def testSplineCurveInverseIsCorrect(self):
"""Tests that the inverse curve is indeed the inverse of the curve."""
x_knot = np.arange(0, 16, 0.01, dtype=np.float64)
with self.session():
alpha = distribution.inv_partition_spline_curve(x_knot).eval()
x_recon = distribution.partition_spline_curve(alpha).eval()
self.assertAllClose(x_recon, x_knot)
def testSplineCurveIsC1Smooth(self):
"""Tests that partition_spline_curve() and its derivative are continuous."""
x1 = np.linspace(0., 8., 10000, dtype=np.float64)
x2 = x1 + 1e-7
x1 = tf.convert_to_tensor(x1)
x2 = tf.convert_to_tensor(x2)
with tf.GradientTape(persistent=True) as tape:
tape.watch(x1)
tape.watch(x2)
y1 = distribution.partition_spline_curve(x1)
y2 = distribution.partition_spline_curve(x2)
dy1 = tape.gradient(tf.reduce_sum(y1), x1)
dy2 = tape.gradient(tf.reduce_sum(y2), x2)
self.assertAllClose(y1, y2)
self.assertAllClose(dy1, dy2)
def testSplineCurveIsC1Smooth(self):
"""Tests that partition_spline_curve() and its derivative are continuous."""
x1 = np.linspace(0., 8., 10000, dtype=np.float64)
x2 = x1 + 1e-7
x_ph = tf.placeholder(x1.dtype, len(x1))
splinefun_ph = distribution.partition_spline_curve(x_ph)
with self.session() as sess:
y1, (dy1,) = sess.run(
(splinefun_ph, tf.gradients(tf.reduce_sum(splinefun_ph), (x_ph))),
feed_dict={x_ph: x1})
y2, (dy2,) = sess.run(
(splinefun_ph, tf.gradients(tf.reduce_sum(splinefun_ph),
(x_ph))), {x_ph: x2})
self.assertAllClose(y1, y2)
self.assertAllClose(dy1, dy2)
