Package org.apache.sysds.runtime.controlprogram.paramserv.dp
package org.apache.sysds.runtime.controlprogram.paramserv.dp
-
ClassesClassDescriptionBalance to Avg Federated scheme When the parameter server runs in federated mode it cannot pull in the data which is already on the workers.Disjoint_Contiguous data partitioner: for each worker, use a right indexing operation X[beg:end,] to obtain contiguous, non-overlapping partitions of rows.Spark Disjoint_Contiguous data partitioner:Data partitioner Disjoint_Random: for each worker, use a permutation multiply P[beg:end,] %*% X, where P is constructed for example with P=table(seq(1,nrow(X)),sample(nrow(X), nrow(X))), i.e., sampling without replacement to ensure disjointness.Disjoint_Round_Robin data partitioner: for each worker, use a permutation multiply or simpler a removeEmpty such as removeEmpty (target=X, margin=rows, select=(seq(1,nrow(X))%%k)==id)Spark Disjoint_Round_Robin data partitioner:Spark data partitioner Disjoint_Random: For the current row block, find all the shifted place for each row (WorkerID => (row block ID, matrix)Keep Data on Worker Federated scheme When the parameter server runs in federated mode it cannot pull in the data which is already on the workers.Data partitioner Overlap_Reshuffle: for each worker, use a new permutation multiply P %*% X, where P is constructed for example with P=table(seq(1,nrow(X),sample(nrow(X), nrow(X))))Spark data partitioner Overlap_Reshuffle:Replicate to Max Federated scheme When the parameter server runs in federated mode it cannot pull in the data which is already on the workers.Shuffle Federated scheme When the parameter server runs in federated mode it cannot pull in the data which is already on the workers.Subsample to Min Federated scheme When the parameter server runs in federated mode it cannot pull in the data which is already on the workers.