CouplingLayer¶

class tfsnippet.layers.CouplingLayer(shift_and_scale_fn, axis=-1, value_ndims=1, secondary=False, scale_type='linear', sigmoid_scale_bias=2.0, epsilon=1e-06, name=None, scope=None)¶

Bases: tfsnippet.layers.flows.base.FeatureMappingFlow

A general implementation of the coupling layer (Dinh et al., 2016).

Basically, a CouplingLayer does the following transformation:

x1, x2 = split(x)
if secondary:
    x1, x2 = x2, x1

y1 = x1

shift, scale = shift_and_scale_fn(x1, x2.shape[axis])
if scale_type == 'exp':
    y2 = (x2 + shift) * exp(scale)
elif scale_type == 'sigmoid':
    y2 = (x2 + shift) * sigmoid(scale + sigmoid_scale_bias)
elif scale_type == 'linear':
    y2 = (x2 + shift) * scale
else:
    y2 = x2 + shift

if secondary:
    y1, y2 = y2, y1
y = tf.concat([y1, y2], axis=axis)

The inverse transformation, and the log-determinants are computed according to the above transformation, respectively.

Attributes Summary

`axis`	Get the feature axis/axes.
`explicitly_invertible`	Whether or not this flow is explicitly invertible?
`name`	Get the name of this object.
`require_batch_dims`	Whether or not this flow requires batch dimensions.
`value_ndims`	Get the number of value dimensions in both x and y.
`variable_scope`	Get the variable scope of this object.
`x_value_ndims`	Get the number of value dimensions in x.
`y_value_ndims`	Get the number of value dimensions in y.

Methods Summary

`__call__`(…) <==> x(…)
`apply`(input)	Apply the layer on input, to produce output.
`build`([input])	Build the layer, creating all required variables.
`inverse_transform`(y[, compute_x, …])	Transform y into x, and compute the log-determinant of f^{-1} at y (i.e., \(\log \det \frac{\partial f^{-1}(y)}{\partial y}\)).
`invert`()	Get the inverted flow from this flow.
`transform`(x[, compute_y, compute_log_det, name])	Transform x into y, and compute the log-determinant of f at x (i.e., \(\log \det \frac{\partial f(x)}{\partial x}\)).

Attributes Documentation

axis¶

Get the feature axis/axes.

Returns:	The feature axis/axes, as is specified in the constructor.
Return type:	int or tuple[int]

explicitly_invertible¶

Whether or not this flow is explicitly invertible?

If a flow is not explicitly invertible, then it only supports to transform x into y, and corresponding \(\log p(x)\) into \(\log p(y)\). It cannot compute \(\log p(y)\) directly without knowing x, nor can it transform x back into y.

Returns:	A boolean indicating whether or not the flow is explicitly invertible.
Return type:	bool

name¶: Get the name of this object.

require_batch_dims¶: Whether or not this flow requires batch dimensions.

value_ndims¶

Get the number of value dimensions in both x and y.

Returns:	The number of value dimensions in both x and y.
Return type:	int

variable_scope¶: Get the variable scope of this object.

x_value_ndims¶

Get the number of value dimensions in x.

Returns:	The number of value dimensions in x.
Return type:	int

y_value_ndims¶

Get the number of value dimensions in y.

Returns:	The number of value dimensions in y.
Return type:	int

Methods Documentation

__call__(...) <==> x(...)¶

apply(input)¶

Apply the layer on input, to produce output.

Parameters:	input (Tensor or list[Tensor]) – The input tensor, or a list of input tensors.
Returns:	The output tensor, or a list of output tensors.

build(input=None)¶

Build the layer, creating all required variables.

Parameters:	input (Tensor or list[Tensor] or None) – If `build()` is called within `apply()`, it will be the input tensor(s). Otherwise if it is called separately, it will be `None`.

inverse_transform(y, compute_x=True, compute_log_det=True, name=None)¶

Transform y into x, and compute the log-determinant of f^{-1} at y (i.e., \(\log \det \frac{\partial f^{-1}(y)}{\partial y}\)).

Parameters:	y (Tensor) – The samples of y. compute_x (bool) – Whether or not to compute \(x = f^{-1}(y)\)? Default `True`. compute_log_det (bool) – Whether or not to compute the log-determinant? Default `True`. name (str) – If specified, will use this name as the TensorFlow operational name scope.
Returns:	x and the (maybe summed) log-determinant. The items in the returned tuple might be `None` if corresponding compute_? argument is set to `False`.
Return type:	(tf.Tensor, tf.Tensor)
Raises:	`RuntimeError` – If both compute_x and compute_log_det are set to `False`. `RuntimeError` – If the flow is not explicitly invertible.

invert()¶

Get the inverted flow from this flow.

The transform() will become the inverse_transform() in the inverted flow, and the inverse_transform() will become the transform() in the inverted flow.

If the current flow has not been initialized, it must be initialized via inverse_transform() in the new flow.

Returns:	The inverted flow.
Return type:	tfsnippet.layers.InvertFlow

transform(x, compute_y=True, compute_log_det=True, name=None)¶

Transform x into y, and compute the log-determinant of f at x (i.e., \(\log \det \frac{\partial f(x)}{\partial x}\)).

Parameters:	x (Tensor) – The samples of x. compute_y (bool) – Whether or not to compute \(y = f(x)\)? Default `True`. compute_log_det (bool) – Whether or not to compute the log-determinant? Default `True`. name (str) – If specified, will use this name as the TensorFlow operational name scope.
Returns:	y and the (maybe summed) log-determinant. The items in the returned tuple might be `None` if corresponding compute_? argument is set to `False`.
Return type:	(tf.Tensor, tf.Tensor)
Raises:	`RuntimeError` – If both compute_y and compute_log_det are set to `False`.