RealNVP 'affine coupling layer' for vector-valued events.
Inherits From: Bijector
oryx.bijectors.RealNVP(
num_masked=None,
fraction_masked=None,
shift_and_log_scale_fn=None,
bijector_fn=None,
is_constant_jacobian=False,
validate_args=False,
name=None
)
Real NVP models a normalizing flow on a D
-dimensional distribution via a
single D-d
-dimensional conditional distribution [(Dinh et al., 2017)][1]:
y[d:D] = x[d:D] * tf.exp(log_scale_fn(x[0:d])) + shift_fn(x[0:d])
y[0:d] = x[0:d]
The last D-d
units are scaled and shifted based on the first d
units only,
while the first d
units are 'masked' and left unchanged. Real NVP's
shift_and_log_scale_fn
computes vector-valued quantities. For
scale-and-shift transforms that do not depend on any masked units, i.e.
d=0
, use the tfb.Scale
and tfb.Shift
bijectors with learned parameters
instead.
Masking is currently only supported for base distributions with
event_ndims=1
. For more sophisticated masking schemes like checkerboard or
channel-wise masking [(Papamakarios et al., 2016)[4], use the tfb.Permute
bijector to re-order desired masked units into the first d
units. For base
distributions with event_ndims > 1
, use the tfb.Reshape
bijector to
flatten the event shape.
Recall that the MAF bijector [(Papamakarios et al., 2016)][4] implements a normalizing flow via an autoregressive transformation. MAF and IAF have opposite computational tradeoffs - MAF can train all units in parallel but must sample units sequentially, while IAF must train units sequentially but can sample in parallel. In contrast, Real NVP can compute both forward and inverse computations in parallel. However, the lack of an autoregressive transformations makes it less expressive on a per-bijector basis.
A 'valid' shift_and_log_scale_fn
must compute each shift
(aka loc
or
'mu' in [Papamakarios et al. (2016)][4]) and log(scale)
(aka 'alpha' in
[Papamakarios et al. (2016)][4]) such that each are broadcastable with the
arguments to forward
and inverse
, i.e., such that the calculations in
forward
, inverse
[below] are possible. For convenience,
real_nvp_default_template
is offered as a possible shift_and_log_scale_fn
function.
NICE [(Dinh et al., 2014)][2] is a special case of the Real NVP bijector
which discards the scale transformation, resulting in a constant-time
inverse-log-determinant-Jacobian. To use a NICE bijector instead of Real
NVP, shift_and_log_scale_fn
should return (shift, None)
, and
is_constant_jacobian
should be set to True
in the RealNVP
constructor.
Calling real_nvp_default_template
with shift_only=True
returns one such
NICE-compatible shift_and_log_scale_fn
.
The bijector_fn
argument allows specifying a more general coupling relation,
such as the LSTM-inspired activation from [5], or Neural Spline Flow [6].
Caching: the scalar input depth D
of the base distribution is not known at
construction time. The first call to any of forward(x)
, inverse(x)
,
inverse_log_det_jacobian(x)
, or forward_log_det_jacobian(x)
memoizes
D
, which is re-used in subsequent calls. This shape must be known prior to
graph execution (which is the case if using tf.layers).
Examples
tfd = tfp.distributions
tfb = tfp.bijectors
# A common choice for a normalizing flow is to use a Gaussian for the base
# distribution. (However, any continuous distribution would work.) E.g.,
nvp = tfd.TransformedDistribution(
distribution=tfd.MultivariateNormalDiag(loc=[0., 0., 0.]),
bijector=tfb.RealNVP(
num_masked=2,
shift_and_log_scale_fn=tfb.real_nvp_default_template(
hidden_layers=[512, 512])))
x = nvp.sample()
nvp.log_prob(x)
nvp.log_prob([0.0, 0.0, 0.0])
For more examples, see [Jang (2018)][3].
