Collective variablesΒΆ

In this section we report the neural network-based collective variables implemented in the library. Note that the linear statistical methods are implemented in mlcolvar.core.stats instead.

Base class

All CVs inherits from this base class, which also implement default methods.

BaseCV(in_features, out_features[, ...])

Base collective variable class.

For each of the specific CV described below there are reported the keys of the expected dataset and the loss function used.

Unsupervised learning

CVs based on the autoencoder architecture. Can be used to reconstruct the original input or an arbitrary reference, with an optional reweighting of the data.

AutoEncoderCV(encoder_layers[, ...])

AutoEncoding Collective Variable.

VariationalAutoEncoderCV(n_cvs, encoder_layers)

Variational AutoEncoder Collective Variable.

Supervised learning

CVs optimized with supervised learning tasks, either classification or regression.

DeepLDA(layers, n_states[, options])

Deep Linear Discriminant Analysis (Deep-LDA) CV.

DeepTDA(n_states, n_cvs, target_centers, ...)

Deep Targeted Discriminant Analysis (Deep-TDA) CV.

RegressionCV(layers[, options])

Example of collective variable obtained with a regression task.

Time-informed learning

CVs which are optimized on pairs of time-lagged configurations, and optional reweighting for the time-correlation functions. Note that also the autoencoder-related CVs can fall in this category when the target reference is the time-lagged data.

DeepTICA(layers[, n_cvs, options])

Neural network-based time-lagged independent component analysis (Deep-TICA).

MultiTask learning

General framework which allows to optimize a single model with different loss functions optimized on different datasets.

MultiTaskCV(main_cv, auxiliary_loss_fns[, ...])

Multi-task collective variable.

Framework for the numerical determination of the committor function based on its variational principle.

Committor

Committor(layers, atomic_masses, alpha[, ...])

Base class for data-driven learning of committor function.