Core modules

These are the building blocks which are used to construct the CVs.

NN

This module implements the architectures with learnable weights that can be used to build CV models.

Descriptors-based

FeedForward(layers[, activation, dropout, ...])

Define a feedforward neural network given the list of layers.

Graphs-based

Base class

BaseGNN(n_out, dataset_for_initialization[, ...])

Base class for Graph Neural Network (GNN) models

Architectures

SchNetModel(n_out[, ...])

The SchNet [1] model.

GVPModel(n_out[, ...])

The Geometric Vector Perceptron (GVP) model [1, 2] with vector gate [2].

Loss

This module implements the loss functions that can be used to optimize CV models.

MSELoss(*args, **kwargs)

(Weighted) Mean Square Error

ELBOGaussiansLoss(*args, **kwargs)

ELBO loss function assuming the latent and reconstruction distributions are Gaussian.

FisherDiscriminantLoss(n_states[, lda_mode, ...])

Fisher's discriminant ratio.

AutocorrelationLoss([reduce_mode, invert_sign])

(Weighted) autocorrelation loss.

ReduceEigenvaluesLoss([mode, n_eig, invert_sign])

Calculate a monotonic function f(x) of the eigenvalues, by default the sum.

TDALoss(n_states, target_centers, target_sigmas)

Compute a loss function as the distance from a simple Gaussian target distribution.

CommittorLoss(alpha, atomic_masses[, gamma, ...])

Compute a loss function based on Kolmogorov's variational principle (or its approximation) for the determination of the committor function

GeneratorLoss(r, eta, friction, alpha[, ...])

Computes the loss function to learn a representation for the resolvent of the infinitesimal generator

SmartDerivatives([setup_device, force_all_atoms])

Utils to compute efficently (time and memory wise) the derivatives of the model output wrt the positions used to compute the input descriptors.

Stats

This module implements statistical methods with learnable weights that can be used in CV models.

Base class

Stats(*args, **kwargs)

Base stats class.

Linear methods

PCA(in_features[, out_features])

Principal Component Analysis class.

LDA(in_features, n_states[, mode])

Fisher's discriminant class.

TICA(in_features[, out_features])

Time-lagged independent component analysis base class.

Transform

This module implements non-learnable pre/postprocessing tools

Base class

Transform(in_features, out_features)

Base transform class.

Descriptors

This submodule implements several descriptors that can be computed starting from atomic positions.

PairwiseDistances(n_atoms, PBC[, cell, ...])

Non duplicated pairwise distances for a set of atoms from their positions.

TorsionalAngles(indices, n_atoms, mode, PBC)

Torsional angle defined by a set of 4 atoms from their positions.

CoordinationNumbers(group_A, group_B, ...[, ...])

Coordination number between the elements of two groups of atoms from their positions

EigsAdjMat(mode, cutoff, n_atoms, PBC[, ...])

Eigenvalues of the adjacency matrix for a set of atoms from their positions

MultipleDescriptors(descriptors_list, n_atoms)

Wrapper class to combine multiple descriptor transform objects acting on the same set of atomic positions

Tools

This submodule implements pre/postporcessing tools.

Normalization(in_features[, mean, range, ...])

Normalizing block, used for computing standardized inputs/outputs.

ContinuousHistogram(in_features, min, max, bins)

Compute continuous histogram using Gaussian kernels

SwitchingFunctions(in_features, name, cutoff)

Common switching functions