References
[1]: Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density Estimation using Real NVP. In International Conference on Learning Representations, 2017. https://arxiv.org/abs/1605.08803
[2]: Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: Non-linear Independent Components Estimation. arXiv preprint arXiv:1410.8516, 2014. https://arxiv.org/abs/1410.8516
[3]: Eric Jang. Normalizing Flows Tutorial, Part 2: Modern Normalizing Flows. Technical Report, 2018. http://blog.evjang.com/2018/01/nf2.html
[4]: George Papamakarios, Theo Pavlakou, and Iain Murray. Masked Autoregressive Flow for Density Estimation. In Neural Information Processing Systems, 2017. https://arxiv.org/abs/1705.07057
[5]: Diederik P Kingma, Tim Salimans, Max Welling. Improving Variational Inference with Inverse Autoregressive Flow. In Neural Information Processing Systems, 2016. https://arxiv.org/abs/1606.04934
[6]: Conor Durkan, Artur Bekasov, Iain Murray, George Papamakarios. Neural Spline Flows, 2019. http://arxiv.org/abs/1906.04032
Raises | |
---|---|
ValueError
|
If both or none of shift_and_log_scale_fn and bijector_fn
are specified.
|
Methods
copy
copy(
**override_parameters_kwargs
)
Creates a copy of the bijector.
Args | |
---|---|
**override_parameters_kwargs
|
String/value dictionary of initialization arguments to override with new values. |
Returns | |
---|---|
bijector
|
A new instance of type(self) initialized from the union
of self.parameters and override_parameters_kwargs, i.e.,
dict(self.parameters, **override_parameters_kwargs) .
|
experimental_batch_shape
experimental_batch_shape(
x_event_ndims=None, y_event_ndims=None
)
Returns the batch shape of this bijector for inputs of the given rank.
The batch shape of a bijector decribes the set of distinct
transformations it represents on events of a given size. For example: the
bijector tfb.Scale([1., 2.])
has batch shape [2]
for scalar events
(event_ndims = 0
), because applying it to a scalar event produces
two scalar outputs, the result of two different scaling transformations.
The same bijector has batch shape []
for vector events, because applying
it to a vector produces (via elementwise multiplication) a single vector
output.
Bijectors that operate independently on multiple state parts, such as
tfb.JointMap
, must broadcast to a coherent batch shape. Some events may
not be valid: for example, the bijector
tfd.JointMap([tfb.Scale([1., 2.]), tfb.Scale([1., 2., 3.])])
does not
produce a valid batch shape when event_ndims = [0, 0]
, since the batch
shapes of the two parts are inconsistent. The same bijector
does define valid batch shapes of []
, [2]
, and [3]
if event_ndims
is [1, 1]
, [0, 1]
, or [1, 0]
, respectively.
Since transforming a single event produces a scalar log-det-Jacobian, the
batch shape of a bijector with non-constant Jacobian is expected to equal
the shape of forward_log_det_jacobian(x, event_ndims=x_event_ndims)
or inverse_log_det_jacobian(y, event_ndims=y_event_ndims)
, for x
or y
of the specified ndims
.
Args | |
---|---|
x_event_ndims
|
Optional Python int (structure) number of dimensions in
a probabilistic event passed to forward ; this must be greater than
or equal to self.forward_min_event_ndims . If None , defaults to
self.forward_min_event_ndims . Mutually exclusive with y_event_ndims .
Default value: None .
|
y_event_ndims
|
Optional Python int (structure) number of dimensions in
a probabilistic event passed to inverse ; this must be greater than
or equal to self.inverse_min_event_ndims . Mutually exclusive with
x_event_ndims .
Default value: None .
|
Returns | |
---|---|
batch_shape
|
TensorShape batch shape of this bijector for a
value with the given event rank. May be unknown or partially defined.
|
experimental_batch_shape_tensor
experimental_batch_shape_tensor(
x_event_ndims=None, y_event_ndims=None
)
Returns the batch shape of this bijector for inputs of the given rank.
The batch shape of a bijector decribes the set of distinct
transformations it represents on events of a given size. For example: the
bijector tfb.Scale([1., 2.])
has batch shape [2]
for scalar events
(event_ndims = 0
), because applying it to a scalar event produces
two scalar outputs, the result of two different scaling transformations.
The same bijector has batch shape []
for vector events, because applying
it to a vector produces (via elementwise multiplication) a single vector
output.
