arcvega/beach_house

A project which aims to simulate the environment stemming from the "Beach House Theory" proposed by Matthew Colbourne

beach_house's Issues

Priority queue for blacklist

Having a priority queue may be nice since rejections caused by a Jim could make it harder for the rejected Matt to get off of Caseys blacklist, hence making rejection by Jim more influential

Potential Casey selection

Currently, Matt always picks the most attractive Casey and hence all matts will start to follow a single Casey instance, completely ignoring all other agents

The way the potential Casey is selected may need to be amended since the majority of Casey agents will remain unpaired indefinitely

Maybe the Matt should first stick to a range of Casey agents which are between his own score and the randomized standard

Swap chasing behavior for Matt and Casey

Matt should be the one chasing Casey, which is not the case

Generic Agent Class

There are several methods in Casey and Matt which do the exact same thing apart from the cast used, ie. Matt casts partner to Casey and Casey casts her partner to the Matt class

Most of this could probably be solved by using Javas generic type and making these methods part of the Agent baseclass to eliminate code duplication

Repulsive force while escaping Jim

It may be nice to introduce a repulsive force that acts on the pair when they escape Jim, so that they don't run towards him as can be the case when using a random walk

Casey not removing edge

There appear to be cases where Casey doesn't remove the edge between herself and a coupled Matt.

This is probably happening when Casey steals a Matt, she uncouples with him but the graph edge between them is not deleted, so they are free to re-pair with other agents but they have a remnant edge connecting them

Blacklist Test cases

It may be time to come up with some test cases which can be used to evaluate the correctness of the blacklist since doing this manually is rather strenuous and ineffective

Simulation hyperparameters

Add the ability to tune hyperparameters such as population size for respective agents as a separate class, to prevent meddling with the Simulation class

Random walk rework

The random walk by the agents is incredibly localized (they don't stray far), it may be worth reworking this to make the exploration factor more extreme

Network linking for agents

To make the relations between between the three different agents more apparent, it may be a good idea to add a network which will help visualize the relations and connections

Normal distribution of affinity

The affinity that is assigned to the agents should be done in accordance to a normal distribution, rather than equally distributed randomness

Casey Blacklist size

Casey is not using her own definition of the blacklist size CASEY_MAXIMUM_BLACKLIST_SIZE, instead MATT_MAXIMUM_BLACKLIST_SIZE is being used which is incorrect