def __init__(self, loc, covariance_matrix=None, precision_matrix=None, scale_tril=None, validate_args=None):
if loc.dim() < 1:
raise ValueError("loc must be at least one-dimensional.")
if (covariance_matrix is not None) + (scale_tril is not None) + (precision_matrix is not None) != 1:
raise ValueError("Exactly one of covariance_matrix or precision_matrix or scale_tril may be specified.")
if scale_tril is not None:
if scale_tril.dim() < 2:
raise ValueError("scale_tril matrix must be at least two-dimensional, "
"with optional leading batch dimensions")
batch_shape = torch.broadcast_shapes(scale_tril.shape[:-2], loc.shape[:-1])
self.scale_tril = scale_tril.expand(batch_shape + (-1, -1))
elif covariance_matrix is not None:
if covariance_matrix.dim() < 2:
raise ValueError("covariance_matrix must be at least two-dimensional, "
"with optional leading batch dimensions")
batch_shape = torch.broadcast_shapes(covariance_matrix.shape[:-2], loc.shape[:-1])
self.covariance_matrix = covariance_matrix.expand(batch_shape + (-1, -1))
else:
if precision_matrix.dim() < 2:
raise ValueError("precision_matrix must be at least two-dimensional, "
"with optional leading batch dimensions")
batch_shape = torch.broadcast_shapes(precision_matrix.shape[:-2], loc.shape[:-1])
self.precision_matrix = precision_matrix.expand(batch_shape + (-1, -1))
self.loc = loc.expand(batch_shape + (-1,))
event_shape = self.loc.shape[-1:]
super(MultivariateNormal, self).__init__(batch_shape, event_shape, validate_args=validate_args)
if scale_tril is not None:
self._unbroadcasted_scale_tril = scale_tril
elif covariance_matrix is not None:
self._unbroadcasted_scale_tril = torch.cholesky(covariance_matrix)
else: # precision_matrix is not None
self._unbroadcasted_scale_tril = _precision_to_scale_tril(precision_matrix)
def meta_cdist_forward(x1, x2, p, compute_mode):
check(
x1.dim() >= 2,
lambda: f"cdist only supports at least 2D tensors, X1 got: {x1.dim()}D",
)
check(
x2.dim() >= 2,
lambda: f"cdist only supports at least 2D tensors, X2 got: {x2.dim()}D",
)
check(
x1.size(-1) == x2.size(-1),
lambda: f"X1 and X2 must have the same number of columns. X1: {x1.size(-1)} X2: {x2.size(-1)}",
)
check(
utils.is_float_dtype(x1.dtype),
lambda: "cdist only supports floating-point dtypes, X1 got: {x1.dtype}",
)
check(
utils.is_float_dtype(x2.dtype),
lambda: "cdist only supports floating-point dtypes, X2 got: {x2.dtype}",
)
check(p >= 0, lambda: "cdist only supports non-negative p values")
check(
compute_mode >= 0 and compute_mode <= 2,
lambda: f"possible modes: 0, 1, 2, but was: {compute_mode}",
)
r1 = x1.size(-2)
r2 = x2.size(-2)
batch_tensor1 = x1.shape[:-2]
batch_tensor2 = x2.shape[:-2]
output_shape = list(torch.broadcast_shapes(batch_tensor1, batch_tensor2))
output_shape.extend([r1, r2])
return x1.new_empty(output_shape)
def broadcast_except(*tensors: Tensor, dim=-1):
shape = broadcast_shapes(*(t.select(dim, 0).shape for t in tensors))
return [
t.expand(*shape[:t.ndim + dim + 1], t.shape[dim],
*shape[t.ndim + dim + 1:])
for t in pad_dims(*tensors, ndim=len(shape) + 1)
]
def __init__(self,
df: Union[torch.Tensor, Number],
covariance_matrix: torch.Tensor = None,
precision_matrix: torch.Tensor = None,
scale_tril: torch.Tensor = None,
validate_args=None):
assert (covariance_matrix is not None) + (scale_tril is not None) + (precision_matrix is not None) == 1, \
"Exactly one of covariance_matrix or precision_matrix or scale_tril may be specified."
