TF 2.0 is out! Get hands-on practice at TF World, Oct 28-31. Use code TF20 for 20% off select passes. Register now

tfp.experimental.substrates.jax.distributions.QuantizedDistribution

View source on GitHub

Class QuantizedDistribution

Distribution representing the quantization Y = ceiling(X).

Inherits From: Distribution

Definition in Terms of Sampling

1. Draw X
2. Set Y <-- ceiling(X)
3. If Y < low, reset Y <-- low
4. If Y > high, reset Y <-- high
5. Return Y

Definition in Terms of the Probability Mass Function

Given scalar random variable X, we define a discrete random variable Y supported on the integers as follows:

P[Y = j] := P[X <= low],  if j == low,
         := P[X > high - 1],  j == high,
         := 0, if j < low or j > high,
         := P[j - 1 < X <= j],  all other j.

Conceptually, without cutoffs, the quantization process partitions the real line R into half open intervals, and identifies an integer j with the right endpoints:

R = ... (-2, -1](-1, 0](0, 1](1, 2](2, 3](3, 4] ...
j = ...      -1      0     1     2     3     4  ...

P[Y = j] is the mass of X within the jth interval. If low = 0, and high = 2, then the intervals are redrawn and j is re-assigned:

R = (-infty, 0](0, 1](1, infty)
j =          0     1     2

P[Y = j] is still the mass of X within the jth interval.

Examples

We illustrate a mixture of discretized logistic distributions [(Salimans et al., 2017)][1]. This is used, for example, for capturing 16-bit audio in WaveNet [(van den Oord et al., 2017)][2]. The values range in a 1-D integer domain of [0, 2**16-1], and the discretization captures P(x - 0.5 < X <= x + 0.5) for all x in the domain excluding the endpoints. The lowest value has probability P(X <= 0.5) and the highest value has probability P(2**16 - 1.5 < X).

Below we assume a wavenet function. It takes as input right-shifted audio samples of shape [..., sequence_length]. It returns a real-valued tensor of shape [..., num_mixtures * 3], i.e., each mixture component has a loc and scale parameter belonging to the logistic distribution, and a logits parameter determining the unnormalized probability of that component.

tfd = tfp.distributions
tfb = tfp.bijectors

net = wavenet(inputs)
loc, unconstrained_scale, logits = tf.split(net,
                                            num_or_size_splits=3,
                                            axis=-1)
scale = tf.math.softplus(unconstrained_scale)

# Form mixture of discretized logistic distributions. Note we shift the
# logistic distribution by -0.5. This lets the quantization capture "rounding"
# intervals, `(x-0.5, x+0.5]`, and not "ceiling" intervals, `(x-1, x]`.
discretized_logistic_dist = tfd.QuantizedDistribution(
    distribution=tfd.TransformedDistribution(
        distribution=tfd.Logistic(loc=loc, scale=scale),
        bijector=tfb.AffineScalar(shift=-0.5)),
    low=0.,
    high=2**16 - 1.)
mixture_dist = tfd.MixtureSameFamily(
    mixture_distribution=tfd.Categorical(logits=logits),
    components_distribution=discretized_logistic_dist)

neg_log_likelihood = -tf.reduce_sum(mixture_dist.log_prob(targets))
train_op = tf.train.AdamOptimizer().minimize(neg_log_likelihood)

After instantiating mixture_dist, we illustrate maximum likelihood by calculating its log-probability of audio samples as target and optimizing.

References

[1]: Tim Salimans, Andrej Karpathy, Xi Chen, and Diederik P. Kingma. PixelCNN++: Improving the PixelCNN with discretized logistic mixture likelihood and other modifications. International Conference on Learning Representations, 2017. https://arxiv.org/abs/1701.05517 [2]: Aaron van den Oord et al. Parallel WaveNet: Fast High-Fidelity Speech Synthesis. arXiv preprint arXiv:1711.10433, 2017. https://arxiv.org/abs/1711.10433

__init__

__init__(
    distribution,
    low=None,
    high=None,
    validate_args=False,
    name='QuantizedDistribution'
)

Construct a Quantized Distribution representing Y = ceiling(X).

Some properties are inherited from the distribution defining X. Example: allow_nan_stats is determined for this QuantizedDistribution by reading the distribution.

Args:

  • distribution: The base distribution class to transform. Typically an instance of Distribution.
  • low: Tensor with same dtype as this distribution and shape able to be added to samples. Should be a whole number. Default None. If provided, base distribution's prob should be defined at low.
  • high: Tensor with same dtype as this distribution and shape able to be added to samples. Should be a whole number. Default None. If provided, base distribution's prob should be defined at high - 1. high must be strictly greater than low.
  • validate_args: Python bool, default False. When True distribution parameters are checked for validity despite possibly degrading runtime performance. When False invalid inputs may silently render incorrect outputs.
  • name: Python str name prefixed to Ops created by this class.

