GTorlai / CDL_RBM_Tutorial

Tutorial on machine learning many-body physics with Boltzmann machines

Restricted Boltzmann Machine

This module describes how to employ a restricted Boltzmann machine (rbm) to learn different physical distributions from raw data, obtained by performing measurements on the systems. Contents of the folder:

• rbm.py: rbm class
• main.py: main script
• data/: folder containing the data of the 2d classical Ising model and the 1d transverse field quantum Ising model
• tutorial/: folder containing the scripts for the tutorial

Usage:

$ python main.py [COMMAND] [ARGUMENTS]

Command:

• train: train the rbm
• sample: sample the rbm given a set of trained parameters

Arguments:

• -nV: number of visible units
• -nH: number of hidden units
• -lr: learning rate
• -CD: number of Gibbs updates in the contrastive divergence algorithm
• -bs: batch size
• -step: number of training steps
• -nC: number of chains sampled in contrastive divergence

Useful references: Training of RBMs: https://www.cs.toronto.edu/~hinton/absps/guideTR.pdf