def testRollStatic(self):
with self.test_session():
with self.assertRaisesRegexp(ValueError, "None values not supported."):
distribution_util.rotate_transpose(None, 1)
for x in (np.ones(1), np.ones((2, 1)), np.ones((3, 2, 1))):
for shift in np.arange(-5, 5):
y = distribution_util.rotate_transpose(x, shift)
self.assertAllEqual(self._np_rotate_transpose(x, shift), y.eval())
self.assertAllEqual(np.roll(x.shape, shift), y.get_shape().as_list())
def testRollStatic(self):
with self.test_session():
with self.assertRaisesRegexp(ValueError,
"None values not supported."):
distribution_util.rotate_transpose(None, 1)
for x in (np.ones(1), np.ones((2, 1)), np.ones((3, 2, 1))):
for shift in np.arange(-5, 5):
y = distribution_util.rotate_transpose(x, shift)
self.assertAllEqual(self._np_rotate_transpose(x, shift),
y.eval())
self.assertAllEqual(np.roll(x.shape, shift),
y.get_shape().as_list())
def make_batch_of_event_sample_matrices(
self, x, name="make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from S+B+E to B_+E_+S_.
Where:
- `B_ = B if B else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
Args:
x: `Tensor`.
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
"""
with self._name_scope(name, values=[x]):
x = ops.convert_to_tensor(x, name="x")
sample_shape, batch_shape, event_shape = self.get_shape(x)
event_shape = distribution_util.pick_vector(
self._event_ndims_is_0, (1, ), event_shape)
batch_shape = distribution_util.pick_vector(
self._batch_ndims_is_0, (1, ), batch_shape)
new_shape = array_ops.concat(0, ((-1, ), batch_shape, event_shape))
x = array_ops.reshape(x, shape=new_shape)
x = distribution_util.rotate_transpose(x, shift=-1)
return x, sample_shape
def make_batch_of_event_sample_matrices(
self, x, expand_batch_dim=True,
name="make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from S+B+E to B_+E_+S_.
Where:
- `B_ = B if B or not expand_batch_dim else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
Args:
x: `Tensor`.
expand_batch_dim: Python `Boolean` scalar. If `True` the batch dims will
be expanded such that batch_ndims>=1.
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
"""
with self._name_scope(name, values=[x]):
x = ops.convert_to_tensor(x, name="x")
sample_shape, batch_shape, event_shape = self.get_shape(x)
event_shape = distribution_util.pick_vector(
self._event_ndims_is_0, [1], event_shape)
if expand_batch_dim:
batch_shape = distribution_util.pick_vector(
self._batch_ndims_is_0, [1], batch_shape)
new_shape = array_ops.concat_v2([[-1], batch_shape, event_shape], 0)
x = array_ops.reshape(x, shape=new_shape)
x = distribution_util.rotate_transpose(x, shift=-1)
return x, sample_shape
def make_batch_of_event_sample_matrices(self, x, name="make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from S+B+E to B_+E_+S_.
Where:
- `B_ = B if B else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
Args:
x: `Tensor`.
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
"""
with self._name_scope(name, values=[x]):
x = ops.convert_to_tensor(x, name="x")
sample_shape, batch_shape, event_shape = self.get_shape(x)
event_shape = distribution_util.pick_vector(self._event_ndims_is_0, (1,), event_shape)
batch_shape = distribution_util.pick_vector(self._batch_ndims_is_0, (1,), batch_shape)
new_shape = array_ops.concat(0, ((-1,), batch_shape, event_shape))
x = array_ops.reshape(x, shape=new_shape)
x = distribution_util.rotate_transpose(x, shift=-1)
return x, sample_shape
def make_batch_of_event_sample_matrices(
self,
x,
expand_batch_dim=True,
name="make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from S+B+E to B_+E_+S_.
Where:
- `B_ = B if B or not expand_batch_dim else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
Args:
x: `Tensor`.
expand_batch_dim: Python `Boolean` scalar. If `True` the batch dims will
be expanded such that batch_ndims>=1.
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
"""
with self._name_scope(name, values=[x]):
x = ops.convert_to_tensor(x, name="x")
sample_shape, batch_shape, event_shape = self.get_shape(x)
event_shape = distribution_util.pick_vector(
self._event_ndims_is_0, [1], event_shape)
if expand_batch_dim:
batch_shape = distribution_util.pick_vector(
self._batch_ndims_is_0, [1], batch_shape)
new_shape = array_ops.concat([[-1], batch_shape, event_shape], 0)
x = array_ops.reshape(x, shape=new_shape)
x = distribution_util.rotate_transpose(x, shift=-1)
return x, sample_shape
def undo_make_batch_of_event_sample_matrices(
self,
x,
sample_shape,
expand_batch_dim=True,
name="undo_make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from B_+E_+S_ to S+B+E.
