CODEWITHSUNDEEP
javaneural-networkencog
Pulkownik
Pulkownik

Reputation: 529

Am i doing it in right wat? predict stock price

I prepared csv file with the input data for neural network, and csv file where i can test my neural network. The results are not satisfactory. I was trying increase/decrease size of input data. Probably i missing something and i would be glad if someone can some tips etc. Here is my encog code:

    //input data
    File file = new File("path to file");
    CSVFormat format = new CSVFormat('.', ',');
    VersatileDataSource source = new CSVDataSource(file, false, format);
    VersatileMLDataSet data = new VersatileMLDataSet(source);
    data.getNormHelper().setFormat(format);

    ColumnDefinition wig20OpenN = data.defineSourceColumn("wig20OpenN", 0, ColumnType.continuous);
    (...)

    ColumnDefinition futureClose = data.defineSourceColumn("futureClose", 81, ColumnType.continuous);

    data.analyze();
    data.defineSingleOutputOthersInput(futureClose);
    EncogModel model = new EncogModel(data);

    //TYPE_RBFNETWORK, TYPE_SVM, TYPE_NEAT, TYPE_FEEDFORWARD <- this type of method i was trying
    model.selectMethod(data, MLMethodFactory.TYPE_SVM);
    model.setReport(new ConsoleStatusReportable());

    data.normalize();

    model.holdBackValidation(0.001, true, 10);
    model.selectTrainingType(data);
    MLRegression bestMethod = (MLRegression)model.crossvalidate(20, true);
    // Display the training and validation errors.
    System.out.println( "Training error: " + model.calculateError(bestMethod, model.getTrainingDataset()));
    System.out.println( "Validation error: " + model.calculateError(bestMethod, model.getValidationDataset()));
    NormalizationHelper helper = data.getNormHelper();

    File testingData = new File("path to testing file");
    ReadCSV csv = new ReadCSV(testingData, false, format);
    String[] line = new String[81];
    MLData input = helper.allocateInputVector();
    while(csv.next()) {
            StringBuilder result = new StringBuilder();

            for(int i = 0; i <81; i++){
            line[i] = csv.get(i);
        }

            String correct = csv.get(81);
            helper.normalizeInputVector(line,input.getData(),false);
            MLData output = bestMethod.compute(input);
            String irisChosen = helper.denormalizeOutputVectorToString(output)[0];

            result.append(Arrays.toString(line));
            result.append(" -> predicted: ");
            result.append(irisChosen);
            result.append("(correct: ");
            result.append(correct);
            result.append(")");

            System.out.println(result.toString());
        }

        // Delete data file and shut down.
        filename.delete();
        Encog.getInstance().shutdown();

What i was trying so far is to change the MLMethodFactory, but had problems here, only TYPE_RBFNETWORK, TYPE_SVM, TYPE_NEAT, TYPE_FEEDFORWARD this type works fine, for example if i changed it to TYPE_PNN i had following exception:

Exception in thread "main" org.encog.EncogError: Please call selectTraining first to choose how to train.

Ok i know from documentation that i should use this method:

selectTraining(VersatileMLDataSet dataset, String trainingType, String trainingArgs)

But the string type for traningtype and triningArgs is confusing.

And last question what about saving the neural after traning to file, and loading it to check on the traning data? As i would like to have this separately.

Edit: I forgot the size of the input data is 1500.

Upvotes: 0

Views: 1082

Answers (1)

Yuriy Zaletskyy
Yuriy Zaletskyy

Reputation: 5151

I see that you not satisfied with your results, but it is relatively fine. I propose you to consider adding scaling to your training. You have 81 columns, and in your input row I see data like 16519.1600, also 2315.94, and even -0.6388282285709328. For neural network it is hard to adjust weights correctly for such different inputs.
P.S. scaling is also normalizing of columns!. As usually in books is described normalizing of rows, but normalizing of columns is also important.

Upvotes: 1

Related Questions