Raises:

  • TypeError: If dist_cls is not a subclass of Distribution or continuous.
  • NotImplementedError: If the base distribution does not implement cdf.

Properties

allow_nan_stats

Python bool describing behavior when a stat is undefined.

Stats return +/- infinity when it makes sense. E.g., the variance of a Cauchy distribution is infinity. However, sometimes the statistic is undefined, e.g., if a distribution's pdf does not achieve a maximum within the support of the distribution, the mode is undefined. If the mean is undefined, then by definition the variance is undefined. E.g. the mean for Student's T for df = 1 is undefined (no clear way to say it is either + or - infinity), so the variance = E[(X - mean)**2] is also undefined.

Returns:

  • allow_nan_stats: Python bool.

batch_shape

Shape of a single sample from a single event index as a TensorShape.

May be partially defined or unknown.

The batch dimensions are indexes into independent, non-identical parameterizations of this distribution.

Returns:

  • batch_shape: TensorShape, possibly unknown.

distribution

Base distribution, p(x).

dtype

The DType of Tensors handled by this Distribution.

event_shape

Shape of a single sample from a single batch as a TensorShape.

May be partially defined or unknown.

Returns:

  • event_shape: TensorShape, possibly unknown.

high

Highest value that quantization returns.

low

Lowest value that quantization returns.

name

Name prepended to all ops created by this Distribution.

parameters

Dictionary of parameters used to instantiate this Distribution.

reparameterization_type

Describes how samples from the distribution are reparameterized.

Currently this is one of the static instances tfd.FULLY_REPARAMETERIZED or tfd.NOT_REPARAMETERIZED.

Returns:

An instance of ReparameterizationType.

trainable_variables

validate_args

Python bool indicating possibly expensive checks are enabled.

variables

Methods

__getitem__

View source

__getitem__(slices)

Slices the batch axes of this distribution, returning a new instance.

b = tfd.Bernoulli(logits=tf.zeros([3, 5, 7, 9]))
b.batch_shape  # => [3, 5, 7, 9]
b2 = b[:, tf.newaxis, ..., -2:, 1::2]
b2.batch_shape  # => [3, 1, 5, 2, 4]

x = tf.random.normal([5, 3, 2, 2])
cov = tf.matmul(x, x, transpose_b=True)
chol = tf.cholesky(cov)
loc = tf.random.normal([4, 1, 3, 1])
mvn = tfd.MultivariateNormalTriL(loc, chol)
mvn.batch_shape  # => [4, 5, 3]
mvn.event_shape  # => [2]
mvn2 = mvn[:, 3:, ..., ::-1, tf.newaxis]
mvn2.batch_shape  # => [4, 2, 3, 1]
mvn2.event_shape  # => [2]

Args:

  • slices: slices from the [] operator

Returns:

  • dist: A new tfd.Distribution instance with sliced parameters.

__iter__

View source

__iter__()

batch_shape_tensor

View source

batch_shape_tensor(name='batch_shape_tensor')

Shape of a single sample from a single event index as a 1-D Tensor.

The batch dimensions are indexes into independent, non-identical parameterizations of this distribution.

Args:

  • name: name to give to the op

Returns:

  • batch_shape: Tensor.

cdf

View source

cdf(
    value,
    name='cdf',
    **kwargs
)

Cumulative distribution function.

Given random variable X, the cumulative distribution function cdf is:

cdf(x) := P[X <= x]

Additional documentation from QuantizedDistribution:

For whole numbers y,

cdf(y) := P[Y <= y]
        = 1, if y >= high,
        = 0, if y < low,
        = P[X <= y], otherwise.

Since Y only has mass at whole numbers, P[Y <= y] = P[Y <= floor(y)]. This dictates that fractional y are first floored to a whole number, and then above definition applies.

The base distribution's cdf method must be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • cdf: a Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

copy

View source

copy(**override_parameters_kwargs)

Creates a deep copy of the distribution.

Args:

  • **override_parameters_kwargs: String/value dictionary of initialization arguments to override with new values.

Returns:

  • distribution: 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).

covariance

View source

covariance(
    name='covariance',
    **kwargs
)

Covariance.

Covariance is (possibly) defined only for non-scalar-event distributions.

For example, for a length-k, vector-valued distribution, it is calculated as,

Cov[i, j] = Covariance(X_i, X_j) = E[(X_i - E[X_i]) (X_j - E[X_j])]

where Cov is a (batch of) k x k matrix, 0 <= (i, j) < k, and E denotes expectation.

