Parameter Containers

Flexible parameter containers that integrate with BrainState’s module system. Param supports bijective transformations for constrained optimization and optional regularization. Const provides non-trainable constant parameters. Both support automatic caching of transformed values for improved performance.

Param	A module has neural network parameters for optional transform and regularization.
Const	A module has non-trainable constant parameter.

Parameter Transforms

Bijective transformations for constrained parameter optimization. These transforms map between unconstrained and constrained spaces, enabling gradient-based optimization of parameters with constraints (positivity, boundedness, simplex, etc.). All transforms implement forward(), inverse(), and optional log_abs_det_jacobian() for probabilistic applications. Use with Param for automatic constraint handling.

Transform	Abstract base class for bijective parameter transformations.
IdentityT	Identity transformation (no-op).
ClipT	Transformation with clipping to specified bounds.
AffineT	Affine (linear) transformation with scaling and shifting.
SigmoidT	Sigmoid transformation mapping unbounded values to a bounded interval.
TanhT	Tanh-based transformation mapping (-inf, +inf) to (lower, upper).
SoftsignT	Softsign-based transformation mapping (-inf, +inf) to (lower, upper).
ScaledSigmoidT	Sigmoid transformation with adjustable sharpness/temperature.
SoftplusT	Softplus transformation mapping unbounded values to positive semi-infinite interval.
NegSoftplusT	Negative softplus transformation mapping unbounded values to negative semi-infinite interval.
LogT	Log transformation mapping (lower, +inf) to (-inf, +inf).
ExpT	Exponential transformation mapping (-inf, +inf) to (lower, +inf).
ReluT	ReLU transform with lower bound: forward(x) = relu(x) + lower_bound
PositiveT	Transformation constraining parameters to be strictly positive (0, +∞).
NegativeT	Transformation constraining parameters to be strictly negative (-∞, 0).
PowerT	Power (Box-Cox) transformation for stabilizing variance.
OrderedT	Transformation ensuring ordered (monotonically increasing) output.
SimplexT	Stick-breaking transformation for simplex constraint.
UnitVectorT	Transformation to unit vectors (L2 norm = 1).
ChainT	Composition of multiple transformations applied sequentially.
MaskedT	Selective transformation using a boolean mask.