Bijectors that operate independently on multiple state parts, such as
tfb.JointMap
, must broadcast to a coherent batch shape. Some events may
not be valid: for example, the bijector
tfd.JointMap([tfb.Scale([1., 2.]), tfb.Scale([1., 2., 3.])])
does not
produce a valid batch shape when event_ndims = [0, 0]
, since the batch
shapes of the two parts are inconsistent. The same bijector
does define valid batch shapes of []
, [2]
, and [3]
if event_ndims
is [1, 1]
, [0, 1]
, or [1, 0]
, respectively.
Since transforming a single event produces a scalar log-det-Jacobian, the
batch shape of a bijector with non-constant Jacobian is expected to equal
the shape of forward_log_det_jacobian(x, event_ndims=x_event_ndims)
or inverse_log_det_jacobian(y, event_ndims=y_event_ndims)
, for x
or y
of the specified ndims
.
Args | |
---|---|
x_event_ndims
|
Optional Python int (structure) number of dimensions in
a probabilistic event passed to forward ; this must be greater than
or equal to self.forward_min_event_ndims . If None , defaults to
self.forward_min_event_ndims . Mutually exclusive with y_event_ndims .
Default value: None .
|
y_event_ndims
|
Optional Python int (structure) number of dimensions in
a probabilistic event passed to inverse ; this must be greater than
or equal to self.inverse_min_event_ndims . Mutually exclusive with
x_event_ndims .
Default value: None .
|
Returns | |
---|---|
batch_shape_tensor
|
integer Tensor batch shape of this bijector for a
value with the given event rank.
|
experimental_compute_density_correction
experimental_compute_density_correction(
x, tangent_space, backward_compat=False, **kwargs
)
Density correction for this transformation wrt the tangent space, at x.
Subclasses of Bijector may call the most specific applicable
method of TangentSpace
, based on whether the transformation is
dimension-preserving, coordinate-wise, a projection, or something
more general. The backward-compatible assumption is that the
transformation is dimension-preserving (goes from R^n to R^n).
Args | |
---|---|
x
|
Tensor (structure). The point at which to calculate the density.
|
tangent_space
|
TangentSpace or one of its subclasses. The tangent to
the support manifold at x .
|
backward_compat
|
bool specifying whether to assume that the Bijector
is dimension-preserving.
|
**kwargs
|
Optional keyword arguments forwarded to tangent space methods. |
Returns | |
---|---|
density_correction
|
Tensor representing the density correction---in log
space---under the transformation that this Bijector denotes.
|
Raises | |
---|---|
TypeError if backward_compat is False but no method of
TangentSpace has been called explicitly.
|
forward
forward(
x, **kwargs
)
forward_dtype
forward_dtype(
dtype=UNSPECIFIED, name='forward_dtype', **kwargs
)
Returns the dtype returned by forward
for the provided input.
forward_event_ndims
forward_event_ndims(
event_ndims, **kwargs
)
Returns the number of event dimensions produced by forward
.
Args | |
---|---|
event_ndims
|
Structure of Python and/or Tensor int s, and/or None
values. The structure should match that of
self.forward_min_event_ndims , and all non-None values must be
greater than or equal to the corresponding value in
self.forward_min_event_ndims .
|
**kwargs
|
Optional keyword arguments forwarded to nested bijectors. |
Returns | |
---|---|
forward_event_ndims
|
Structure of integers and/or None values matching
self.inverse_min_event_ndims . These are computed using 'prefer static'
semantics: if any inputs are None , some or all of the outputs may be
None , indicating that the output dimension could not be inferred
(conversely, if all inputs are non-None , all outputs will be
non-None ). If all input event_ndims are Python int s, all of the
(non-None ) outputs will be Python int s; otherwise, some or
all of the outputs may be Tensor int s.
|
forward_event_shape
forward_event_shape(
input_shape
)
Shape of a single sample from a single batch as a TensorShape
.
Same meaning as forward_event_shape_tensor
. May be only partially defined.
Args | |
---|---|
input_shape
|
TensorShape (structure) indicating event-portion shape
passed into forward function.
|
Returns | |
---|---|
forward_event_shape_tensor
|
TensorShape (structure) indicating
event-portion shape after applying forward . Possibly unknown.
|
forward_event_shape_tensor
forward_event_shape_tensor(
input_shape, name='forward_event_shape_tensor'
)
Shape of a single sample from a single batch as an int32
1D Tensor
.
