def testScalarCongruency(self):
bijector_test_util.assert_scalar_congruency(tfb.GumbelCDF(loc=0.3,
scale=20.),
lower_x=1.,
upper_x=100.,
eval_func=self.evaluate,
rtol=0.05)
def testVariableScale(self):
x = tf.Variable(1.)
b = tfb.GumbelCDF(loc=0., scale=x, validate_args=True)
self.evaluate(x.initializer)
self.assertIs(x, b.scale)
self.assertEqual((), self.evaluate(b.forward(-3.)).shape)
with self.assertRaisesOpError("Argument `scale` must be positive."):
with tf.control_dependencies([x.assign(-1.)]):
self.evaluate(b.forward(-3.))
def testBijectiveAndFinite(self):
bijector = tfb.GumbelCDF(loc=0., scale=3.0, validate_args=True)
x = np.linspace(-10., 10., num=10).astype(np.float32)
y = np.linspace(0.01, 0.99, num=10).astype(np.float32)
bijector_test_util.assert_bijective_and_finite(bijector,
x,
y,
eval_func=self.evaluate,
event_ndims=0,
rtol=1e-3)
def testBijector(self):
loc = 0.3
scale = 5.
bijector = tfb.GumbelCDF(loc=loc, scale=scale, validate_args=True)
self.assertStartsWith(bijector.name, "gumbel")
x = np.array([[[-3.], [0.], [0.5], [4.2], [12.]]], dtype=np.float32)
# Gumbel distribution
gumbel_dist = stats.gumbel_r(loc=loc, scale=scale)
y = gumbel_dist.cdf(x).astype(np.float32)
self.assertAllClose(y, self.evaluate(bijector.forward(x)))
self.assertAllClose(x, self.evaluate(bijector.inverse(y)))
self.assertAllClose(
np.squeeze(gumbel_dist.logpdf(x), axis=-1),
self.evaluate(bijector.forward_log_det_jacobian(x, event_ndims=1)))
self.assertAllClose(
self.evaluate(-bijector.inverse_log_det_jacobian(y, event_ndims=1)),
self.evaluate(bijector.forward_log_det_jacobian(x, event_ndims=1)),
rtol=1e-4,
atol=0.)