gluonts.mx.distribution.bijection module#

class gluonts.mx.distribution.bijection.AffineTransformation(loc: Optional[Union[NDArray, Symbol]] = None, scale: Optional[Union[NDArray, Symbol]] = None)[source]#

Bases: Bijection

An affine transformation consisting of a scaling and a translation.

If translation is specified loc, and the scaling by scale, then this transformation computes y = scale * x + loc, where all operations are element-wise.

Parameters

loc – Translation parameter. If unspecified or None, this will be zero.
scale – Scaling parameter. If unspecified or None, this will be one.

property event_dim: int#

f(x: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Forward transformation x -> y.

f_inv(y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Inverse transformation y -> x.

log_abs_det_jac(x: Union[NDArray, Symbol], y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Receives (x, y) and returns log of the absolute value of the Jacobian determinant.

\log | d y / d x |

Note that this is the Jacobian determinant of the forward transformation x -> y.

property sign: Union[float, NDArray, Symbol]#: Return the sign of the Jacobian’s determinant.

class gluonts.mx.distribution.bijection.Bijection[source]#

Bases: object

A bijective transformation.

This is defined through the forward transformation (computed by the f method) and the inverse transformation (f_inv).

property event_dim: int#

f(x: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Forward transformation x -> y.

f_inv(y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Inverse transformation y -> x.

inverse_bijection() → Bijection[source]#: Returns a Bijection instance that represents the inverse of this transformation.

log_abs_det_jac(x: Union[NDArray, Symbol], y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Receives (x, y) and returns log of the absolute value of the Jacobian determinant.

\log | d y / d x |

Note that this is the Jacobian determinant of the forward transformation x -> y.

property sign: Union[float, NDArray, Symbol]#: Return the sign of the Jacobian’s determinant.

class gluonts.mx.distribution.bijection.BijectionHybridBlock(prefix=None, params=None)[source]#

Bases: HybridBlock, Bijection

Allows a Bijection to have parameters.

class gluonts.mx.distribution.bijection.ComposedBijection(bijections: Optional[List[Bijection]] = None)[source]#

Bases: Bijection

Encapsulates a series of bijections and implements functions associated to their composition.

property event_dim#

property event_shape#

f(x: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Computes the forward transform of the composition of bijections.

Parameters: x – Input Tensor for the forward transform.
Returns: Transformation of x by the forward composition of bijections
Return type: Tensor

f_inv(y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Computes the inverse transform of a composition of bijections.

Parameters: y – Input Tensor for the inverse function
Returns: Transformation of y by the inverse composition of bijections
Return type: Tensor

log_abs_det_jac(x: Union[NDArray, Symbol], y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Logarithm of the absolute value of the Jacobian determinant corresponding to the composed bijection.

Parameters

x – input of the forward transformation or output of the inverse transform
y – output of the forward transform or input of the inverse transform

Returns

Jacobian evaluated for x as input or y as output

Return type

Tensor

class gluonts.mx.distribution.bijection.ComposedBijectionHybridBlock(bij_blocks: Optional[List[Bijection]] = None, *args, **kwargs)[source]#

Bases: BijectionHybridBlock, ComposedBijection

Allows a ComposedBijection object to have parameters.

class gluonts.mx.distribution.bijection.InverseBijection(bijection: Bijection)[source]#

Bases: Bijection

The inverse of a given transformation.

This is a wrapper around bijective transformations, that inverts the role of f and f_inv, and modifies other related methods accordingly.

Parameters: bijection – The transformation to invert.

property event_dim: int#

f(x: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Forward transformation x -> y.

f_inv(y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#: Inverse transformation y -> x.

inverse_bijection() → Bijection[source]#: Returns a Bijection instance that represents the inverse of this transformation.

log_abs_det_jac(x: Union[NDArray, Symbol], y: Union[NDArray, Symbol]) → Union[NDArray, Symbol][source]#

Receives (x, y) and returns log of the absolute value of the Jacobian determinant.

\log | d y / d x |

Note that this is the Jacobian determinant of the forward transformation x -> y.

property sign: Union[float, NDArray, Symbol]#: Return the sign of the Jacobian’s determinant.