mxLSTMtrain: mxLSTMtrain #' @param model mxSymbol as returned by...

In MarkusBonsch/mxLSTM: LSTM model for regression

Description

train the LSTM

Usage

mxLSTMtrain(datTrain, datEval, symbol, batchSize, num.hidden,
  optimizeFullSequence, num.rounds, optimizer = "rmsprop",
  learning.rate = data.table(epoch = 1L, lr = 0.002),
  initializer = mx.init.Xavier(), initialModel = NULL, shuffle = TRUE,
  gammaInit, epoch.end.callback = NULL, ...)

Arguments

datTrain	list with entries 'data' and 'label'. 'data' is a 3D array of dimension num.features:seq.length:number of observations 'label is a 3D array of dimension num.outputs:seq.length:number of observations
datEval	similar to datTrain, but for evaluation instead of training.
symbol	mxSymbol as returned by mxLSTMcreate
batchSize	see mxLSTM
num.hidden	see mxLSTM
optimizeFullSequence	see mxLSTM
num.rounds	see num.epoch argument in mxLSTM
optimizer	see mxLSTM
learning.rate	data.table with two columns: epoch (integer) tells from which epoch onwards this learningrate is active. Has to start with 1 lr (numeric) the actual value for the learningrate epoch
initializer	see mxLSTM
initialModel	see mxLSTM
shuffle	see mxLSTM
gammaInit	see mxLSTM
epoch.end.callback	function to be called at the end of each epoch.
...	further arguments passed to optimizer

Value

object of class 'mxLSTM' list: 'model' is the actual symbol, 'arg.params' and 'aux.params' are the parameters, 'log' is the training log, 'optimizerEnv' is an optional environment with optimizer parameters.

MarkusBonsch/mxLSTM documentation built on May 28, 2019, 6:40 a.m.