public virtual void differentIntervalsTest()
{
DoubleArray xValues = DoubleArray.of(1.0328724558967068, 1.2692381049172323, 2.8611430465380905, 4.296118458251132, 7.011992052151352, 7.293354144919639, 8.557971037612713, 8.77306861567384, 10.572470371584489, 12.96945799507056);
DoubleArray[] yValues = new DoubleArray[] { DoubleArray.of(0.11593075755231343, 0.2794957672828094, 0.4674733634811079, 0.5517689918508841, 0.6138447304104604, 0.6264375977142906, 0.6581666492568779, 0.8378685055774037, 0.10005246918325483, 0.10468304334744241), DoubleArray.of(0.995780079114617, 0.8733013195721913, 0.8192165283188197, 0.6539369493529048, 0.63868683960757515, 0.4700471352238411, 0.4555354921077598, 0.3780781869340659, 0.2299369456202763, 0.9182441378327986) };
int nData = xValues.size();
int nKeys = 100 * nData;
double[] xKeys = new double[nKeys];
double xMin = 0.0;
double xMax = xValues.get(nData - 1) + 2d;
double step = (xMax - xMin) / nKeys;
for (int i = 0; i < nKeys; ++i)
{
xKeys[i] = xMin + step * i;
}
int yDim = yValues.Length;
for (int k = 0; k < yDim; ++k)
{
BoundCurveInterpolator bci = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, yValues[k], DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
// Check C0 continuity
assertEquals(bci.interpolate(xValues.get(0) - TOL), bci.interpolate(xValues.get(0)), TOL * 1.0e2);
// Check C1 continuity
assertEquals(bci.firstDerivative(xValues.get(0) - TOL), bci.firstDerivative(xValues.get(0)), Math.Sqrt(TOL));
// Test sensitivity
double[] yValues1Up = yValues[k].toArray();
double[] yValues1Dw = yValues[k].toArray();
for (int j = 0; j < nData; ++j)
{
yValues1Up[j] = yValues[k].get(j) * (1d + EPS);
yValues1Dw[j] = yValues[k].get(j) * (1d - EPS);
BoundCurveInterpolator bciUp = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, DoubleArray.ofUnsafe(yValues1Up), DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
BoundCurveInterpolator bciDw = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, DoubleArray.ofUnsafe(yValues1Dw), DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
for (int i = 0; i < nKeys; ++i)
{
double res1 = 0.5 * (bciUp.interpolate(xKeys[i]) - bciDw.interpolate(xKeys[i])) / EPS / yValues[k].get(j);
assertEquals(res1, bci.parameterSensitivity(xKeys[i]).get(j), Math.Max(Math.Abs(yValues[k].get(j)) * EPS, EPS) * 1e2); //because gradient is NOT exact
}
yValues1Up[j] = yValues[k].get(j);
yValues1Dw[j] = yValues[k].get(j);
}
}
}
public virtual void sameIntervalsTest()
{
DoubleArray xValues = DoubleArray.of(1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0);
DoubleArray[] yValues = new DoubleArray[] { DoubleArray.of(0.001, 0.001, 0.001, 0.001, 0.001, 0.001, 0.001, 0.001), DoubleArray.of(-11.0, 8.0, 5.0, 1.001, -1.001, 5.0, -8.0, 11.0), DoubleArray.of(0.001, -0.001, 5.0, 9.0, 9.0, 12.0, 18.0, 18.0) };
int nData = xValues.size();
int nKeys = 100 * nData;
double[] xKeys = new double[nKeys];
double xMin = 0.0;
double xMax = xValues.get(nData - 1) + 2d;
double step = (xMax - xMin) / nKeys;
for (int i = 0; i < nKeys; ++i)
{
xKeys[i] = xMin + step * i;
}
int yDim = yValues.Length;
for (int k = 0; k < yDim; ++k)
{
BoundCurveInterpolator bci = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, yValues[k], DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
// Check C0 continuity
assertEquals(bci.interpolate(xValues.get(0) - TOL), bci.interpolate(xValues.get(0)), TOL * 1.0e2);
// Check C1 continuity
assertEquals(bci.firstDerivative(xValues.get(0) - TOL), bci.firstDerivative(xValues.get(0)), Math.Sqrt(TOL));
// Test sensitivity
double[] yValues1Up = yValues[k].toArray();
double[] yValues1Dw = yValues[k].toArray();
for (int j = 0; j < nData; ++j)
{
yValues1Up[j] = yValues[k].get(j) * (1d + EPS);
yValues1Dw[j] = yValues[k].get(j) * (1d - EPS);
BoundCurveInterpolator bciUp = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, DoubleArray.ofUnsafe(yValues1Up), DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
BoundCurveInterpolator bciDw = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(xValues, DoubleArray.ofUnsafe(yValues1Dw), DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, PRODUCT_LINEAR);
for (int i = 2; i < nKeys; ++i)
{
double exp = 0.5 * (bciUp.interpolate(xKeys[i]) - bciDw.interpolate(xKeys[i])) / EPS / yValues[k].get(j);
assertEquals(bci.parameterSensitivity(xKeys[i]).get(j), exp, Math.Max(Math.Abs(yValues[k].get(j)) * EPS, EPS) * 1e3); //because gradient is NOT exact TODO
}
yValues1Up[j] = yValues[k].get(j);
yValues1Dw[j] = yValues[k].get(j);
}
}
}
//-------------------------------------------------------------------------
public virtual void e2eTest()
{
BoundCurveInterpolator bciZero = PRODUCT_NATURAL_SPLINE_MONOTONE_CUBIC.bind(X_VALUES, Y_VALUES, DISCOUNT_FACTOR_QUADRATIC_LEFT_ZERO_RATE, DISCOUNT_FACTOR_LINEAR_RIGHT_ZERO_RATE);
BoundCurveInterpolator bciDf = LOG_NATURAL_SPLINE_MONOTONE_CUBIC.bind(X_VALUES, DSC_VALUES, QUADRATIC_LEFT, LINEAR);
int nKeys = 170;
for (int i = 0; i < nKeys; ++i)
{
double key = -0.1d + 0.05d * i;
double zero = bciZero.interpolate(key);
double df = bciDf.interpolate(key);
assertEquals(Math.Exp(-key * zero), df, TOL_E2E);
double zeroGrad = bciZero.firstDerivative(key);
double dfGrad = bciDf.firstDerivative(key);
assertEquals(-zero * df - key * df * zeroGrad, dfGrad, TOL_E2E);
DoubleArray zeroSensi = bciZero.parameterSensitivity(key);
DoubleArray dfSensi = bciDf.parameterSensitivity(key);
for (int j = 0; j < X_VALUES.size(); ++j)
{
assertEquals(key * df * zeroSensi.get(j) / (X_VALUES.get(j) * DSC_VALUES.get(j)), dfSensi.get(j), TOL_E2E * 10d);
}
}
}