param = next(p for p in (covariance_matrix, precision_matrix, scale_tril) if p is not None)
if param.dim() < 2:
raise ValueError("scale_tril must be at least two-dimensional, with optional leading batch dimensions")
if isinstance(df, Number):
batch_shape = torch.Size(param.shape[:-2])
self.df = torch.tensor(df, dtype=param.dtype, device=param.device)
else:
batch_shape = torch.broadcast_shapes(param.shape[:-2], df.shape)
self.df = df.expand(batch_shape)
event_shape = param.shape[-2:]
if self.df.le(event_shape[-1] - 1).any():
raise ValueError(f"Value of df={df} expected to be greater than ndim - 1 = {event_shape[-1]-1}.")
if scale_tril is not None:
self.scale_tril = param.expand(batch_shape + (-1, -1))
elif covariance_matrix is not None:
self.covariance_matrix = param.expand(batch_shape + (-1, -1))
elif precision_matrix is not None:
self.precision_matrix = param.expand(batch_shape + (-1, -1))
self.arg_constraints['df'] = constraints.greater_than(event_shape[-1] - 1)
if self.df.lt(event_shape[-1]).any():
warnings.warn("Low df values detected. Singular samples are highly likely to occur for ndim - 1 < df < ndim.")
super(Wishart, self).__init__(batch_shape, event_shape, validate_args=validate_args)
self._batch_dims = [-(x + 1) for x in range(len(self._batch_shape))]
if scale_tril is not None:
self._unbroadcasted_scale_tril = scale_tril
elif covariance_matrix is not None:
self._unbroadcasted_scale_tril = torch.linalg.cholesky(covariance_matrix)
else: # precision_matrix is not None
self._unbroadcasted_scale_tril = _precision_to_scale_tril(precision_matrix)
# Chi2 distribution is needed for Bartlett decomposition sampling
self._dist_chi2 = torch.distributions.chi2.Chi2(
df=(
self.df.unsqueeze(-1)
- torch.arange(
self._event_shape[-1],
dtype=self._unbroadcasted_scale_tril.dtype,
device=self._unbroadcasted_scale_tril.device,
).expand(batch_shape + (-1,))
)
)
def broadcastable(*shapes) -> bool:
r"""Returns whether `shapes` are broadcastable.
Example:
>>> x = torch.rand(3, 2, 1)
>>> y = torch.rand(1, 2, 3)
>>> z = torch.rand(2, 2, 2)
>>> broadcastable(x.shape, y.shape)
True
>>> broadcastable(y.shape, z.shape)
False
"""
try:
torch.broadcast_shapes(*shapes)
except RuntimeError as e:
return False
else:
return True
def _dense_unflatten(self, flat_samples: torch.Tensor) -> Dict[str, torch.Tensor]:
# Convert a single flattened sample to a dict of shaped samples.
sample_shape = flat_samples.shape[:-1]
samples = {}
pos = 0
for d, (batch_shape, event_shape) in self._dense_shapes.items():
end = pos + (batch_shape + event_shape).numel()
flat_sample = flat_samples[..., pos:end]
pos = end
# Assumes sample shapes are left of batch shapes.
samples[d] = flat_sample.reshape(
torch.broadcast_shapes(sample_shape, batch_shape) + event_shape
)
return samples
def broadcast_gather(input, dim, index, sparse_grad=False, index_ndim=1):
"""
input: Size(batch_shape..., N, event_shape...)
index: Size(batch_shape..., index_shape...)
->: Size(batch_shape..., index_shape..., event_shape...)
