} // pure static class
/// <summary>
/// Raw-level ARIMA forecasting function.
/// </summary>
/// <param name="data"> UNMODIFIED, list of double numbers representing time-series with constant time-gap </param>
/// <param name="forecastSize"> integer representing how many data points AFTER the data series to be
/// forecasted </param>
/// <param name="params"> ARIMA parameters </param>
/// <returns> a ForecastResult object, which contains the forecasted values and/or error message(s) </returns>
//JAVA TO C# CONVERTER WARNING: 'final' parameters are not available in .NET:
//ORIGINAL LINE: public static TimeSeries.Forecast.TimeSeries.Arima.struct.ForecastResult forecast_arima(final double[] data, final int forecastSize, TimeSeries.Forecast.TimeSeries.Arima.struct.ArimaParams params)
public static ForecastResult forecast_arima(double[] data, int forecastSize, ArimaParams @params)
{
try
{
int p = @params.p;
int d = @params.d;
int q = @params.q;
int P = @params.P;
int D = @params.D;
int Q = @params.Q;
int m = @params.m;
ArimaParams paramsForecast = new ArimaParams(p, d, q, P, D, Q, m);
ArimaParams paramsXValidation = new ArimaParams(p, d, q, P, D, Q, m);
// estimate ARIMA model parameters for forecasting
ArimaModel fittedModel = ArimaSolver.estimateARIMA(paramsForecast, data, data.Length, data.Length + 1);
// compute RMSE to be used in confidence interval computation
double rmseValidation = ArimaSolver.computeRMSEValidation(data, ForecastUtil.testSetPercentage, paramsXValidation);
fittedModel.RMSE = rmseValidation;
ForecastResult forecastResult = fittedModel.Forecast(forecastSize);
// populate confidence interval
forecastResult.Sigma2AndPredicationInterval = fittedModel.Params;
// add logging messages
forecastResult.Log("{" + "\"Best ModelInterface Param\" : \"" + fittedModel.Params.summary() + "\"," + "\"Forecast Size\" : \"" + forecastSize + "\"," + "\"Input Size\" : \"" + data.Length + "\"" + "}");
// successfully built ARIMA model and its forecast
return(forecastResult);
}
catch (Exception ex)
{
// failed to build ARIMA model
throw new Exception("Failed to build ARIMA forecast: " + ex.Message);
}
}
/// <summary>
/// Estimate ARMA(p,q) parameters, i.e. AR-parameters: \phi_1, ... , \phi_p
/// MA-parameters: \theta_1, ... , \theta_q
/// Input data is assumed to be stationary, has zero-mean, aligned, and imputed
/// </summary>
/// <param name="data_orig"> original data </param>
/// <param name="params"> ARIMA parameters </param>
/// <param name="forecast_length"> forecast length </param>
/// <param name="maxIteration"> maximum number of iteration </param>
//JAVA TO C# CONVERTER WARNING: 'final' parameters are not available in .NET:
//ORIGINAL LINE: public static void estimateARMA(final double[] data_orig, final TimeSeries.Forecast.TimeSeries.Arima.struct.ArimaParams params, final int forecast_length, final int maxIteration)
public static void estimateARMA(double[] data_orig, ArimaParams @params, int forecast_length, int maxIteration)
{
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[] data = new double[data_orig.length];
double[] data = new double[data_orig.Length];
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final int total_length = data.length;
int total_length = data.Length;
Array.Copy(data_orig, 0, data, 0, total_length);
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final int r = (params.getDegreeP() > params.getDegreeQ()) ? 1 + params.getDegreeP() : 1 + params.getDegreeQ();
int r = (@params.DegreeP > @params.DegreeQ) ? 1 + @params.DegreeP : 1 + @params.DegreeQ;
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final int length = total_length - forecast_length;
int length = total_length - forecast_length;
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final int size = length - r;
int size = length - r;
if (length < 2 * r)
{
throw new Exception("not enough data points: length=" + length + ", r=" + r);
}
// step 1: apply Yule-Walker method and estimate AR(r) model on input data
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[] errors = new double[length];
double[] errors = new double[length];
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[] yuleWalkerParams = applyYuleWalkerAndGetInitialErrors(data, r, length, errors);
double[] yuleWalkerParams = applyYuleWalkerAndGetInitialErrors(data, r, length, errors);
for (int j = 0; j < r; ++j)
{
errors[j] = 0;
}
// step 2: iterate Least-Square fitting until the parameters converge
// instantiate Z-matrix
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[][] matrix = new double[params.getNumParamsP() + params.getNumParamsQ()][size];
//JAVA TO C# CONVERTER NOTE: The following call to the 'RectangularArrays' helper class reproduces the rectangular array initialization that is automatic in Java:
//ORIGINAL LINE: double[][] matrix = new double[params.NumParamsP + params.NumParamsQ][size];
double[][] matrix = RectangularArrays.ReturnRectangularDoubleArray(@params.NumParamsP + @params.NumParamsQ, size);
double bestRMSE = -1; // initial value
int remainIteration = maxIteration;
InsightsVector bestParams = null;
while (--remainIteration >= 0)
{
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final TimeSeries.Forecast.matrix.InsightsVector estimatedParams = iterationStep(params, data, errors, matrix, r, length, size);
InsightsVector estimatedParams = iterationStep(@params, data, errors, matrix, r, length, size);
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final TimeSeries.Forecast.matrix.InsightsVector originalParams = params.getParamsIntoVector();
InsightsVector originalParams = @params.ParamsIntoVector;
@params.ParamsFromVector = estimatedParams;
// forecast for validation data and compute RMSE
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[] forecasts = ArimaSolver.forecastARMA(params, data, length, data.length);
double[] forecasts = ArimaSolver.forecastARMA(@params, data, length, data.Length);
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double anotherRMSE = ArimaSolver.computeRMSE(data, forecasts, length, 0, forecast_length);
double anotherRMSE = ArimaSolver.ComputeRMSE(data, forecasts, length, 0, forecast_length);
// update errors
//JAVA TO C# CONVERTER WARNING: The original Java variable was marked 'final':
//ORIGINAL LINE: final double[] train_forecasts = ArimaSolver.forecastARMA(params, data, r, data.length);
double[] train_forecasts = ArimaSolver.forecastARMA(@params, data, r, data.Length);
for (int j = 0; j < size; ++j)
{
errors[j + r] = data[j + r] - train_forecasts[j];
}
if (bestRMSE < 0 || anotherRMSE < bestRMSE)
{
bestParams = estimatedParams;
bestRMSE = anotherRMSE;
}
}
@params.ParamsFromVector = bestParams;
}