Network to be trained
Weight update algorithm to be applied
Input Operation that supervises how to manipulate input as matrices. This also controls how to compute actual network. (default: VectorType)
Training criteria (default: SimpleTrainingCriteria)
Training Pair Type
Training Pair Type
Sampler Type
Sampler Type
Implicit weight operation
Implicit weight operation
Accumulator variable for counter
Accumulator variable for counter
Accumulator variable for networks
Accumulator variable for networks
Weight update algorithm to be applied
Weight update algorithm to be applied
Do mini-batch
Do mini-batch
Spark distributed networks
Spark distributed networks
Fetch weights
Iterate over given number of test instances
Iterate over given number of test instances
number of random sampled instances
iteratee function
Indicates whether the asynchronous update is finished or not.
Indicates whether the asynchronous update is finished or not.
future object of update
Logger
Logger
Input Operation that supervises how to manipulate input as matrices.
Input Operation that supervises how to manipulate input as matrices. This also controls how to compute actual network. (default: VectorType)
Fraction of negative samples
Fraction of negative samples
Negative Sampler
Negative Sampler
Partitioner for negative samples
Partitioner for negative samples
Network to be trained
Network to be trained
Training criteria (default: SimpleTrainingCriteria)
Training criteria (default: SimpleTrainingCriteria)
Set negative sampling method.
Set negative sampling method.
all training outputs that will be used for negative training
Set negative sampling method.
Set negative sampling method.
all training outputs that will be used for negative training
Set training instances
Set training instances
Set training instances
Sequence of training set
Set testing instances
Set testing instances
Set testing instances
Sequence of testing set
Non-blocking pending, until all assigned batches are finished
Non-blocking pending, until all assigned batches are finished
Test Set
Test Set
Size of test set
Size of test set
Fraction of mini-batch
Fraction of mini-batch
Training set
Training set
Unpersist all
Send update of weights
Calculate validation error
Calculate validation error
validation error
Trainer : Stochastic-Style, Multi-Threaded using Spark.
This is not a implementation using DistBelief Paper. This is between DistBeliefTrainStyle(DBTS) and SingleThreadTrainStyle(STTS). The major difference is whether "updating" is asynchronous(DBTS) or not(MTTS).