def testShapeGetters(self):
with self.test_session():
x = tensor_shape.TensorShape([4])
y = tensor_shape.TensorShape([5])
bijector = SoftmaxCentered(validate_args=True)
self.assertAllEqual(y, bijector.forward_event_shape(x))
self.assertAllEqual(y.as_list(),
bijector.forward_event_shape_tensor(
x.as_list()).eval())
self.assertAllEqual(x, bijector.inverse_event_shape(y))
self.assertAllEqual(x.as_list(),
bijector.inverse_event_shape_tensor(
y.as_list()).eval())
def _compute_quantiles():
"""Helper to build quantiles."""
# Omit {0, 1} since they might lead to Inf/NaN.
zero = array_ops.zeros([], dtype=dist.dtype)
edges = math_ops.linspace(zero, 1., quadrature_size + 3)[1:-1]
# Expand edges so its broadcast across batch dims.
edges = array_ops.reshape(edges, shape=array_ops.concat([
[-1], array_ops.ones([batch_ndims], dtype=dtypes.int32)], axis=0))
quantiles = dist.quantile(edges)
quantiles = SoftmaxCentered().forward(quantiles)
# Cyclically permute left by one.
perm = array_ops.concat([
math_ops.range(1, 1 + batch_ndims), [0]], axis=0)
quantiles = array_ops.transpose(quantiles, perm)
quantiles.set_shape(_get_final_shape(quadrature_size + 1))
return quantiles
def _compute_quantiles():
"""Helper to build quantiles."""
# Omit {0, 1} since they might lead to Inf/NaN.
zero = array_ops.zeros([], dtype=dist.dtype)
edges = math_ops.linspace(zero, 1., quadrature_size + 3)[1:-1]
# Expand edges so its broadcast across batch dims.
edges = array_ops.reshape(edges, shape=array_ops.concat([
[-1], array_ops.ones([batch_ndims], dtype=dtypes.int32)], axis=0))
quantiles = dist.quantile(edges)
quantiles = SoftmaxCentered().forward(quantiles)
# Cyclically permute left by one.
perm = array_ops.concat([
math_ops.range(1, 1 + batch_ndims), [0]], axis=0)
quantiles = array_ops.transpose(quantiles, perm)
quantiles.set_shape(_get_final_shape(quadrature_size + 1))
return quantiles
def testShapeGetters(self):
with self.cached_session():
chain = Chain([
SoftmaxCentered(validate_args=True),
SoftmaxCentered(validate_args=True),
])
x = tensor_shape.TensorShape([1])
y = tensor_shape.TensorShape([2 + 1])
self.assertAllEqual(y, chain.forward_event_shape(x))
self.assertAllEqual(
y.as_list(),
chain.forward_event_shape_tensor(x.as_list()).eval())
self.assertAllEqual(x, chain.inverse_event_shape(y))
self.assertAllEqual(
x.as_list(),
chain.inverse_event_shape_tensor(y.as_list()).eval())
def testShapeGetters(self):
with self.test_session():
for x, y, b in ((tensor_shape.TensorShape([]),
tensor_shape.TensorShape([2]),
SoftmaxCentered(
event_ndims=0, validate_args=True)),
(tensor_shape.TensorShape([4]),
tensor_shape.TensorShape([5]),
SoftmaxCentered(
event_ndims=1, validate_args=True))):
self.assertAllEqual(y, b.forward_event_shape(x))
self.assertAllEqual(y.as_list(),
b.forward_event_shape_tensor(x.as_list()).eval())
self.assertAllEqual(x, b.inverse_event_shape(y))
self.assertAllEqual(x.as_list(),
b.inverse_event_shape_tensor(y.as_list()).eval())
def testShapeGetters(self):
with self.test_session():
bijector = Chain([
SoftmaxCentered(event_ndims=1, validate_args=True),
SoftmaxCentered(event_ndims=0, validate_args=True)
])
x = tensor_shape.TensorShape([])
y = tensor_shape.TensorShape([2 + 1])
self.assertAllEqual(y, bijector.forward_event_shape(x))
self.assertAllEqual(
y.as_list(),
bijector.forward_event_shape_tensor(x.as_list()).eval())
self.assertAllEqual(x, bijector.inverse_event_shape(y))
self.assertAllEqual(
x.as_list(),
bijector.inverse_event_shape_tensor(y.as_list()).eval())
def testBijectorUnknownShape(self):
with self.test_session():
softmax = SoftmaxCentered()
self.assertEqual("softmax_centered", softmax.name)
x = array_ops.placeholder(shape=[2, None], dtype=dtypes.float32)
real_x = np.log([[2., 3, 4], [4., 8, 12]])
y = array_ops.placeholder(shape=[2, None], dtype=dtypes.float32)
real_y = [[0.2, 0.3, 0.4, 0.1], [0.16, 0.32, 0.48, 0.04]]
self.assertAllClose(real_y,
softmax.forward(x).eval(feed_dict={x: real_x}))
self.assertAllClose(real_x,
softmax.inverse(y).eval(feed_dict={y: real_y}))
self.assertAllClose(-np.sum(np.log(real_y), axis=1),
softmax.inverse_log_det_jacobian(
y,
event_ndims=1).eval(feed_dict={y: real_y}),
atol=0.,
rtol=1e-7)
self.assertAllClose(-softmax.inverse_log_det_jacobian(
y, event_ndims=1).eval(feed_dict={y: real_y}),
softmax.forward_log_det_jacobian(
x,
event_ndims=1).eval(feed_dict={x: real_x}),
atol=0.,
rtol=1e-7)
def testBijectiveAndFinite(self):
with self.test_session():
softmax = SoftmaxCentered(event_ndims=1)
x = np.linspace(-50, 50, num=10).reshape(5, 2).astype(np.float32)