Alternatively, for non-vector, multivariate distributions (e.g., matrix-valued, Wishart), Covariance shall return a (batch of) matrices under some vectorization of the events, i.e.,

Cov[i, j] = Covariance(Vec(X)_i, Vec(X)_j) = [as above]

where Cov is a (batch of) k' x k' matrices, 0 <= (i, j) < k' = reduce_prod(event_shape), and Vec is some function mapping indices of this distribution's event dimensions to indices of a length-k' vector.

Args:

  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • covariance: Floating-point Tensor with shape [B1, ..., Bn, k', k'] where the first n dimensions are batch coordinates and k' = reduce_prod(self.event_shape).

cross_entropy

View source

cross_entropy(
    other,
    name='cross_entropy'
)

Computes the (Shannon) cross entropy.

Denote this distribution (self) by P and the other distribution by Q. Assuming P, Q are absolutely continuous with respect to one another and permit densities p(x) dr(x) and q(x) dr(x), (Shannon) cross entropy is defined as:

H[P, Q] = E_p[-log q(X)] = -int_F p(x) log q(x) dr(x)

where F denotes the support of the random variable X ~ P.

Args:

Returns:

  • cross_entropy: self.dtype Tensor with shape [B1, ..., Bn] representing n different calculations of (Shannon) cross entropy.

entropy

View source

entropy(
    name='entropy',
    **kwargs
)

Shannon entropy in nats.

event_shape_tensor

View source

event_shape_tensor(name='event_shape_tensor')

Shape of a single sample from a single batch as a 1-D int32 Tensor.

Args:

  • name: name to give to the op

Returns:

  • event_shape: Tensor.

is_scalar_batch

View source

is_scalar_batch(name='is_scalar_batch')

Indicates that batch_shape == [].

Args:

  • name: Python str prepended to names of ops created by this function.

Returns:

  • is_scalar_batch: bool scalar Tensor.

is_scalar_event

View source

is_scalar_event(name='is_scalar_event')

Indicates that event_shape == [].

Args:

  • name: Python str prepended to names of ops created by this function.

Returns:

  • is_scalar_event: bool scalar Tensor.

kl_divergence

View source

kl_divergence(
    other,
    name='kl_divergence'
)

Computes the Kullback--Leibler divergence.

Denote this distribution (self) by p and the other distribution by q. Assuming p, q are absolutely continuous with respect to reference measure r, the KL divergence is defined as:

KL[p, q] = E_p[log(p(X)/q(X))]
         = -int_F p(x) log q(x) dr(x) + int_F p(x) log p(x) dr(x)
         = H[p, q] - H[p]

where F denotes the support of the random variable X ~ p, H[., .] denotes (Shannon) cross entropy, and H[.] denotes (Shannon) entropy.

Args:

Returns:

  • kl_divergence: self.dtype Tensor with shape [B1, ..., Bn] representing n different calculations of the Kullback-Leibler divergence.

log_cdf

View source

log_cdf(
    value,
    name='log_cdf',
    **kwargs
)

Log cumulative distribution function.

Given random variable X, the cumulative distribution function cdf is:

log_cdf(x) := Log[ P[X <= x] ]

Often, a numerical approximation can be used for log_cdf(x) that yields a more accurate answer than simply taking the logarithm of the cdf when x << -1.

Additional documentation from QuantizedDistribution:

For whole numbers y,

cdf(y) := P[Y <= y]
        = 1, if y >= high,
        = 0, if y < low,
        = P[X <= y], otherwise.

Since Y only has mass at whole numbers, P[Y <= y] = P[Y <= floor(y)]. This dictates that fractional y are first floored to a whole number, and then above definition applies.

The base distribution's log_cdf method must be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • logcdf: a Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

log_prob

View source

log_prob(
    value,
    name='log_prob',
    **kwargs
)

Log probability density/mass function.

Additional documentation from QuantizedDistribution:

For whole numbers y,

P[Y = y] := P[X <= low],  if y == low,
         := P[X > high - 1],  y == high,
         := 0, if j < low or y > high,
         := P[y - 1 < X <= y],  all other y.

The base distribution's log_cdf method must be defined on y - 1. If the base distribution has a log_survival_function method results will be more accurate for large values of y, and in this case the log_survival_function must also be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • log_prob: a Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

log_survival_function

View source

log_survival_function(
    value,
    name='log_survival_function',
    **kwargs
)

Log survival function.

Given random variable X, the survival function is defined:

log_survival_function(x) = Log[ P[X > x] ]
                         = Log[ 1 - P[X <= x] ]
                         = Log[ 1 - cdf(x) ]

Typically, different numerical approximations can be used for the log survival function, which are more accurate than 1 - cdf(x) when x >> 1.

