tfep.nn.flows

Normalizing flow models for PyTorch.

All the layers defined in this module are invertible and implement an inverse() method (not to be comfused with the Tensor’s backward() method which backpropagate the gradients).

The forward propagation of the modules here return both the transformation of the input plus the log determinant of the Jacobian.

Modules

autoregressive	Base autoregressive flow layer for PyTorch.
centroid	Transformation that constrains the (weighted) centroid of the DOFs.
continuous	Continuous normalizing flow layer for PyTorch.
maf	Masked autoregressive flow layer for PyTorch.
oriented	Transformation that constrains the rotational degrees of freedom.
partial	Normalizing flow mapping only a subset of the input degrees of freedom.
pca	Normalizing flow transforming to and from PCA-whitened space.
sequential	Normalizing flow concatenating multiple normalizing flows.