-
Notifications
You must be signed in to change notification settings - Fork 0
alsonkemp/neural_net_heatbugs
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
= Neural Net Heatbugs This is an old project which some folks might find of interest, so I'm putting it up on GitHub. MIT licensed. It's an old and heavily incremented project which I never expected anyone to see. I've added docs to the code, but I'm not proud of the code quality. Brief description: * This is RePast's heatbugs agent simulation that's been sexed up with genetic algorithms, neural networks, etc. * Each bug has a "genotype" that's just a bitstring. * Neural networks are constructed from the genotype. * The genotype or bitstring is parsed in order to generate a list of rules. * The rules are repeatedly applied to a "predicate", or starting, neural network in order to build the neural network the bug uses. * The resulting neural network is used to process inputs to the bug and to produce outputs to move the bug. == Heatbugs Simple: each agent should move around to find the spot that gets them the perfect temperature. * There are a number of heatsources on the grid. * Each heatbug also gives off heat (configurabl). * After a period of time, the Reaper runs and removes bugs from the space. Bugs can get reaped for various reasons: * Not having a nice temperature after their "childhood". * Never having moved during "childhood". == Build + Run You probably don't *need* NetBeans to build it, but I've only ever used NetBeans, so: * Open the project in NetBeans and run it... * Assuming that it runs, you'll get two windows. Click the play button to run the simulation. * A bunch of windows will then pop up. * Note!: the actual graphical display will minimize itself to speed the simulation. * Wait 5 minutes. * Check out the simulation to see if agents are exhibiting interesting behavior... == You should try... hmm... I should add interesting parameters here... == Architecture Code: * src/uchicago/src/sim/nnheatBugs/HeatBugsModel.java - this is the simulation and is where everything is controlled. * src/uchicago/src/sim/nnheatBugs/HeatBug.java - this is a bug. * src/uchicago/src/sim/network/ActivationFunction.java - this is an activation function for a neuron. * src/uchicago/src/sim/network/NeuralNetworkNode.java - this is a neuron. * src/uchicago/src/sim/network/CHHNeuralNetworkNode.java - this is a particular type of neuron. * src/uchicago/src/sim/network/SigmoidNeuralNetworkNode.java - this is a particular type of neuron. * src/uchicago/src/sim/graphgrammar/* - a giant flail... But the gist is that this is the code that converts genomes -> neural networks... There are a number of ways it can do the conversion... Terminology: * Heatbug/Agent - the thing that's moving around. * Genotype - a BitString representing the genes of a bug. * Rule - matches a spot on a Matrix and might expand or overlap a set of values on matrix. * Matrix - a (probably sparce) matrix of node-node connections. All inputs are on one side and all outputs are on the other; a value V at (M,N) says that node M is connected to node N with value V. * Structure - this is just a string describing some basic characteristics of a neural net. This was introduced in order to prevent a single neural net from dominating and is used to prevent inbreeding. == Warnings * It had been long enough since I'd coded that I'd kinda forgotten about recursion, so the code to apply the grammar to the connection matrix is fabulously complicated. * I recycle frequently used objects to prevent GCs, so you'll see Factory objects for Matrices, etc.
About
No description, website, or topics provided.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published