private static void compareNormToGeo() {
FlierCatcher gameGeo = new FlierCatcher(size);
makeModelerGeoNet(gameGeo, trainTurns, new int[] {5});
System.out.println("Learning Geo Net");
gameGeo.modeler.learnFromMemory(0.3, 0.5, 0, false, learnIterations, 10000);
FlierCatcher gameNorm = new FlierCatcher(size);
gameNorm.modeler =
TransferTestUtils.loadOrCreate(
null, gameGeo, trainTurns, new int[] {size * size * 3}, null);
System.out.println("Learning Normal Net");
gameNorm.modeler.learnFromMemory(0.3, 0.5, 0, false, learnIterations, 10000);
TransferTestUtils.compareTwoModelers(gameNorm, 100, gameNorm.modeler, gameGeo.modeler, false);
}
public static ModelLearnerHeavy makeModelerGeoNet(
FlierCatcher game, int turns, int[] hiddenPerOutput) {
game.modeler =
TransferTestUtils.loadOrCreate(null, game, trainTurns, new int[] {size * size * 3}, null);
game.modeler.learnFromMemory(0, 0, 0, false, 1, 10000);
RelationManager relMngr = RelationManager.createFromGridGamePredictor(game, game.modeler);
FlierCatcher.trainModeler(
game.modeler, turns, game, 0, game.actionChoices, GridExploreGame.actionTranslator);
FFNeuralNetwork ann = game.modeler.getModelVTA().getNeuralNetwork();
ModularizationUtils.initializeANNOnlyConnectingRelatedVars(ann, relMngr, hiddenPerOutput);
return game.modeler;
}