bijx.Bijection

class bijx.Bijection

Bases: Module

Base class for all bijective transformations.

A bijection represents an invertible mapping between two spaces, equipped with forward and reverse transformations that properly track log-density changes according to the change of variables formula.

Note

This is an abstract base class. Subclasses must override forward() and reverse() methods to implement the actual transformation logic.

__init__(*args, **kwargs)

Methods

apply(x, log_density[, reverse])	Apply the bijection in either direction.
forward(x, log_density, **kwargs)	Apply forward transformation.
invert()	Create an inverted version of this bijection.
reverse(x, log_density, **kwargs)	Apply reverse (inverse) transformation.

forward(x, log_density, **kwargs)

Apply forward transformation.

Transforms input through the bijection and updates log-density according to the change of variables formula.

For convenience Bijection() gives the default identity bijection.

Parameters:

• x – Input data of any pytree structure.

• log_density – Log density values corresponding to the input.

• **kwargs – Additional transformation-specific arguments.

Returns:

Tuple of (transformed_data, updated_log_density) where the log-density incorporates the log absolute determinant of the transformation Jacobian.

reverse(x, log_density, **kwargs)

Apply reverse (inverse) transformation.

Transforms input through the inverse bijection and updates log-density accordingly.

Parameters:

• x – Input data of any pytree structure.

• log_density – Log density values corresponding to the input.

• **kwargs – Additional transformation-specific arguments.

Returns:

Tuple of (inverse_transformed_data, updated_log_density) where the log-density change has the opposite sign compared to forward().

invert()

Create an inverted version of this bijection.

Returns:

New bijection where forward() and reverse() methods are swapped. See Inverse.

apply(x, log_density, reverse=False, **kwargs)

Apply the bijection in either direction.

Dispatches to forward() or reverse() based on reverse. If reverse is a Python bool the dispatch is static (only the chosen branch is traced); otherwise jax.lax.cond is used so a traced boolean array is supported, at the cost of tracing both branches.

Parameters:

• x – Input data of any pytree structure.

• log_density – Log density values corresponding to the input.

• reverse – If True, apply reverse transformation; if False, forward. May be a Python bool or a JAX boolean scalar.

• **kwargs – Additional transformation-specific arguments.

Returns:

Tuple of (transformed_data, updated_log_density).