public static double Area(NumSharp.NDArray pointSet)
{
int d = pointSet.size / 3;
double area = 0;
for (int i = 0; i < d - 1; i++)
{
area += pointSet[i, 0] * pointSet[i + 1, 1] - pointSet[i + 1, 0] * pointSet[i, 1];
}
area += pointSet[d - 1, 0] * pointSet[0, 1] - pointSet[0, 0] * pointSet[d - 1, 1];
return(0.5 * Math.Abs(area));
}
public void SquaredDifference_1D()
{
// Calcute the gradient of (x1-x2)^2
// by Automatic Differentiation in Eager mode
// Expected is 2*(abs(x1-x2))
Tensor x1 = new NumSharp.NDArray(new float[] { 1, 3, 5, 21, 19, 17 });
Tensor x2 = new NumSharp.NDArray(new float[] { 29, 27, 23, 7, 11, 13 });
float[] expected = new float[] {
(29 - 1) * 2,
(27 - 3) * 2,
(23 - 5) * 2,
(7 - 21) * 2,
(11 - 19) * 2,
(13 - 17) * 2
};
// Sanity check
using (var tape = tf.GradientTape())
{
tape.watch(x1);
tape.watch(x2);
var loss = tf.multiply((x1 - x2), (x1 - x2));
var result = tape.gradient(loss, x2);
CollectionAssert.AreEqual(result.ToArray <float>(), expected);
}
// Actual test
using (var tape = tf.GradientTape())
{
tape.watch(x1);
tape.watch(x2);
var loss = tf.squared_difference(x1, x2);
var result = tape.gradient(loss, x2);
CollectionAssert.AreEqual(result.ToArray <float>(), expected);
}
}
