def testOddRank(self):
w = np.random.rand(2)
x = np.random.rand(2, 2, 2)
self.assertRaises(ValueError, self.diagPartOp, w, np.float32, 0)
self.assertRaises(ValueError, self.diagPartOp, x, np.float32, 0)
with self.assertRaises(ValueError):
array_ops.diag_part(0.0)
def __call__(self, shape, dtype=None, partition_info=None):
if dtype is None:
dtype = self.dtype
# Check the shape
if len(shape) < 3 or len(shape) > 5:
raise ValueError("The tensor to initialize must be at least "
"three-dimensional and at most five-dimensional")
if shape[-2] > shape[-1]:
raise ValueError("In_filters cannot be greater than out_filters.")
# Generate a random matrix
a = random_ops.random_normal([shape[-1], shape[-1]],
dtype=dtype, seed=self.seed)
# Compute the qr factorization
q, r = linalg_ops.qr(a, full_matrices=False)
# Make Q uniform
d = array_ops.diag_part(r)
q *= math_ops.sign(d)
q = q[:shape[-2], :]
q *= math_ops.sqrt(math_ops.cast(self.gain, dtype=dtype))
if len(shape) == 3:
weight = array_ops.scatter_nd([[(shape[0]-1)//2]],
array_ops.expand_dims(q, 0), shape)
elif len(shape) == 4:
weight = array_ops.scatter_nd([[(shape[0]-1)//2, (shape[1]-1)//2]],
array_ops.expand_dims(q, 0), shape)
else:
weight = array_ops.scatter_nd([[(shape[0]-1)//2, (shape[1]-1)//2,
(shape[2]-1)//2]],
array_ops.expand_dims(q, 0), shape)
return weight
def __call__(self, shape, dtype=None, partition_info=None):
if dtype is None:
dtype = self.dtype
# Check the shape
if len(shape) < 2:
raise ValueError("The tensor to initialize must be "
"at least two-dimensional")
# Flatten the input shape with the last dimension remaining
# its original shape so it works for conv2d
num_rows = 1
for dim in shape[:-1]:
num_rows *= dim
num_cols = shape[-1]
flat_shape = (num_cols, num_rows) if num_rows < num_cols else (num_rows,
num_cols)
# Generate a random matrix
a = random_ops.random_normal(flat_shape, dtype=dtype, seed=self.seed)
# Compute the qr factorization
q, r = linalg_ops.qr(a, full_matrices=False)
# Make Q uniform
d = array_ops.diag_part(r)
q *= math_ops.sign(d)
if num_rows < num_cols:
q = array_ops.matrix_transpose(q)
return self.gain * array_ops.reshape(q, shape)
def _diagPartOp(self, tensor, dtype, expected_ans, use_gpu):
with self.cached_session(use_gpu=use_gpu):
tensor = ops.convert_to_tensor(tensor.astype(dtype))
tf_ans_inv = array_ops.diag_part(tensor)
inv_out = self.evaluate(tf_ans_inv)
self.assertAllClose(inv_out, expected_ans)
self.assertShapeEqual(expected_ans, tf_ans_inv)
def __call__(self, shape, dtype=None, partition_info=None):
if dtype is None:
dtype = self.dtype
# Check the shape
if len(shape) < 2:
raise ValueError("The tensor to initialize must be "
"at least two-dimensional")
# Flatten the input shape with the last dimension remaining
# its original shape so it works for conv2d
num_rows = 1
for dim in shape[:-1]:
num_rows *= dim
num_cols = shape[-1]
flat_shape = (num_rows, num_cols)
# Generate a random matrix
a = random_ops.random_normal(flat_shape, dtype=dtype, seed=self.seed)
# Compute the qr factorization
q, r = linalg_ops.qr(a, full_matrices=False)
# Make Q uniform
square_len = math_ops.minimum(num_rows, num_cols)
d = array_ops.diag_part(r[:square_len, :square_len])
ph = d / math_ops.abs(d)
q *= ph
# Pad zeros to Q (if rows smaller than cols)
if num_rows < num_cols:
padding = array_ops.zeros([num_rows, num_cols - num_rows], dtype=dtype)
q = array_ops.concat([q, padding], 1)
return self.gain * array_ops.reshape(q, shape)
def __call__(self, shape, dtype=dtypes.float32):
"""Returns a tensor object initialized as specified by the initializer.
Args:
shape: Shape of the tensor.
dtype: Optional dtype of the tensor. Only floating point types are
supported.
Raises:
ValueError: If the dtype is not floating point or the input shape is not
valid.
"""
dtype = _assert_float_dtype(dtype)
# Check the shape
if len(shape) < 2:
raise ValueError("The tensor to initialize must be "
"at least two-dimensional")
# Flatten the input shape with the last dimension remaining
# its original shape so it works for conv2d
num_rows = 1
for dim in shape[:-1]:
num_rows *= dim
num_cols = shape[-1]
flat_shape = (max(num_cols, num_rows), min(num_cols, num_rows))
# Generate a random matrix
a = random_ops.random_normal(flat_shape, dtype=dtype, seed=self.seed)
# Compute the qr factorization
q, r = gen_linalg_ops.qr(a, full_matrices=False)
# Make Q uniform
d = array_ops.diag_part(r)
q *= math_ops.sign(d)
if num_rows < num_cols:
q = array_ops.matrix_transpose(q)
return self.gain * array_ops.reshape(q, shape)
def trace(x, name=None):
""" Compute the trace of a tensor `x`.