Where:
- `B_ = B if B or not expand_batch_dim else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
This function "reverses" `make_batch_of_event_sample_matrices`.
Args:
x: `Tensor` of shape `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
expand_batch_dim: Python `bool`. If `True` the batch dims will be expanded
such that `batch_ndims>=1`.
name: Python `str`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `S+B+E`.
"""
with self._name_scope(name, values=[x, sample_shape]):
x = ops.convert_to_tensor(x, name="x")
# x.shape: _B+_E+[prod(S)]
sample_shape = ops.convert_to_tensor(sample_shape,
name="sample_shape")
x = distribution_util.rotate_transpose(x, shift=1)
# x.shape: [prod(S)]+_B+_E
if self._is_all_constant_helper(self.batch_ndims,
self.event_ndims):
if self._batch_ndims_is_0 or self._event_ndims_is_0:
squeeze_dims = []
if self._event_ndims_is_0:
squeeze_dims += [-1]
if self._batch_ndims_is_0 and expand_batch_dim:
squeeze_dims += [1]
if squeeze_dims:
x = array_ops.squeeze(x, squeeze_dims=squeeze_dims)
# x.shape: [prod(S)]+B+E
_, batch_shape, event_shape = self.get_shape(x)
else:
s = (x.get_shape().as_list() if
x.get_shape().is_fully_defined() else array_ops.shape(x))
batch_shape = s[1:1 + self.batch_ndims]
# Since sample_dims=1 and is left-most, we add 1 to the number of
# batch_ndims to get the event start dim.
event_start = array_ops.where(
math_ops.logical_and(expand_batch_dim,
self._batch_ndims_is_0), 2,
1 + self.batch_ndims)
event_shape = s[event_start:event_start + self.event_ndims]
new_shape = array_ops.concat(
[sample_shape, batch_shape, event_shape], 0)
x = array_ops.reshape(x, shape=new_shape)
# x.shape: S+B+E
return x
def testRollDynamic(self):
with self.test_session() as sess:
x = tf.placeholder(tf.float32)
shift = tf.placeholder(tf.int32)
for x_value in (np.ones(1, dtype=x.dtype.as_numpy_dtype()),
np.ones((2, 1), dtype=x.dtype.as_numpy_dtype()),
np.ones((3, 2, 1), dtype=x.dtype.as_numpy_dtype())):
for shift_value in np.arange(-5, 5):
self.assertAllEqual(
self._np_rotate_transpose(x_value, shift_value),
sess.run(distribution_util.rotate_transpose(x, shift),
feed_dict={x: x_value, shift: shift_value}))
def undo_make_batch_of_event_sample_matrices(
self, x, sample_shape, expand_batch_dim=True,
name="undo_make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from B_+E_+S_ to S+B+E.
Where:
- `B_ = B if B or not expand_batch_dim else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
This function "reverses" `make_batch_of_event_sample_matrices`.
Args:
x: `Tensor` of shape `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
expand_batch_dim: Python `bool`. If `True` the batch dims will be expanded
such that `batch_ndims>=1`.
name: Python `str`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `S+B+E`.
"""
with self._name_scope(name, values=[x, sample_shape]):
x = ops.convert_to_tensor(x, name="x")
# x.shape: _B+_E+[prod(S)]
sample_shape = ops.convert_to_tensor(sample_shape, name="sample_shape")
x = distribution_util.rotate_transpose(x, shift=1)
# x.shape: [prod(S)]+_B+_E
if self._is_all_constant_helper(self.batch_ndims, self.event_ndims):
if self._batch_ndims_is_0 or self._event_ndims_is_0:
squeeze_dims = []
if self._event_ndims_is_0:
squeeze_dims += [-1]
if self._batch_ndims_is_0 and expand_batch_dim:
squeeze_dims += [1]
if squeeze_dims:
x = array_ops.squeeze(x, squeeze_dims=squeeze_dims)
# x.shape: [prod(S)]+B+E
_, batch_shape, event_shape = self.get_shape(x)
else:
s = (x.get_shape().as_list() if x.get_shape().is_fully_defined()
else array_ops.shape(x))
batch_shape = s[1:1+self.batch_ndims]
