Search-based Generalization and Refinement of Code Templates - README
=====================================================================

This archive contains all data for the design pattern generalization experiment, as well as the implementation of our evolutionary search algorithm. In this experiment, we have used our algorithm to automatically generalize one instance of a design pattern into a template that describes all instances of that design pattern. (All instances are known by the algorithm in order to compute the fitness of a given template.)

We have performed this experiment for 8 design patterns in the JHotDraw 5.1 and Nutch 0.4 projects:

- JHotDraw:	Observer, Prototype, Template method, Strategy, Factory method
- Nutch:	Template method, Strategy, Bridge

This archive contains the following files:

 - Experiment reproduction.txt					Instructions to reproduce the experiment
 - Experiment configuration.txt					Describes the complete configuration used for each experiment
 - README.txt									This file

 - Source code									Implementation of the evolutionary search algorithm

 - Evolve										Experiment data using the evolutionary algorithm
 	- For each design pattern:					The results of each pattern are archived in a .zip file (~700MB when unzipped)
 		- fitness-generation.pdf				Best fitness per generation for all runs of this pattern
 		- time-generation.pdf					Execution time per generation for all runs of this pattern
 		- For each run of the experiment:
 			- results.csv						CSV with various measurements per generation
 			- success.ekt						The solution Ekeko/X template that was found, if any ; can be opened with Ekeko/X
 			- success.ekt.txt					The solution Ekeko/X template that was found, if any ; as a plain text file
 			- For each generation:
 				- population.csv 				CSV with fitness measurements of each template group, and which mutation operator was applied
 				- best.ekt						The template group with the best fitness of this generation

 - Hillclimb									Experiment data using the hillclimbing algorithm
 	- For each design pattern:
 		- For each run of the experiment:
 			- iteration-*.ekt					Template file of each iteration
 			- results.CSV 						CSV with fitness and time measurements per iteration
 			
 - Random search 								Experiment data using the random search algorithm
 	- For each design pattern:
 		- For each run of the experiment:
 			- iteration-*.ekt					Template file of each iteration
 			- results.CSV 						CSV with fitness and time measurements per iteration