ds.monitored_fitrbm: Fit an RBM model
In stefan-m-lenz/dsBoltzmannMachinesClient: Client for dsBoltzmannMachines

Fits an RBM model using Stochastic Gradient Descent (SGD) on the data with Contrastive Divergence (CD). 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).

ds.monitored_fitrbm(
  datasources,
  data = "D",
  newobj = "rbm",
  monitoring = "reconstructionerror",
  monitoringdata = NULL,
  nhidden = NULL,
  epochs = NULL,
  upfactor = NULL,
  downfactor = NULL,
  learningrate = NULL,
  learningrates = NULL,
  pcd = NULL,
  cdsteps = NULL,
  categories = NULL,
  batchsize = NULL,
  rbmtype = NULL,
  startrbm = NULL
)

`datasources`	A list of Opal object(s) as a handle to the server-side session
`data`	The name of the variable that holds the training data on the server-side. Defaults to `"D"`.
`newobj`	The name for the variable in which the trained RBM will be stored. Defaults to `"rbm"`.
`monitoring`	Name(s) for monitoring options used for RBM training. Possible options: `"reconstructionerror"`: Calculates the reconstruction error (Default) `"loglikelihood"`: Estimates the loglikelihood via annealed importance sampling (AIS) `"exactloglikelihood"`: Exact calculation of log-likelihood. This is only feasible for very small models. `NULL`: No monitoring
`monitoringdata`	A vector of names for server-side data sets that are to be used for monitoring
`nhidden`	The number of hidden units of the returned RBM
`epochs`	The number of training epochs
`upfactor`	If this function is used for pretraining a part of a DBM, it is necessary to multiply the input from the visible layer of the RBM with a factor.
`downfactor`	If this function is used for pretraining a part of a DBM, it is necessary to multiply the input from the hidden layer of the RBM with a factor.
`learningrate`	The learning rate for the weights and biases can be specified as a single value, used throughout all epochs. Defaults to 0.005.
`learningrates`	The learning rate for the weights and biases can also be specified as a vector that contains a value for each epoch.
`pcd`	Indicating whether Persistent Contrastive Divergence (PCD) is to be used (`TRUE`, default) or simple CD that initializes the Gibbs chain with the training sample (`FALSE`)
`cdsteps`	The number of Gibbs sampling steps for (persistent) contrastive divergence. Defaults to 1.
`categories`	Only relevant if `rbmtype` is `"Softmax0BernoulliRBM"`. The number of categories, if all variables have the same number of categories, or a vector that contains the number of categories for the i'th categorical variable in the i'th entry.
`batchsize`	The number of samples that are used for making one step in the stochastic gradient descent optimizer algorithm. Default is 1.
`rbmtype`	The type of the RBM that is to be trained. This must be a subtype of `AbstractRBM`. Defaults to `BernoulliRBM`.
`startrbm`	A name for an RBM object at the server side that is used as starting value for training. If this argument is specified, `nhidden` and `rbmtype` are ignored.

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

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