# Since sample_dims=1 and is left-most, we add 1 to the number of
# batch_ndims to get the event start dim.
event_start = array_ops.where(
math_ops.logical_and(expand_batch_dim, self._batch_ndims_is_0),
2, 1 + self.batch_ndims)
event_shape = s[event_start:event_start+self.event_ndims]
new_shape = array_ops.concat([sample_shape, batch_shape, event_shape], 0)
x = array_ops.reshape(x, shape=new_shape)
# x.shape: S+B+E
return x
def testRollDynamic(self):
with self.test_session() as sess:
x = tf.placeholder(tf.float32)
shift = tf.placeholder(tf.int32)
for x_value in (np.ones(1, dtype=x.dtype.as_numpy_dtype()),
np.ones((2, 1), dtype=x.dtype.as_numpy_dtype()),
np.ones((3, 2, 1), dtype=x.dtype.as_numpy_dtype())):
for shift_value in np.arange(-5, 5):
self.assertAllEqual(
self._np_rotate_transpose(x_value, shift_value),
sess.run(distribution_util.rotate_transpose(x, shift),
feed_dict={x: x_value, shift: shift_value}))
def undo_make_batch_of_event_sample_matrices(
self,
x,
sample_shape,
name="undo_make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from B_+E_+S_ to S+B+E.
Where:
- `B_ = B if B else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
This function "reverses" `make_batch_of_event_sample_matrices`.
Args:
x: `Tensor` of shape `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `S+B+E`.
"""
with self._name_scope(name, values=[x, sample_shape]):
x = ops.convert_to_tensor(x, name="x")
sample_shape = ops.convert_to_tensor(sample_shape,
name="sample_shape")
x = distribution_util.rotate_transpose(x, shift=1)
if self._is_all_constant_helper(self.batch_ndims,
self.event_ndims):
if self._batch_ndims_is_0 or self._event_ndims_is_0:
b = ((min(-2, -1 - self._event_ndims_static), )
if self._batch_ndims_is_0 else ())
e = (-1, ) if self._event_ndims_is_0 else ()
x = array_ops.squeeze(x, squeeze_dims=b + e)
_, batch_shape, event_shape = self.get_shape(x)
else:
s = (x.get_shape().as_list() if
x.get_shape().is_fully_defined() else array_ops.shape(x))
batch_shape = array_ops.slice(s, (1, ), (self.batch_ndims, ))
# Since sample_dims=1 and is left-most, we add 1 to the number of
# batch_ndims to get the event start dim.
event_start = math_ops.select(self._batch_ndims_is_0, 2,
1 + self.batch_ndims)
event_shape = array_ops.slice(s, (event_start, ),
(self.event_ndims, ))
new_shape = array_ops.concat(
0, (sample_shape, batch_shape, event_shape))
x = array_ops.reshape(x, shape=new_shape)
return x
def undo_make_batch_of_event_sample_matrices(
self, x, sample_shape, name="undo_make_batch_of_event_sample_matrices"):
"""Reshapes/transposes `Distribution` `Tensor` from B_+E_+S_ to S+B+E.
Where:
- `B_ = B if B else [1]`,
- `E_ = E if E else [1]`,
- `S_ = [tf.reduce_prod(S)]`.
This function "reverses" `make_batch_of_event_sample_matrices`.
Args:
x: `Tensor` of shape `B_+E_+S_`.
sample_shape: `Tensor` (1D, `int32`).
name: `String`. The name to give this op.
Returns:
x: `Tensor`. Input transposed/reshaped to `S+B+E`.
"""
with self._name_scope(name, values=[x, sample_shape]):
x = ops.convert_to_tensor(x, name="x")
sample_shape = ops.convert_to_tensor(sample_shape, name="sample_shape")
x = distribution_util.rotate_transpose(x, shift=1)
if self._is_all_constant_helper(self.batch_ndims, self.event_ndims):
if self._batch_ndims_is_0 or self._event_ndims_is_0:
b = ((min(-2, -1 - self._event_ndims_static),)
if self._batch_ndims_is_0 else ())
e = (-1,) if self._event_ndims_is_0 else ()
x = array_ops.squeeze(x, squeeze_dims=b + e)
_, batch_shape, event_shape = self.get_shape(x)
else:
s = (x.get_shape().as_list() if x.get_shape().is_fully_defined()
else array_ops.shape(x))
batch_shape = array_ops.slice(s, (1,), (self.batch_ndims,))
# Since sample_dims=1 and is left-most, we add 1 to the number of
# batch_ndims to get the event start dim.
event_start = array_ops.where(
self._batch_ndims_is_0, 2, 1 + self.batch_ndims)
event_shape = array_ops.slice(s, (event_start,), (self.event_ndims,))
new_shape = array_ops.concat(0, (sample_shape, batch_shape, event_shape))
x = array_ops.reshape(x, shape=new_shape)
return x