"""
index_shape = index.shape[-index_ndim:]
index = index.flatten(-index_ndim)
batch_shape = broadcast_shapes(input.shape[:dim], index.shape[:-1])
input = input.expand(*batch_shape, *input.shape[dim:])
index = index.expand(*batch_shape, index.shape[-1])
return torch.gather(
input,
dim,
index.reshape(*index.shape,
*(input.ndim - index.ndim) *
(1, )).expand(*index.shape, *input.shape[index.ndim:])
if input.ndim > index.ndim else index,
sparse_grad=sparse_grad).reshape(*index.shape[:-1], *index_shape,
*input.shape[dim % input.ndim + 1:])
def _masked_observe(name, fn, obs, obs_mask, *args, **kwargs):
# Split into two auxiliary sample sites.
with poutine.mask(mask=obs_mask):
observed = sample(f"{name}_observed", fn, *args, **kwargs, obs=obs)
with poutine.mask(mask=~obs_mask):
unobserved = sample(f"{name}_unobserved", fn, *args, **kwargs)
# Interleave observed and unobserved events.
shape = obs_mask.shape + (1, ) * fn.event_dim
batch_mask = obs_mask.reshape(shape)
try:
value = torch.where(batch_mask, observed, unobserved)
except RuntimeError as e:
if "must match the size of tensor" in str(e):
shape = torch.broadcast_shapes(observed.shape, unobserved.shape)
batch_shape = shape[:len(shape) - fn.event_dim]
raise ValueError(
f"Invalid obs_mask shape {tuple(obs_mask.shape)}; should be "
f"broadcastable to batch_shape = {tuple(batch_shape)}") from e
raise
return deterministic(name, value)
def _unpack_latent(self, latent):
"""
Unpacks a packed latent tensor, iterating over tuples of the form::
(site, unconstrained_value)
"""
batch_shape = latent.shape[:
-1] # for plates outside of _setup_prototype, e.g. parallel particles
pos = 0
for name, site in self.prototype_trace.iter_stochastic_nodes():
constrained_shape = site["value"].shape
unconstrained_shape = self._unconstrained_shapes[name]
size = _product(unconstrained_shape)
event_dim = (site["fn"].event_dim + len(unconstrained_shape) -
len(constrained_shape))
unconstrained_shape = torch.broadcast_shapes(
unconstrained_shape, batch_shape + (1, ) * event_dim)
unconstrained_value = latent[..., pos:pos +
size].view(unconstrained_shape)
yield site, unconstrained_value
pos += size
if not torch._C._get_tracing_state():
assert pos == latent.size(-1)
def __init__(self,
base_dist: TorchDistribution,
skewness,
validate_args=None):
assert (
base_dist.event_shape == skewness.shape[-1:]
), "Sine Skewing is only valid with a skewness parameter for each dimension of `base_dist.event_shape`."
if (skewness.abs().sum(-1) > 1.0).any():
warnings.warn("Total skewness weight shouldn't exceed one.",
UserWarning)
batch_shape = broadcast_shapes(base_dist.batch_shape,
skewness.shape[:-1])
event_shape = skewness.shape[-1:]
self.skewness = skewness.broadcast_to(batch_shape + event_shape)
self.base_dist = base_dist.expand(batch_shape)
super().__init__(batch_shape, event_shape, validate_args=validate_args)
if self._validate_args and base_dist.mean.device != skewness.device:
raise ValueError(
f"base_density: {base_dist.__class__.__name__} and SineSkewed "
f"must be on same device.")
def broadcast_shape(shape_a: Tuple[int, ...],
shape_b: Tuple[int, ...]) -> Tuple[int, ...]:
"""
Infers, if possible, the broadcast output shape of two operands a and b. Inspired by stackoverflow post:
https://stackoverflow.com/questions/24743753/test-if-an-array-is-broadcastable-to-a-shape
Parameters
----------
shape_a : Tuple[int,...]
Shape of first operand
shape_b : Tuple[int,...]
Shape of second operand
Raises
-------
ValueError
If the two shapes cannot be broadcast.
