin pruning memories once it reaches its bound (
python run_bounded.py)
After running the experiments the results can be visualized using /notebooks/plots.ipynb
An environment.yml file is provided to create a conda environment for the experiments.