Skip to main content

Table 2 Overview of client-side functions for training and using DBM models

From: Deep generative models in DataSHIELD

Function name Short description
ds.monitored_fitrbm Monitored training of an RBM model
ds.monitored_stackrbms Monitored training of a stack of RBMs. Can be used for pre-training a DBM or for training a DBN
ds.monitored_fitdbm Monitored training of a DBM, including pre-training and fine-tuning
ds.setJuliaSeed Set a seed for the random number generator
ds.dbm.samples/
ds.rbm.samples
Generate samples from a DBM/RBM.
This also allows conditional sampling.
ds.bm.defineLayer Define training parameters individually for a RBM layer in a DBM or DBN
ds.bm.definePartitionedLayer Define a partitioned layer using other layers as parts
ds.dbm.top2LatentDims Get a two-dimensional representation of latent features
ds.rbm.loglikelihood Estimates the partition function of an RBM with AIS and then calculates the log-likelihood
ds.dbm.loglikelihood Performs a separate AIS run for each of the samples to estimate the log-likelihood of a DBM
ds.dbm.logproblowerbound Estimates the variational lower bound of the likelihood of a DBM with AIS
ds.rbm.exactloglikelihood/
ds.dbm.exactloglikelihood
Calculates the log-likelihood for a RBM/DBM (exponential complexity)