def testShapesAreCorrect(self):
k = psd_kernels.ExpSinSquared(amplitude=1., length_scale=1., period=3.)
x = np.ones([4, 3], np.float32)
y = np.ones([5, 3], np.float32)
self.assertAllEqual([4, 5], k.matrix(x, y).shape)
self.assertAllEqual(
[2, 4, 5],
k.matrix(tf.stack([x]*2), tf.stack([y]*2)).shape)
k = psd_kernels.ExpSinSquared(
amplitude=np.ones([2, 1, 1], np.float32),
length_scale=np.ones([1, 3, 1], np.float32),
period=np.ones([2, 1, 1, 1], np.float32))
self.assertAllEqual(
[2, 2, 3, 2, 4, 5],
#`-----' | `--'
# | | `- matrix shape
# | `- from input batch shapes
# `- from broadcasting kernel params
k.matrix(
tf.stack([x]*2), # shape [2, 4, 3]
tf.stack([y]*2) # shape [2, 5, 3]
).shape)
def testValidateArgs(self):
with self.assertRaises(tf.errors.InvalidArgumentError):
k = psd_kernels.ExpSinSquared(
amplitude=-1., length_scale=-1., period=-1., validate_args=True)
self.evaluate(k.amplitude)
if not tf.executing_eagerly():
with self.assertRaises(tf.errors.InvalidArgumentError):
self.evaluate(k.length_scale)
with self.assertRaises(tf.errors.InvalidArgumentError):
self.evaluate(k.period)
# But `None`'s are ok
k = psd_kernels.ExpSinSquared(
amplitude=None, length_scale=None, period=None, validate_args=True)
self.evaluate(k.apply([1.], [1.]))
def testShapesAreCorrect(self):
k = psd_kernels.ExpSinSquared(amplitude=1., length_scale=1., period=3.)
x = np.ones([4, 3], np.float32)
y = np.ones([5, 3], np.float32)
self.assertAllEqual(k.matrix(x, y).shape, [4, 5])
self.assertAllEqual(
k.matrix(tf.stack([x] * 2), tf.stack([y] * 2)).shape, [2, 4, 5])
k = psd_kernels.ExpSinSquared(amplitude=np.ones([2, 1, 1], np.float32),
length_scale=np.ones([1, 3, 1],
np.float32),
period=np.ones([2, 1, 1, 1], np.float32))
self.assertAllEqual(
k.matrix(
tf.stack([x] * 2), # shape [2, 4, 3]
tf.stack([y] * 2) # shape [2, 5, 3]
).shape,
[2, 2, 3, 2, 4, 5])
def testValuesAreCorrect(self, feature_ndims, dtype, dims):
amplitude = np.array(5., dtype=dtype)
period = np.array(1., dtype=dtype)
length_scale = np.array(.2, dtype=dtype)
np.random.seed(42)
k = psd_kernels.ExpSinSquared(amplitude=amplitude,
length_scale=length_scale,
period=period,
feature_ndims=feature_ndims)
shape = [dims] * feature_ndims
for _ in range(5):
x = np.random.uniform(-3., 3., size=shape).astype(dtype)
y = np.random.uniform(-3., 3., size=shape).astype(dtype)
self.assertAllClose(self.evaluate(k.apply(x, y)),
self._exp_sin_squared_kernel(
amplitude, length_scale, period, x, y),
rtol=1e-4)
def testNoneShapes(self):
k = psd_kernels.ExpSinSquared(amplitude=np.reshape([1.] * 6, [3, 2]))
self.assertEqual([3, 2], k.batch_shape.as_list())
def testMismatchedFloatTypesAreBad(self):
psd_kernels.ExpSinSquared(1, 1) # Should be OK (float32 fallback).
psd_kernels.ExpSinSquared(1, np.float64(1)) # Should be OK.
with self.assertRaises(TypeError):
psd_kernels.ExpSinSquared(1, np.float64(1),
np.float32(1)) # Should fail.
