time-lagged autoencoder

A toolbox for dimension reduction of time series data with a time-lagged autoencoder-type deep neural network.

Installation

Make sure to install pytorch via conda, instructions on http://pytorch.org, before you install the present module with

python setup.py test
python setup.py install

To run the included benchmarks, you also need to install the packages pyemma and mdshare.

Methods

This package implements

• principal component analysis (PCA),
• time-lagged independent component analysis (TICA),
• time-lagged canonical correlation analysis (via TICA),
• kinetic maps (via TICA), and
• an autoencoder-type neural network (AE) trained in a time-lagged manner.

Example

Assume that data is a single numpy.ndarray(shape=[n_frames, n_features]) object or list thereof, n_frames refers to the number of timesteps in the/each trajectory, and n_features refers to the number of features extracted from the original molecular dyamics (MD) data. Now choose a target dimensionality dim and a transformation lag time lag, and run:

import tae

# run PCA
pca_transformed_data, pca_train_loss, pca_val_loss = tae.pca(data, dim=dim)

# run TICA
tica_transformed_data, tica_train_loss, tica_val_loss = tae.tica(data, dim=dim, lag=lag)

# run AE
ae_transformed_data, ae_train_loss, ae_val_loss = tae.ae(data, dim=dim, lag=lag)

# run VAE
vae_transformed_data, vae_train_loss, vae_val_loss = tae.vae(data, dim=dim, lag=lag)

# run AE on a GPU
ae_transformed_data, ae_train_loss, ae_val_loss = tae.ae(data, dim=dim, lag=lag, cuda=True)

In this example, we get *_val_loss=None because we are training on the full data set. To exclude a randomly chosen fraction fval of the data from the training, add the parameter validation_split=fval to the function calls, e.g.:

ae_transformed_data, ae_train_loss, ae_val_loss = tae.ae(
    data, dim=dim, lag=lag, validation_split=fval, cuda=True)

Citation

@article{time-lagged-autoencoder,
	Author = {Christoph Wehmeyer and Frank No{\'{e}}},
	Doi = {10.1063/1.5011399},
	Journal = {J. Chem. Phys.},
	Month = {jun},
	Number = {24},
	Pages = {241703},
	Publisher = {{AIP} Publishing},
	Title = {Time-lagged autoencoders: Deep learning of slow collective variables for molecular kinetics},
	Volume = {148},
	Year = {2018}}

Development system

This project was developed using the following python environment:

package	version		channel
python	3.6.1	2
conda	4.3.29	py36_0	conda-forge
numpy	1.13.3	py36_blas_openblas_200 [blas_openblas]	conda-forge
pytorch	0.2.0	py36_4cu75	soumith
pyemma	2.4	np113py36_1	conda-forge