`trace(x)` returns the sum of along the diagonal.
For example:
```python
# 'x' is [[1, 1],
# [1, 1]]
tf.trace(x) ==> 2
# 'x' is [[1,2,3],
# [4,5,6],
# [7,8,9]]
tf.trace(x) ==> 15
```
Args:
x: 2-D tensor.
name: A name for the operation (optional).
Returns:
The trace of input tensor.
"""
with ops.op_scope([x], name, "Trace") as name:
x = ops.convert_to_tensor(x, name="x")
if len(x.get_shape()) != 2:
raise ValueError("Expected a tensor with rank 2, rank %d tensor received"
% len(x.get_shape()))
return reduce_sum(array_ops.diag_part(x), name=name)
def diagOp(self, diag, dtype, expected_ans, use_gpu=False):
with self.test_session(use_gpu=use_gpu):
tf_ans = array_ops.diag(ops.convert_to_tensor(diag.astype(dtype)))
out = tf_ans.eval()
tf_ans_inv = array_ops.diag_part(expected_ans)
inv_out = tf_ans_inv.eval()
self.assertAllClose(out, expected_ans)
self.assertAllClose(inv_out, diag)
self.assertShapeEqual(expected_ans, tf_ans)
self.assertShapeEqual(diag, tf_ans_inv)
def _diagOp(self, diag, dtype, expected_ans, use_gpu):
with self.cached_session(use_gpu=use_gpu):
tf_ans = array_ops.diag(ops.convert_to_tensor(diag.astype(dtype)))
out = self.evaluate(tf_ans)
tf_ans_inv = array_ops.diag_part(expected_ans)
inv_out = self.evaluate(tf_ans_inv)
self.assertAllClose(out, expected_ans)
self.assertAllClose(inv_out, diag)
self.assertShapeEqual(expected_ans, tf_ans)
self.assertShapeEqual(diag, tf_ans_inv)
def testRankFourFloatTensorUnknownShape(self):
x = np.random.rand(3, 3)
i = np.arange(3)
expected_ans = x[i, i]
for shape in None, (None, 3), (3, None):
with self.cached_session(use_gpu=False):
t = ops.convert_to_tensor(x.astype(np.float32))
t.set_shape(shape)
tf_ans = array_ops.diag_part(t)
out = self.evaluate(tf_ans)
self.assertAllClose(out, expected_ans)
self.assertShapeEqual(expected_ans, tf_ans)
def _define_maximization_operation(self, num_batches):
"""Maximization operations."""
# TODO(xavigonzalvo): some of these operations could be moved to C++.
# Compute the effective number of data points assigned to component k.
with ops.control_dependencies(self._w):
points_in_k = array_ops.squeeze(
math_ops.add_n(self._points_in_k), axis=[0])
# Update alpha.
if 'w' in self._params:
final_points_in_k = points_in_k / num_batches
num_examples = math_ops.cast(math_ops.reduce_sum(final_points_in_k),
dtypes.float32)
self._alpha_op = self._alpha.assign(final_points_in_k /
(num_examples + MEPS))
else:
self._alpha_op = control_flow_ops.no_op()
self._train_ops = [self._alpha_op]
# Update means.
points_in_k_expanded = array_ops.reshape(points_in_k,
[self._num_classes, 1, 1])
if 'm' in self._params:
self._means_op = self._means.assign(
math_ops.div(
math_ops.add_n(self._w_mul_x), points_in_k_expanded + MEPS))
else:
self._means_op = control_flow_ops.no_op()
# means are (num_classes x 1 x dims)
# Update covariances.
with ops.control_dependencies([self._means_op]):
b = math_ops.add_n(self._w_mul_x2) / (points_in_k_expanded + MEPS)
new_covs = []
for k in range(self._num_classes):
mean = self._means.value()[k, :, :]
square_mean = math_ops.matmul(mean, mean, transpose_a=True)
new_cov = b[k, :, :] - square_mean + self._min_var
if self._covariance_type == FULL_COVARIANCE:
new_covs.append(array_ops.expand_dims(new_cov, 0))
elif self._covariance_type == DIAG_COVARIANCE:
new_covs.append(
array_ops.expand_dims(array_ops.diag_part(new_cov), 0))
new_covs = array_ops.concat(new_covs, 0)
if 'c' in self._params:
# Train operations don't need to take care of the means
# because covariances already depend on it.
with ops.control_dependencies([self._means_op, new_covs]):
self._train_ops.append(
state_ops.assign(
self._covs, new_covs, validate_shape=False))
def testDiagPartGrad(self):
np.random.seed(0)
shapes = ((3, 3), (3, 3, 3, 3))
dtypes = (dtypes_lib.float32, dtypes_lib.float64)
with self.test_session(use_gpu=False):
errors = []
for shape in shapes:
for dtype in dtypes:
x1 = constant_op.constant(np.random.rand(*shape), dtype=dtype)
y = array_ops.diag_part(x1)
error = gradient_checker.compute_gradient_error(
x1, x1.get_shape().as_list(), y, y.get_shape().as_list())
tf_logging.info("error = %f", error)
self.assertLess(error, 1e-4)
def _initialize_variables(self, data, initial_means=None):
"""Initializes variables.
