utils

Type Members

1. case class DistriOptimizerPerfParam(batchSize: Int = 128, maxEpoch: Int = 5, dataType: String = "float", module: String = "inception_v1", inputData: String = "random") extends Product with Serializable

   

   The parameters of a distributed optimizer

   The parameters of a distributed optimizer

   batchSize

   batch size

   maxEpoch

   how many epochs to run

   dataType

   data type (double / float)

   module

   module name

   inputData

   inputData input data type (constant / random)

2. case class LocalOptimizerPerfParam(batchSize: Int = 128, coreNumber: Int = ..., iteration: Int = 50, dataType: String = "float", module: String = "inception_v1", inputData: String = "random") extends Product with Serializable

   

   Local Optimizer Performance Parameters

   Local Optimizer Performance Parameters

   batchSize

   batch size

   coreNumber

   core number

   iteration

   how many iterations to run

   dataType

   data type (double / float)

   module

   module name

   inputData

   input data type (constant / random)

3. class ModelBroadcast[T] extends Serializable

   

   ModelBroadcast is used to broadcast model when doing model inference.

   ModelBroadcast is used to broadcast model when doing model inference. Note: do not use it in model training since the broadcast models share weights and biases It shortens the broadcast time, which is especially useful when the model size is large

   T

   data type