Class for managing the training process by logging and storing the state of the current epoch.


onmt.EpochState(epoch, numIterations, learningRate, lastValidPpl, status)

Initialize for epoch epoch and training status (current loss)


onmt.EpochState:update(batch, loss)

Update training status. Takes batch (described in data.lua) and last loss.


onmt.EpochState:log(batchIndex, json)

Log to status stdout.

Undocumented methods

* onmt.EpochState:getTrainPpl() * onmt.EpochState:getTime() * onmt.EpochState:getStatus() * onmt.EpochState:getMinFreememory()