Additional documentation from QuantizedDistribution:

For whole numbers y,

survival_function(y) := P[Y > y]
                      = 0, if y >= high,
                      = 1, if y < low,
                      = P[X <= y], otherwise.

Since Y only has mass at whole numbers, P[Y <= y] = P[Y <= floor(y)]. This dictates that fractional y are first floored to a whole number, and then above definition applies.

The base distribution's log_cdf method must be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

mean

View source

mean(
    name='mean',
    **kwargs
)

Mean.

mode

View source

mode(
    name='mode',
    **kwargs
)

Mode.

param_shapes

View source

param_shapes(
    cls,
    sample_shape,
    name='DistributionParamShapes'
)

Shapes of parameters given the desired shape of a call to sample().

This is a class method that describes what key/value arguments are required to instantiate the given Distribution so that a particular shape is returned for that instance's call to sample().

Subclasses should override class method _param_shapes.

Args:

  • sample_shape: Tensor or python list/tuple. Desired shape of a call to sample().
  • name: name to prepend ops with.

Returns:

dict of parameter name to Tensor shapes.

param_static_shapes

View source

param_static_shapes(
    cls,
    sample_shape
)

param_shapes with static (i.e. TensorShape) shapes.

This is a class method that describes what key/value arguments are required to instantiate the given Distribution so that a particular shape is returned for that instance's call to sample(). Assumes that the sample's shape is known statically.

Subclasses should override class method _param_shapes to return constant-valued tensors when constant values are fed.

Args:

  • sample_shape: TensorShape or python list/tuple. Desired shape of a call to sample().

Returns:

dict of parameter name to TensorShape.

Raises:

  • ValueError: if sample_shape is a TensorShape and is not fully defined.

prob

View source

prob(
    value,
    name='prob',
    **kwargs
)

Probability density/mass function.

Additional documentation from QuantizedDistribution:

For whole numbers y,

P[Y = y] := P[X <= low],  if y == low,
         := P[X > high - 1],  y == high,
         := 0, if j < low or y > high,
         := P[y - 1 < X <= y],  all other y.

The base distribution's cdf method must be defined on y - 1. If the base distribution has a survival_function method, results will be more accurate for large values of y, and in this case the survival_function must also be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • prob: a Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

quantile

View source

quantile(
    value,
    name='quantile',
    **kwargs
)

Quantile function. Aka 'inverse cdf' or 'percent point function'.

Given random variable X and p in [0, 1], the quantile is:

quantile(p) := x such that P[X <= x] == p

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • quantile: a Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

sample

View source

sample(
    sample_shape=(),
    seed=None,
    name='sample',
    **kwargs
)

Generate samples of the specified shape.

Note that a call to sample() without arguments will generate a single sample.

Args:

  • sample_shape: 0D or 1D int32 Tensor. Shape of the generated samples.
  • seed: Python integer seed for RNG
  • name: name to give to the op.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • samples: a Tensor with prepended dimensions sample_shape.

stddev

View source

stddev(
    name='stddev',
    **kwargs
)

Standard deviation.

Standard deviation is defined as,

stddev = E[(X - E[X])**2]**0.5

where X is the random variable associated with this distribution, E denotes expectation, and stddev.shape = batch_shape + event_shape.

Args:

  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • stddev: Floating-point Tensor with shape identical to batch_shape + event_shape, i.e., the same shape as self.mean().

survival_function

View source

survival_function(
    value,
    name='survival_function',
    **kwargs
)

Survival function.

Given random variable X, the survival function is defined:

survival_function(x) = P[X > x]
                     = 1 - P[X <= x]
                     = 1 - cdf(x).

Additional documentation from QuantizedDistribution:

For whole numbers y,

survival_function(y) := P[Y > y]
                      = 0, if y >= high,
                      = 1, if y < low,
                      = P[X <= y], otherwise.

Since Y only has mass at whole numbers, P[Y <= y] = P[Y <= floor(y)]. This dictates that fractional y are first floored to a whole number, and then above definition applies.

The base distribution's cdf method must be defined on y - 1.

Args:

  • value: float or double Tensor.
  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

Tensor of shape sample_shape(x) + self.batch_shape with values of type self.dtype.

variance

View source

variance(
    name='variance',
    **kwargs
)

Variance.

Variance is defined as,

Var = E[(X - E[X])**2]

where X is the random variable associated with this distribution, E denotes expectation, and Var.shape = batch_shape + event_shape.

Args:

  • name: Python str prepended to names of ops created by this function.
  • **kwargs: Named arguments forwarded to subclass implementation.

Returns:

  • variance: Floating-point Tensor with shape identical to batch_shape + event_shape, i.e., the same shape as self.mean().