ds.monitored_stackrbms: Train a stack of RBMs

In stefan-m-lenz/dsBoltzmannMachinesClient: Client for dsBoltzmannMachines

Description

Performs greedy layerwise training for deep belief networks or greedy layerwise pretraining for deep Boltzmann machines. During the training, monitoring data is collected by default. The monitoring data is returned to the user. The trained model is stored on the server side (see parameter newobj).

Usage

ds.monitored_stackrbms(
  datasources,
  data = "D",
  newobj = "rbmstack",
  monitoring = "reconstructionerror",
  monitoringdata = NULL,
  nhiddens = NULL,
  epochs = NULL,
  predbm = NULL,
  samplehidden = NULL,
  learningrate = NULL,
  batchsize = NULL,
  trainlayers = NULL
)

Arguments

datasources	A list of Opal object(s) as a handle to the server-side session
data	The name of the variable that holds the data on the server-side. Defaults to "D".
newobj	The name of the variable in which the trained RBM will be stored. Defaults to "rbmstack".
monitoring	Name(s) for monitoring options used for RBM training. For possible options, see ds.monitored_fitrbm
monitoringdata	A vector of names for server-side data sets that are to be used for monitoring. The data is propagated forward through the network to monitor higher levels.
nhiddens	A vector containing the number of nodes of the i'th hidden layer in the i'th entry
epochs	The number of training epochs
predbm	logical value indicating that the greedy layerwise training is pre-training for a DBM. If its value is FALSE (default), a DBN is trained.
samplehidden	logical value indicating that consequent layers are to be trained with sampled values instead of the deterministic potential. Using the deterministic potential (FALSE) is the default.
learningrate	The learningrate used for training the RBMs. Defaults to 0.005.
batchsize	The size of the training minibatches. Defaults to 1.
trainlayers	A vector of names of TrainLayer objects. With this argument it is possible to specify the training parameters for each layer/RBM individually. If the number of training epochs and the learning rate are not specified explicitly for a layer, the values of epochs, learningrate and batchsize are used. For more information see help of ds.bm.defineLayer.

Details

If the option dsBoltzmannMachines.shareModels is set to TRUE by an administrator at the server side, the model itself is returned in addition.

stefan-m-lenz/dsBoltzmannMachinesClient documentation built on May 2, 2021, 10:46 a.m.