https://github.com/diego-vicente/ann-ntnu
Artificial Neural Networks visualization for IT3708@NTNU
https://github.com/diego-vicente/ann-ntnu
ann artificial-intelligence artificial-neural-networks bio-inspired ntnu visualization
Last synced: about 2 months ago
JSON representation
Artificial Neural Networks visualization for IT3708@NTNU
- Host: GitHub
- URL: https://github.com/diego-vicente/ann-ntnu
- Owner: diego-vicente
- License: mit
- Created: 2017-01-14T21:51:13.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2017-07-09T10:40:14.000Z (almost 8 years ago)
- Last Synced: 2025-03-26T00:36:04.312Z (2 months ago)
- Topics: ann, artificial-intelligence, artificial-neural-networks, bio-inspired, ntnu, visualization
- Language: Python
- Homepage:
- Size: 417 KB
- Stars: 7
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.org
- License: LICENSE
Awesome Lists containing this project
README
#+TITLE: Artificial Neural Networks visualization
#+AUTHOR: Diego Vicente Martín
#+EMAIL: [email protected]
This code implements a simple Neural Network that can be run on a graphical
environment using ~pygame~. This project implements 4 different agents: A
greedy agent that follows simple rules (that can be used as baseline), an agent
that learns by supervision of the greedy agent, an agent that learns by
reinforcement, and a enhanced agent that learns by reinforcement but having
more information about the surroundings.
The simulation window also includes a graphical representation of the neurons
and their synapses, that changes live accordingly to the game progress: a green
synapse means a positive connection between the input neuron and the output
neuron (that input triggers the output) and a red synapse mean a negative
connection (the input inhibits the output).
[[./report/img/input.png]]
The simulation can be controlled using the arrow keys to move step by step, or
using ~SPC~ to run a game automatically. By pressing ~n~ a new board is
randomly generated and loaded. By pressing ~t~ a visual training can be
performed: the agent rapidly executes a batch of games and shows the evolution
of the brain to the user. Since the visualization can be confusing with certain
values of the connections, ~w~ can be used to print the weights in the terminal
that is running the simulation in text format.
The ~src/~ folder contains all the source code. To run it, just do: ~python3
src/run.py~ and you will be prompted with several options that were needed for
the demo presentation. In the repository there is also a report that explained
the details about the project as well as some study on the agents' learning and
performance.
[[./report/img/agents.png]]
-----
This code was presented as the Project 1 for the course Bio-Inspired Artificial
Intelligence (IT3708) @ NTNU (Spring 2017).