Examples
--------
>>> import heat as ht
>>> ht.core.stride_tricks.broadcast_shape((5,4),(4,))
(5, 4)
>>> ht.core.stride_tricks.broadcast_shape((1,100,1),(10,1,5))
(10, 100, 5)
>>> ht.core.stride_tricks.broadcast_shape((8,1,6,1),(7,1,5,))
(8,7,6,5))
>>> ht.core.stride_tricks.broadcast_shape((2,1),(8,4,3))
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "heat/core/stride_tricks.py", line 42, in broadcast_shape
"operands could not be broadcast, input shapes {} {}".format(shape_a, shape_b)
ValueError: operands could not be broadcast, input shapes (2, 1) (8, 4, 3)
"""
try:
resulting_shape = torch.broadcast_shapes(shape_a, shape_b)
except AttributeError: # torch < 1.8
it = itertools.zip_longest(shape_a[::-1], shape_b[::-1], fillvalue=1)
resulting_shape = max(len(shape_a), len(shape_b)) * [None]
for i, (a, b) in enumerate(it):
if a == 0 and b == 1 or b == 0 and a == 1:
resulting_shape[i] = 0
elif a == 1 or b == 1 or a == b:
resulting_shape[i] = max(a, b)
else:
raise ValueError(
"operands could not be broadcast, input shapes {} {}".
format(shape_a, shape_b))
return tuple(resulting_shape[::-1])
except TypeError:
raise TypeError("operand 1 must be tuple of ints, not {}".format(
type(shape_a)))
except NameError:
raise TypeError(
"operands must be tuples of ints, not {} and {}".format(
shape_a, shape_b))
except RuntimeError:
raise ValueError(
"operands could not be broadcast, input shapes {} {}".format(
shape_a, shape_b))
return tuple(resulting_shape)
def __init__(
self,
height: torch.Tensor,
width: torch.Tensor,
x: torch.Tensor,
y: torch.Tensor,
target_locs: torch.Tensor,
background: torch.Tensor,
gain: torch.Tensor,
offset_samples: torch.Tensor,
offset_logits: torch.Tensor,
P: int,
m: torch.Tensor = None,
alpha: torch.Tensor = None,
use_pykeops: bool = True,
validate_args=None,
):
# shapes for cosmos and crosstalk models
self.height = height # (N, F, C, K) or (N, F, Q, K)
self.width = width # (N, F, C, K) or (N, F, Q, K)
self.x = x
self.y = y
self.target_locs = target_locs # (N, F, C, 2)
self.m = m # (N, F, C, K) or (N, F, Q, K)
self.background = background[..., None, None] # (N, F, C, P, P)
if alpha is not None:
C = alpha.shape[-1]
self.gain = gain[..., None, None, None] # (1, K, P, P)
self.height = (self.height.unsqueeze(-2) * alpha[..., None]
) # (N, F, Q, C, K)
self.width = self.width.unsqueeze(-2) # (N, F, Q, 1, K)
self.x = self.x.unsqueeze(-2) # (N, F, Q, 1, K)
self.y = self.y.unsqueeze(-2) # (N, F, Q, 1, K)
self.m = self.m.unsqueeze(-2) # (N, F, Q, 1, K)
self.target_locs = self.target_locs.unsqueeze(
-3) # (N, F, 1, C, 2)
else:
self.gain = gain[..., None, None] # (1, P, P)
self.alpha = alpha
self.rate = 1 / self.gain
self.offset_samples = offset_samples
self.offset_logits = offset_logits
self.P = P
self.use_pykeops = use_pykeops
if self.use_pykeops:
device = self.target_locs.device.type
self.device_pykeops = "GPU" if device == "cuda" else "CPU"
# calculate batch shape
batch_shape = torch.broadcast_shapes(
height.shape, width.shape, x.shape,
y.shape) # (N, F, C, K) or (N, F, Q, K)
if m is not None:
batch_shape = torch.broadcast_shapes(
batch_shape, m.shape) # (N, F, C, K) or (N, F, Q, K)
event_shape = torch.Size([P, P]) # (P, P)
bg_shape = background.shape # (N, F, C)
target_shape = target_locs.shape[:-1] # (N, F, C)
# remove K dim
batch_shape = batch_shape[:-1] # (N, F, C) or (N, F, Q)
if alpha is not None:
# remove Q dim
batch_shape = batch_shape[:-1] # (N, F)
# add C dim
event_shape = (C, ) + event_shape # (C, P, P)
# remove C dim
bg_shape = bg_shape[:-1] # (N, F)
