grnt426 / optimal-business-time-partitioning Goto Github PK

Project Name:	Optimal Business Time Partitioning

Description:	An agent will contain all the information about which actions
				to perform each day.  Each day a total of 16 actions can be
				performed, with 8 different actions: Search for Raw Materials
				(RM), Risky Search for RMs, Produce Low-Quality Goods 
				(LQG), Produce Medium-Quality Goods (MQG), Produce High-Quality
				Goods (HQG), Sell Goods, Sell Goods in Risky Environment, and
				Store Unused Goods (UG). After 100 days, all the agents in a
				generation will be assessed for their overall performance.
				A new generation of agents is spawned and the process repeated
				until end-conditions are met (time, generation count, or 
				plataeu	of performance).  All unused raw materials must be
				stored by the end of the day.
				
########################
Technical Details of Algorithm*:
				> Performace will be rated based on trending revenue growth.
					> Highly Likely to change or other factors thrown in
				> Elite agents will be carried over to the next generation
				untouched, while all other agents will either be replaced with
				newly generated agents, agents of mixed chromosomes, or mutated
				agents. 
				> Each chromosome will be composed of 16 3-bit strings to
				represent the division of time among the 8 different tasks.
				Crossover will involve taking 2+ parents and taking 1+
				subsequent series of tasks and recombining them. Mutation will
				be the random application of bit-flipping.
					> Actions are encoded as all values between 000:111
				> All "Risky" actions have predetermined and unchanging
				"Lucky Days".  In this way, if there is a 50% chance for a
				risky investment to pay off, then exactly 50 of the 100 days
				will be "Lucky Days", and the others failures.  The exact
				distribution of the lucky days is determined either as random
				placement, alternating, all early, all late, or a mix of any of
				the stated distribution methods.
					> Successful risky searches will give higher RMs at lower
					costs.
					> Successful risky sales will give higher returns.
					> Unsuccessful risky investments will simply have the
					opposite effect.
				> LQGs differ from MQGs and HQGs in the number of RMs consumed
				and the sale price obtained in the market.
				> During a sale action, LQ, MQ, and HQ goods will be sold (in
				decreasing amounts)
				> If an agent fails to store UGs by the end of the day then all
				UGs are lost during the next day.

########################
User Interface Details:
				> The environment will be completely alterable at the user 
				level, with possible alterable variables such as:
					> Cost/Return on RMs and FGs
					> exchange rate of RMs to Finished Goods (FGs)
					> risky pay-offs and odds
					> number of parents during pollination
					> total number of agents per generation
					> total number of elites carried over
					> total number of "failed entrepeneurs" carried over
					> total number of randomly generated agents added each
					generation
					> total number of days in a performance assessment period
					> Possibly Others
				> Start, Stop, Restart, Step, Save com.Agent Information, Save
				Environment State, and Save Everything will all be actions
				the user can perform on a simulation