Args | |
---|---|
input_shape
|
Tensor , int32 vector (structure) indicating event-portion
shape passed into forward function.
|
name
|
name to give to the op |
Returns | |
---|---|
forward_event_shape_tensor
|
Tensor , int32 vector (structure)
indicating event-portion shape after applying forward .
|
forward_log_det_jacobian
forward_log_det_jacobian(
x, event_ndims=None, name='forward_log_det_jacobian', **kwargs
)
Returns both the forward_log_det_jacobian.
Args | |
---|---|
x
|
Tensor (structure). The input to the 'forward' Jacobian determinant
evaluation.
|
event_ndims
|
Optional number of dimensions in the probabilistic events
being transformed; this must be greater than or equal to
self.forward_min_event_ndims . If event_ndims is specified, the
log Jacobian determinant is summed to produce a
scalar log-determinant for each event. Otherwise
(if event_ndims is None ), no reduction is performed.
Multipart bijectors require structured event_ndims, such that the
batch rank rank(y[i]) - event_ndims[i] is the same for all
elements i of the structured input. In most cases (with the
exception of tfb.JointMap ) they further require that
event_ndims[i] - self.inverse_min_event_ndims[i] is the same for
all elements i of the structured input.
Default value: None (equivalent to self.forward_min_event_ndims ).
|
name
|
The name to give this op. |
**kwargs
|
Named arguments forwarded to subclass implementation. |
Returns | |
---|---|
Tensor (structure), if this bijector is injective.
If not injective this is not implemented.
|
Raises | |
---|---|
TypeError
|
if y 's dtype is incompatible with the expected output dtype.
|
NotImplementedError
|
if neither _forward_log_det_jacobian
nor {_inverse , _inverse_log_det_jacobian } are implemented, or
this is a non-injective bijector.
|
ValueError
|
if the value of event_ndims is not valid for this bijector.
|
inverse
inverse(
x, **kwargs
)
inverse_dtype
inverse_dtype(
dtype=UNSPECIFIED, name='inverse_dtype', **kwargs
)
Returns the dtype returned by inverse
for the provided input.
inverse_event_ndims
inverse_event_ndims(
event_ndims, **kwargs
)
Returns the number of event dimensions produced by inverse
.
Args | |
---|---|
event_ndims
|
Structure of Python and/or Tensor int s, and/or None
values. The structure should match that of
self.inverse_min_event_ndims , and all non-None values must be
greater than or equal to the corresponding value in
self.inverse_min_event_ndims .
|
**kwargs
|
Optional keyword arguments forwarded to nested bijectors. |
Returns | |
---|---|
inverse_event_ndims
|
Structure of integers and/or None values matching
self.forward_min_event_ndims . These are computed using 'prefer static'
semantics: if any inputs are None , some or all of the outputs may be
None , indicating that the output dimension could not be inferred
(conversely, if all inputs are non-None , all outputs will be
non-None ). If all input event_ndims are Python int s, all of the
(non-None ) outputs will be Python int s; otherwise, some or
all of the outputs may be Tensor int s.
|
inverse_event_shape
inverse_event_shape(
output_shape
)
Shape of a single sample from a single batch as a TensorShape
.
Same meaning as inverse_event_shape_tensor
. May be only partially defined.
Args | |
---|---|
output_shape
|
TensorShape (structure) indicating event-portion shape
passed into inverse function.
|
Returns | |
---|---|
inverse_event_shape_tensor
|
TensorShape (structure) indicating
event-portion shape after applying inverse . Possibly unknown.
|
inverse_event_shape_tensor
inverse_event_shape_tensor(
output_shape, name='inverse_event_shape_tensor'
)
Shape of a single sample from a single batch as an int32
1D Tensor
.
Args | |
---|---|
output_shape
|
Tensor , int32 vector (structure) indicating
event-portion shape passed into inverse function.
|
name
|
name to give to the op |
Returns | |
---|---|
inverse_event_shape_tensor
|
Tensor , int32 vector (structure)
indicating event-portion shape after applying inverse .
|
inverse_log_det_jacobian
inverse_log_det_jacobian(
y, event_ndims=None, name='inverse_log_det_jacobian', **kwargs
)
Returns the (log o det o Jacobian o inverse)(y).