target_shape = target_shape[:-1] # (N, F)
batch_shape = torch.broadcast_shapes(
batch_shape, bg_shape, target_shape) # (N, F, C) or (N, F)
super().__init__(batch_shape, event_shape, validate_args=validate_args)
def broadcast_left(*tensors, ndim):
shape = broadcast_shapes(*(t.shape[:ndim] for t in tensors))
return (t.expand(*shape, *t.shape[ndim:]) for t in tensors)
def forward_shape(self, shape):
return torch.broadcast_shapes(shape, getattr(self.exponent, "shape",
()))
def inverse_shape(self, shape):
return torch.broadcast_shapes(shape, getattr(self.exponent, "shape",
()))
def inverse_shape(self, shape):
return torch.broadcast_shapes(shape, getattr(self.loc, "shape", ()),
getattr(self.scale, "shape", ()))
def get_deltas(self, save_params=None):
deltas = {}
aux_values = {}
compute_density = poutine.get_mask() is not False
for name, site in self._sorted_sites:
if save_params is not None and name not in save_params:
continue
# Sample zero-mean blockwise independent Delta/Normal/MVN.
log_density = 0.0
loc = deep_getattr(self.locs, name)
zero = torch.zeros_like(loc)
conditional = self.conditionals[name]
if callable(conditional):
aux_value = deep_getattr(self.conds, name)()
elif conditional == "delta":
aux_value = zero
elif conditional == "normal":
aux_value = pyro.sample(
name + "_aux",
dist.Normal(zero, 1).to_event(1),
infer={"is_auxiliary": True},
)
scale = deep_getattr(self.scales, name)
aux_value = aux_value * scale
if compute_density:
log_density = (-scale.log()).expand_as(aux_value)
elif conditional == "mvn":
# This overparametrizes by learning (scale,scale_tril),
# enabling faster learning of the more-global scale parameter.
aux_value = pyro.sample(
name + "_aux",
dist.Normal(zero, 1).to_event(1),
infer={"is_auxiliary": True},
)
scale = deep_getattr(self.scales, name)
scale_tril = deep_getattr(self.scale_trils, name)
aux_value = aux_value @ scale_tril.T * scale
if compute_density:
log_density = (
-scale_tril.diagonal(dim1=-2, dim2=-1).log() -
scale.log()).expand_as(aux_value)
else:
raise ValueError(
f"Unsupported conditional type: {conditional}")
# Accumulate upstream dependencies.
# Note: by accumulating upstream dependencies before updating the
# aux_values dict, we encode a block-sparse structure of the
# precision matrix; if we had instead accumulated after updating
# aux_values, we would encode a block-sparse structure of the
# covariance matrix.
# Note: these shear transforms have no effect on the Jacobian
# determinant, and can therefore be excluded from the log_density
# computation below, even for nonlinear dep().
deps = deep_getattr(self.deps, name)
for upstream in self.dependencies.get(name, {}):
dep = deep_getattr(deps, upstream)
aux_value = aux_value + dep(aux_values[upstream])
aux_values[name] = aux_value
# Shift by loc and reshape.
batch_shape = torch.broadcast_shapes(aux_value.shape[:-1],
self._batch_shapes[name])
unconstrained = (
aux_value +
loc).reshape(batch_shape +
self._unconstrained_event_shapes[name])
if not is_identically_zero(log_density):
log_density = log_density.reshape(batch_shape + (-1, )).sum(-1)
# Transform to constrained space.
transform = biject_to(site["fn"].support)
value = transform(unconstrained)
if compute_density and conditional != "delta":
assert transform.codomain.event_dim == site["fn"].event_dim
log_density = log_density + transform.inv.log_abs_det_jacobian(
value, unconstrained)
# Create a reparametrized Delta distribution.
deltas[name] = dist.Delta(value, log_density, site["fn"].event_dim)
return deltas
