https://github.com/baggepinnen/kalmantree.jl

Bayesian function approximation using KD-trees with Kalman-filter cells
https://github.com/baggepinnen/kalmantree.jl

Last synced: 7 months ago
JSON representation

Bayesian function approximation using KD-trees with Kalman-filter cells

• Host: GitHub
• URL: https://github.com/baggepinnen/kalmantree.jl
• Owner: baggepinnen
• Created: 2019-03-01T17:42:44.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2020-12-13T07:57:44.000Z (almost 5 years ago)
• Last Synced: 2025-01-22T04:13:26.119Z (9 months ago)
• Language: Julia
• Size: 96.7 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# KalmanTree.jl

*Not yet ready for consumption*

[![Build Status](https://travis-ci.org/baggepinnen/KalmanTree.jl.svg?branch=master)](https://travis-ci.org/baggepinnen/KalmanTree.jl)

[![codecov](https://codecov.io/gh/baggepinnen/KalmanTree.jl/branch/master/graph/badge.svg)](https://codecov.io/gh/baggepinnen/KalmanTree.jl)

This is so far a toy project. The goal is to develop a function approximator particularily suitable for representing value functions `Q(s,a)` (from control and reinforcement learning). Key points are that the approximator is easy to update recursively and to take the `argmaxₐ(Q)` of.

The approximator consists of a KD-tree, where each cell is a quadratic model. The models are updated recursively once a new datapoint is available. Every now and then, the cell/model with highest covariance (in some sense) is split into two new cells/models.

![window](figs/demo.png)