Mathematically, returns: log(det(dX/dY))(Y)
. (Recall that: X=g^{-1}(Y)
.)
Note that forward_log_det_jacobian
is the negative of this function,
evaluated at g^{-1}(y)
.
Args | |
---|---|
y
|
Tensor (structure). The input to the 'inverse' Jacobian determinant
evaluation.
|
event_ndims
|
Optional number of dimensions in the probabilistic events
being transformed; this must be greater than or equal to
self.inverse_min_event_ndims . If event_ndims is specified, the
log Jacobian determinant is summed to produce a
scalar log-determinant for each event. Otherwise
(if event_ndims is None ), no reduction is performed.
Multipart bijectors require structured event_ndims, such that the
batch rank rank(y[i]) - event_ndims[i] is the same for all
elements i of the structured input. In most cases (with the
exception of tfb.JointMap ) they further require that
event_ndims[i] - self.inverse_min_event_ndims[i] is the same for
all elements i of the structured input.
Default value: None (equivalent to self.inverse_min_event_ndims ).
|
name
|
The name to give this op. |
**kwargs
|
Named arguments forwarded to subclass implementation. |
Returns | |
---|---|
ildj
|
Tensor , if this bijector is injective.
If not injective, returns the tuple of local log det
Jacobians, log(det(Dg_i^{-1}(y))) , where g_i is the restriction
of g to the ith partition Di .
|
Raises | |
---|---|
TypeError
|
if x 's dtype is incompatible with the expected inverse-dtype.
|
NotImplementedError
|
if _inverse_log_det_jacobian is not implemented.
|
ValueError
|
if the value of event_ndims is not valid for this bijector.
|
parameter_properties
@classmethod
parameter_properties( dtype=tf.float32 )
Returns a dict mapping constructor arg names to property annotations.
This dict should include an entry for each of the bijector's
Tensor
-valued constructor arguments.
Args | |
---|---|
dtype
|
Optional float dtype to assume for continuous-valued parameters.
Some constraining bijectors require advance knowledge of the dtype
because certain constants (e.g., tfb.Softplus.low ) must be
instantiated with the same dtype as the values to be transformed.
|
Returns | |
---|---|
parameter_properties
|
A
str -> tfp.python.internal.parameter_properties.ParameterPropertiesdict mapping constructor argument names to ParameterProperties`
instances.
|
__call__
__call__(
value, name=None, **kwargs
)
Applies or composes the Bijector
, depending on input type.
This is a convenience function which applies the Bijector
instance in
three different ways, depending on the input:
- If the input is a
tfd.Distribution
instance, returntfd.TransformedDistribution(distribution=input, bijector=self)
. - If the input is a
tfb.Bijector
instance, returntfb.Chain([self, input])
. - Otherwise, return
self.forward(input)
Args | |
---|---|
value
|
A tfd.Distribution , tfb.Bijector , or a (structure of) Tensor .
|
name
|
Python str name given to ops created by this function.
|
**kwargs
|
Additional keyword arguments passed into the created
tfd.TransformedDistribution , tfb.Bijector , or self.forward .
|
Returns | |
---|---|
composition
|
A tfd.TransformedDistribution if the input was a
tfd.Distribution , a tfb.Chain if the input was a tfb.Bijector , or
a (structure of) Tensor computed by self.forward .
|
Examples
sigmoid = tfb.Reciprocal()(
tfb.Shift(shift=1.)(
tfb.Exp()(
tfb.Scale(scale=-1.))))
# ==> `tfb.Chain([
# tfb.Reciprocal(),
# tfb.Shift(shift=1.),
# tfb.Exp(),
# tfb.Scale(scale=-1.),
# ])` # ie, `tfb.Sigmoid()`
log_normal = tfb.Exp()(tfd.Normal(0, 1))
# ==> `tfd.TransformedDistribution(tfd.Normal(0, 1), tfb.Exp())`
tfb.Exp()([-1., 0., 1.])
# ==> tf.exp([-1., 0., 1.])
__eq__
__eq__(
other
)
Return self==value.
__getitem__
__getitem__(
slices
)
__iter__
__iter__()