public static void useNeuralNetwork() {
List<Action> actions = new ArrayList<>();
actions.add(new TradingAction(0, 0, TradingAction.Type.NOTHING));
int tradeActionsCount = 1;
for (int i = 0; i < tradeActionsCount; i++) {
actions.add(new TradingAction(i + 1, (int) Math.pow(2, i), TradingAction.Type.BUY));
}
for (int i = 0; i < tradeActionsCount; i++) {
actions.add(
new TradingAction(
i + 1 + tradeActionsCount, (int) Math.pow(2, i), TradingAction.Type.SELL));
}
double learningRate = 0.0001;
double discountFactor = 0.5;
double randomFactor = 0.8;
int sizeOfSeenVector = 5;
List<Feature> features = new ArrayList<>();
features.add(new MoneyFeature());
features.add(new CommodityFeature());
features.add(new IsActionDoable());
for (int i = 0; i < sizeOfSeenVector; i++) {
features.add(new DeltaPriceFeature(i));
}
FeedForwardHeuristic heuristic = new FeedForwardHeuristic();
heuristic.learningRate = 0.0000001;
heuristic.l2RegularizationConstant = 0.0;
heuristic.dynamicKillingWeights = false;
heuristic.momentum = 0.000;
heuristic.BOOSTING_ETA_COEFF = 0.0;
heuristic.dynamicBoostingEtas = false;
OnlineFeedForwardNetwork neuralNetwork =
new OnlineFeedForwardNetwork(
"asdf",
Arrays.asList(/*features.size(), */ features.size(), features.size(), 1),
Arrays.asList(
/*new ParametrizedHyperbolicTangens(),*/ new ParametrizedHyperbolicTangens(),
new LinearFunction()),
new GaussianRndWeightsFactory(0.1, 0),
heuristic);
OnlineFeedForwardNetwork neuralNetwork2 =
new OnlineFeedForwardNetwork(
"asdf2",
Arrays.asList(/*features.size(),*/ features.size(), features.size(), 1),
Arrays.asList(
/*new ParametrizedHyperbolicTangens(),*/ new ParametrizedHyperbolicTangens(),
new LinearFunction()),
new GaussianRndWeightsFactory(0.1, 0),
heuristic);
Trading trading =
new Trading(
actions,
new Random(0),
learningRate,
discountFactor,
randomFactor,
new NonLinearStateEvaluatorBasedOnNeuralNetwork(features, neuralNetwork),
new NonLinearStateEvaluatorBasedOnNeuralNetwork(features, neuralNetwork2),
10,
0,
sizeOfSeenVector,
new ParameterizedSinus(0, 2, 1, 1),
0);
/*trading.run(1000, 1000, 0);
trading.setRandomFactor(0.5);
trading.initialize(10, 0, 0);
trading.run(10000, 1000, 0);*/
for (int i = 0; i < 1000; i++) {
QLearning.LOGGER.info("------------------------------------------------------");
trading.setLearningRate(0.001);
trading.setDiscountFactor(0.9);
trading.setMode(QLearning.Mode.TRAINING);
randomFactor = Math.pow(Math.E, (double) -i / 20.0) / 3;
QLearning.LOGGER.info("Random Factor: {}", randomFactor);
trading.setRandomFactor(randomFactor);
trading.initialize(10, 0, 0);
trading.run(10000, 1000, i);
int startingMoney = 10;
int startingCommodity = 0;
trading.setRandomFactor(0.0);
trading.setMode(QLearning.Mode.TESTING);
trading.initialize(startingMoney, startingCommodity, 0);
trading.run(1, 1000);
TradingState state = trading.getLastState();
QLearning.LOGGER.info(
"Money {}, Commodity: {}",
state.getMoney() - startingMoney,
state.getCommodity() - startingCommodity);
// try {
// System.in.read();
// } catch (IOException e) {
// e.printStackTrace();
// }
/*trading.reseLearningAtributes();
trading.initialize(startingMoney, startingCommodity, 1);
trading.run(1, 100100);
state = trading.getLastState();
QLearning.LOGGER.info("Money {}, Commodity: {}", state.getMoney() - startingMoney, state.getCommodity() - startingCommodity);
trading.reseLearningAtributes();
trading.initialize(startingMoney, startingCommodity, 2);
trading.run(1, 101000);
state = trading.getLastState();
QLearning.LOGGER.info("Money {}, Commodity: {}", state.getMoney() - startingMoney, state.getCommodity() - startingCommodity);*/
// QLearning.LOGGER.info("Weights {}", neuralNetwork.getWeights());
}
QLearning.LOGGER.info("PRODUCTION");
/*trading.reseLearningAtributes();
trading.initialize(15, 5, 3);
trading.run(1, 10000);
trading.reseLearningAtributes();
trading.initialize(15, 5, 4);
trading.run(1, 10000);
trading.reseLearningAtributes();
trading.initialize(15, 5, 5);
trading.run(1, 10000);*/
}
