Eigen Memory Trees

The eigen memory trees (EMT) paper can be found at [1].

Motivation

EMTs were developed to possess three qualities:

1. iterative growth (i.e., the tree's memory bank grows one memory at a time)
2. iterative learning (i.e., the tree's search function improves one example at a time)
3. sublinear complexity (i.e., the tree's insertion and query complexity is O(log n))

Experiments

This repository contains two experiments using 206 datasets to evaluate EMT on contextual bandit problems.

There are two experiments provided for EMT:

1. Unbounded -- the tree keeps all the memories it is given (python run_unbounded.py)
2. Bounded -- the tree must begin pruning memories once it reaches its bound (python run_bounded.py)

Results

After running the experiments the results can be visualized using /notebooks/plots.ipynb

Dependencies

An environment.yml file is provided to create a conda environment for the experiments.