The GEP Framework - Alpha Edition

How to use this version:

First, compile the program using the Make utility (makefile included):

$ make

Then, run the program by using the GEPMain class:

$ java GEPMain

This launches the main GUI interface for the program. In some cases, the output window may not be visible (the right side, which should appear as a textarea). If this happens, resize the window, and it should immediately set itself to be visible.

Once running, many settings are available to be set. How each option works will be explained in the next section, but for now, it is recommended that the user try one of the example datasets, included in the 'dataset' directory.

To load these, go to File->Open, and in the main directory, test001.config and test002.config should be visible. Load either of these. To ensure that the dataset is loaded correctly, make sure you reload the directory by pressing the "Load DataSet" button, and selecting the appropriate dataset in the dataset directory (iris.data for test001.config, and letter-recognition.data for test002.config).

Press Launch, at the bottom, to see that these are working correctly.

Using the configuration editor:

Title: Optional piece of information to give a name to a specific test.

DataSet: The dataset being loaded to train the classifier on. Use the "Load DataSet" button to load a new file.

DataSet Loader: The class being used to turn the dataset file into a DataSet object in the evolver. This can be the built in "NumericalDataSetLoader", or a custom one built by the user which meets the DataSetLoader interface.

Parameters: What parameters should be passed into the DataSet Loader.

Number of Inputs: How many inputs the DataSet Loader should be told exist in the dataset.

Number of Classes: How many potential classes the dataset is classified into.

Training Percentage: From 0.1 to 1.0, what percentage of the dataset should be used for training (the rest saved in a test-set).

Number of Runs: How many runs of the evolver should be done.

Number of Generations: How many generations each run should have.

Population Size: Size of population the genetic algorithm will have.

Functions: The list of functions the classifier neural network can use. To remove a function, click on the big X button beside the name. To give a function parameters, simple fill in the text under the function. To add a new function, click the "Add Function" button, and select one from the list, or load an external function.

Head Size: The size of the head for each karva-string for each node (see accompanying paper for more details on head size, and GEP algorithm).

Number of RNCs: The number of random numerical constants in each node (see paper for more details).

Number of Layers: Number of layers in the neural network.

Layers: Number of nodes in each layer of the network.

Fitness Process: Which fitness process is used during the genetic algorithm. Parameters can be added by filling in the text field under the name. Loading a different fitness process can be done by pressing the Load Fitness Process button.

Selection Process: Similar to the Fitness Process, but used to choose the selection process.

Mutators: List of mutators used by the algorithm, with a weight for how often they should be used. Like with the functions, the X button can remove a mutator, and the Add Mutator button will add a new one.

Mutation Rate: From 0.1 or higher (where 0.1 indicates 10%), what percentage of individuals should have mutations applied to them during the algorithm.

Crossovers: Similar to the Mutators list, but loads which crossover methods to use.

Evolver State Processes: Start, End, Start of Run, End of Run, Before Generation, After Generation. For each of these times, loads a list of Evolver State Processes to be run at that time. Can be given parameters, and removed with the X button.

Launch: Initiates the GEP Framework, outputting things to the output window on the right side.

Stop: Stops the currently running GEP Framework process.