Training style that supervises how to train. There are two styles, one is SingleThreadTrainStyle and the other is DistBeliefTrainStyle.
Stopping Criteria that controls the threshold for stopping. (Default : StoppingCriteria)
Name used for logging.
Best Loss Iteration Number
Best Loss Iteration Number
Best Parameter History
Best Parameter History
Logger
Logger
Name used for logging.
Print validation result into logger
Print validation result into logger
Restore best parameters
Restore best parameters
Store best parameters
Store best parameters
current iteration epoch. (1 iteration = 1 validation freq)
previous loss
current patience, i.e. loop until at least this epoch.
Stopping Criteria that controls the threshold for stopping.
Stopping Criteria that controls the threshold for stopping. (Default : StoppingCriteria)
Training style that supervises how to train.
Training style that supervises how to train. There are two styles, one is SingleThreadTrainStyle and the other is DistBeliefTrainStyle.
Train using given RDD sequence.
Train using given RDD sequence.
RDD of training set
RDD of validation set
Train using given RDD sequence.
Train using given RDD sequence.
RDD of training set
Train given sequence, and validate with another sequence.
Train given sequence, and validate with another sequence.
Full Sequence of training set
Full Sequence of validation set
Training error (loss)
Train given sequence, and validate with given sequence.
Train given sequence, and validate with given sequence.
Full Sequence of training set
Training error (loss)
Tail Recursive : Train each batch
Tail Recursive : Train each batch
current iteration epoch. (1 iteration = 1 validation freq)
previous loss
current patience, i.e. loop until at least this epoch.
Total Loss when train is finished
Period of validation
Period of validation
General Trainer Implementation.
This class trains with help of Training Style and Input Operation.
the type of input. Currently, kr.ac.kaist.ir.deep.fn.ScalarMatrix and DAG are supported
the type of output Currently, kr.ac.kaist.ir.deep.fn.ScalarMatrix and Null are supported
To train an autoencoder, you can provide same training set as validation set.
,This trainer is generalized class. Further implementation, you should see several styles.