# Make y values on the simplex with a wide range.
y_0 = np.ones(5).astype(np.float32)
y_1 = (1e-5 * rng.rand(5)).astype(np.float32)
y_2 = (1e1 * rng.rand(5)).astype(np.float32)
y = np.array([y_0, y_1, y_2])
y /= y.sum(axis=0)
y = y.T # y.shape = [5, 3]
assert_bijective_and_finite(softmax, x, y)
def testBijectorUnknownShape(self):
with self.test_session():
softmax = SoftmaxCentered()
self.assertEqual("softmax_centered", softmax.name)
x = array_ops.placeholder(shape=[2, None], dtype=dtypes.float32)
real_x = np.log([[2., 3, 4], [4., 8, 12]])
y = array_ops.placeholder(shape=[2, None], dtype=dtypes.float32)
real_y = [[0.2, 0.3, 0.4, 0.1], [0.16, 0.32, 0.48, 0.04]]
self.assertAllClose(real_y, softmax.forward(x).eval(
feed_dict={x: real_x}))
self.assertAllClose(real_x, softmax.inverse(y).eval(
feed_dict={y: real_y}))
self.assertAllClose(
-np.sum(np.log(real_y), axis=1),
softmax.inverse_log_det_jacobian(y).eval(
feed_dict={y: real_y}),
atol=0.,
rtol=1e-7)
self.assertAllClose(
-softmax.inverse_log_det_jacobian(y).eval(
feed_dict={y: real_y}),
softmax.forward_log_det_jacobian(x).eval(
feed_dict={x: real_x}),
atol=0.,
rtol=1e-7)
def testBijectorVector(self):
with self.test_session():
softmax = SoftmaxCentered(event_ndims=1)
self.assertEqual("softmax_centered", softmax.name)
x = np.log([[2., 3, 4], [4., 8, 12]])
y = [[0.2, 0.3, 0.4, 0.1], [0.16, 0.32, 0.48, 0.04]]
self.assertAllClose(y, softmax.forward(x).eval())
self.assertAllClose(x, softmax.inverse(y).eval())
self.assertAllClose(-np.sum(np.log(y), axis=1),
softmax.inverse_log_det_jacobian(y).eval(),
atol=0.,
rtol=1e-7)
self.assertAllClose(-softmax.inverse_log_det_jacobian(y).eval(),
softmax.forward_log_det_jacobian(x).eval(),
atol=0.,
rtol=1e-7)
def testShapeGetters(self):
with self.test_session():
x = tensor_shape.TensorShape([4])
y = tensor_shape.TensorShape([5])
bijector = SoftmaxCentered(validate_args=True)
self.assertAllEqual(y, bijector.forward_event_shape(x))
self.assertAllEqual(
y.as_list(),
bijector.forward_event_shape_tensor(x.as_list()).eval())
self.assertAllEqual(x, bijector.inverse_event_shape(y))
self.assertAllEqual(
x.as_list(),
bijector.inverse_event_shape_tensor(y.as_list()).eval())
def testBijectorScalar(self):
with self.test_session():
softmax = SoftmaxCentered() # scalar by default
self.assertEqual("softmax_centered", softmax.name)
x = np.log([[2., 3, 4], [4., 8, 12]])
y = [[[2. / 3, 1. / 3], [3. / 4, 1. / 4], [4. / 5, 1. / 5]],
[[4. / 5, 1. / 5], [8. / 9, 1. / 9], [12. / 13, 1. / 13]]]
self.assertAllClose(y, softmax.forward(x).eval())
self.assertAllClose(x, softmax.inverse(y).eval())
self.assertAllClose(-np.sum(np.log(y), axis=2),
softmax.inverse_log_det_jacobian(y).eval(),
atol=0.,
rtol=1e-7)
self.assertAllClose(-softmax.inverse_log_det_jacobian(y).eval(),
softmax.forward_log_det_jacobian(x).eval(),
atol=0.,
rtol=1e-7)
def testBijectorVector(self):
with self.test_session():
softmax = SoftmaxCentered(event_ndims=1)
self.assertEqual("softmax_centered", softmax.name)
x = np.log([[2., 3, 4], [4., 8, 12]])
y = [[0.2, 0.3, 0.4, 0.1], [0.16, 0.32, 0.48, 0.04]]
self.assertAllClose(y, softmax.forward(x).eval())
self.assertAllClose(x, softmax.inverse(y).eval())
self.assertAllClose(
-np.sum(np.log(y), axis=1),
softmax.inverse_log_det_jacobian(y).eval(),
atol=0.,
rtol=1e-7)
self.assertAllClose(
-softmax.inverse_log_det_jacobian(y).eval(),
softmax.forward_log_det_jacobian(x).eval(),
atol=0.,
rtol=1e-7)
def testBijectorScalar(self):
with self.test_session():
softmax = SoftmaxCentered() # scalar by default
self.assertEqual("softmax_centered", softmax.name)
x = np.log([[2., 3, 4],
[4., 8, 12]])
y = [[[2. / 3, 1. / 3],
[3. / 4, 1. / 4],
[4. / 5, 1. / 5]],
[[4. / 5, 1. / 5],
[8. / 9, 1. / 9],
[12. / 13, 1. / 13]]]
self.assertAllClose(y, softmax.forward(x).eval())
self.assertAllClose(x, softmax.inverse(y).eval())
self.assertAllClose(
-np.sum(np.log(y), axis=2),
softmax.inverse_log_det_jacobian(y).eval(),
atol=0.,
rtol=1e-7)
self.assertAllClose(
-softmax.inverse_log_det_jacobian(y).eval(),
softmax.forward_log_det_jacobian(x).eval(),
atol=0.,
rtol=1e-7)