Args:
data: a list of Tensors with data, each row is a new example.
initial_means: a Tensor with a matrix of means.
"""
first_shard = data[0]
# Initialize means: num_classes X 1 X dimensions.
if initial_means is not None:
means = array_ops.expand_dims(initial_means, 1)
else:
# Sample data randomly
means = array_ops.expand_dims(
_init_clusters_random(data, self._num_classes, self._random_seed), 1)
# Initialize covariances.
if self._covariance_type == FULL_COVARIANCE:
cov = _covariance(first_shard, False) + self._min_var
# A matrix per class, num_classes X dimensions X dimensions
covs = array_ops.tile(
array_ops.expand_dims(cov, 0), [self._num_classes, 1, 1])
elif self._covariance_type == DIAG_COVARIANCE:
cov = _covariance(first_shard, True) + self._min_var
# A diagonal per row, num_classes X dimensions.
covs = array_ops.tile(
array_ops.expand_dims(array_ops.diag_part(cov), 0),
[self._num_classes, 1])
with ops.colocate_with(self._cluster_centers_initialized):
initialized = control_flow_ops.with_dependencies(
[means, covs],
array_ops.identity(self._cluster_centers_initialized))
self._init_ops = []
with ops.colocate_with(self._means):
init_means = state_ops.assign(self._means, means, validate_shape=False)
init_means = control_flow_ops.with_dependencies(
[init_means],
state_ops.assign(self._cluster_centers_initialized, True))
self._init_ops.append(control_flow_ops.cond(initialized,
control_flow_ops.no_op,
lambda: init_means).op)
with ops.colocate_with(self._covs):
init_covs = state_ops.assign(self._covs, covs, validate_shape=False)
init_covs = control_flow_ops.with_dependencies(
[init_covs],
state_ops.assign(self._cluster_centers_initialized, True))
self._init_ops.append(control_flow_ops.cond(initialized,
control_flow_ops.no_op,
lambda: init_covs).op)
def _orthogonal_matrix(self, n):
"""Construct an n x n orthogonal matrix.
Args:
n: dimension.
Returns:
a n x n orthogonal matrix.
"""
a = random_ops.random_normal([n, n], dtype=self.dtype, seed=self.seed)
if self.seed:
self.seed += 1
q, r = linalg_ops.qr(a)
d = array_ops.diag_part(r)
# make q uniform
q *= math_ops.sign(d)
return q
def _create_variables(self, data, initial_means=None):
"""Initializes GMM algorithm.
Args:
data: a list of Tensors with data, each row is a new example.
initial_means: a Tensor with a matrix of means.
"""
first_shard = data[0]
# Initialize means: num_classes X 1 X dimensions.
if initial_means is not None:
self._means = variables.Variable(
array_ops.expand_dims(initial_means, 1),
name=self.CLUSTERS_VARIABLE,
validate_shape=False,
dtype=dtypes.float32)
else:
# Sample data randomly
self._means = variables.Variable(
array_ops.expand_dims(
_init_clusters_random(data, self._num_classes, self._random_seed),
1),
name=self.CLUSTERS_VARIABLE,
validate_shape=False)
# Initialize covariances.
if self._covariance_type == FULL_COVARIANCE:
cov = _covariance(first_shard, False) + self._min_var
# A matrix per class, num_classes X dimensions X dimensions
covs = array_ops.tile(
array_ops.expand_dims(cov, 0), [self._num_classes, 1, 1])
elif self._covariance_type == DIAG_COVARIANCE:
cov = _covariance(first_shard, True) + self._min_var
# A diagonal per row, num_classes X dimensions.
covs = array_ops.tile(
array_ops.expand_dims(array_ops.diag_part(cov), 0),
[self._num_classes, 1])
self._covs = variables.Variable(
covs, name=self.CLUSTERS_COVS_VARIABLE, validate_shape=False)
# Mixture weights, representing the probability that a randomly
# selected unobservable data (in EM terms) was generated by component k.
self._alpha = variables.Variable(
array_ops.tile([1.0 / self._num_classes], [self._num_classes]),
name=self.CLUSTERS_WEIGHT,
validate_shape=False)
def _DiagGrad(_, grad): return array_ops.diag_part(grad)
def _DiagGrad(_, grad):
return array_ops.diag_part(grad)
def loop_fn(i): inp = array_ops.gather(x, i) # pylint: disable=redefined-builtin return array_ops.diag_part(inp)
