org.apache.sysml.runtime.io.ReaderBinaryBlock Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of systemml Show documentation
Show all versions of systemml Show documentation
Declarative Machine Learning
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.apache.sysml.runtime.io;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.SequenceFile;
import org.apache.hadoop.mapred.JobConf;
import org.apache.sysml.conf.ConfigurationManager;
import org.apache.sysml.runtime.DMLRuntimeException;
import org.apache.sysml.runtime.matrix.data.MatrixBlock;
import org.apache.sysml.runtime.matrix.data.MatrixIndexes;
import org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;
import org.apache.sysml.runtime.matrix.mapred.MRJobConfiguration;
public class ReaderBinaryBlock extends MatrixReader
{
protected boolean _localFS = false;
public ReaderBinaryBlock( boolean localFS )
{
_localFS = localFS;
}
public void setLocalFS(boolean flag) {
_localFS = flag;
}
@Override
public MatrixBlock readMatrixFromHDFS(String fname, long rlen, long clen, int brlen, int bclen, long estnnz)
throws IOException, DMLRuntimeException
{
//allocate output matrix block
MatrixBlock ret = createOutputMatrixBlock(rlen, clen, estnnz, false, false);
//prepare file access
JobConf job = new JobConf(ConfigurationManager.getCachedJobConf());
FileSystem fs = _localFS ? FileSystem.getLocal(job) : FileSystem.get(job);
Path path = new Path( (_localFS ? "file:///" : "") + fname);
//check existence and non-empty file
checkValidInputFile(fs, path);
//core read
readBinaryBlockMatrixFromHDFS(path, job, fs, ret, rlen, clen, brlen, bclen);
//finally check if change of sparse/dense block representation required
if( !AGGREGATE_BLOCK_NNZ )
ret.recomputeNonZeros();
ret.examSparsity();
return ret;
}
/**
*
* @param fname
* @param rlen
* @param clen
* @param brlen
* @param bclen
* @param estnnz
* @return
* @throws IOException
* @throws DMLRuntimeException
*/
public ArrayList readIndexedMatrixBlocksFromHDFS(String fname, long rlen, long clen, int brlen, int bclen)
throws IOException, DMLRuntimeException
{
//allocate output matrix block collection
ArrayList ret = new ArrayList();
//prepare file access
JobConf job = new JobConf(ConfigurationManager.getCachedJobConf());
FileSystem fs = _localFS ? FileSystem.getLocal(job) : FileSystem.get(job);
Path path = new Path( (_localFS ? "file:///" : "") + fname);
//check existence and non-empty file
checkValidInputFile(fs, path);
//core read
readBinaryBlockMatrixBlocksFromHDFS(path, job, fs, ret, rlen, clen, brlen, bclen);
return ret;
}
/**
* Note: For efficiency, we directly use SequenceFile.Reader instead of SequenceFileInputFormat-
* InputSplits-RecordReader (SequenceFileRecordReader). First, this has no drawbacks since the
* SequenceFileRecordReader internally uses SequenceFile.Reader as well. Second, it is
* advantageous if the actual sequence files are larger than the file splits created by
* informat.getSplits (which is usually aligned to the HDFS block size) because then there is
* overhead for finding the actual split between our 1k-1k blocks. This case happens
* if the read matrix was create by CP or when jobs directly write to large output files
* (e.g., parfor matrix partitioning).
*
* @param path
* @param job
* @param fs
* @param dest
* @param rlen
* @param clen
* @param brlen
* @param bclen
* @throws IOException
* @throws IllegalAccessException
* @throws InstantiationException
* @throws DMLRuntimeException
*/
@SuppressWarnings("deprecation")
private static void readBinaryBlockMatrixFromHDFS( Path path, JobConf job, FileSystem fs, MatrixBlock dest, long rlen, long clen, int brlen, int bclen )
throws IOException, DMLRuntimeException
{
boolean sparse = dest.isInSparseFormat();
MatrixIndexes key = new MatrixIndexes();
MatrixBlock value = new MatrixBlock();
long lnnz = 0; //aggregate block nnz
//set up preferred custom serialization framework for binary block format
if( MRJobConfiguration.USE_BINARYBLOCK_SERIALIZATION )
MRJobConfiguration.addBinaryBlockSerializationFramework( job );
for( Path lpath : getSequenceFilePaths(fs, path) ) //1..N files
{
//directly read from sequence files (individual partfiles)
SequenceFile.Reader reader = new SequenceFile.Reader(fs,lpath,job);
try
{
//note: next(key, value) does not yet exploit the given serialization classes, record reader does but is generally slower.
while( reader.next(key, value) )
{
//empty block filter (skip entire block)
if( value.isEmptyBlock(false) )
continue;
int row_offset = (int)(key.getRowIndex()-1)*brlen;
int col_offset = (int)(key.getColumnIndex()-1)*bclen;
int rows = value.getNumRows();
int cols = value.getNumColumns();
//bound check per block
if( row_offset + rows < 0 || row_offset + rows > rlen || col_offset + cols<0 || col_offset + cols > clen )
{
throw new IOException("Matrix block ["+(row_offset+1)+":"+(row_offset+rows)+","+(col_offset+1)+":"+(col_offset+cols)+"] " +
"out of overall matrix range [1:"+rlen+",1:"+clen+"].");
}
//copy block to result
if( sparse )
{
//note: append requires final sort (but prevents repeated shifting)
dest.appendToSparse(value, row_offset, col_offset);
}
else
{
dest.copy( row_offset, row_offset+rows-1,
col_offset, col_offset+cols-1,
value, false );
}
//maintain nnz as aggregate of block nnz
lnnz += value.getNonZeros();
}
}
finally
{
IOUtilFunctions.closeSilently(reader);
}
}
//post-processing
dest.setNonZeros( lnnz );
if( sparse && clen>bclen ){
//no need to sort if 1 column block since always sorted
dest.sortSparseRows();
}
}
/**
*
* @param path
* @param job
* @param fs
* @param dest
* @param rlen
* @param clen
* @param brlen
* @param bclen
* @throws IOException
* @throws IllegalAccessException
* @throws InstantiationException
*/
@SuppressWarnings("deprecation")
private void readBinaryBlockMatrixBlocksFromHDFS( Path path, JobConf job, FileSystem fs, Collection dest, long rlen, long clen, int brlen, int bclen )
throws IOException
{
MatrixIndexes key = new MatrixIndexes();
MatrixBlock value = new MatrixBlock();
//set up preferred custom serialization framework for binary block format
if( MRJobConfiguration.USE_BINARYBLOCK_SERIALIZATION )
MRJobConfiguration.addBinaryBlockSerializationFramework( job );
for( Path lpath : getSequenceFilePaths(fs, path) ) //1..N files
{
//directly read from sequence files (individual partfiles)
SequenceFile.Reader reader = new SequenceFile.Reader(fs,lpath,job);
try
{
while( reader.next(key, value) )
{
int row_offset = (int)(key.getRowIndex()-1)*brlen;
int col_offset = (int)(key.getColumnIndex()-1)*bclen;
int rows = value.getNumRows();
int cols = value.getNumColumns();
//bound check per block
if( row_offset + rows < 0 || row_offset + rows > rlen || col_offset + cols<0 || col_offset + cols > clen )
{
throw new IOException("Matrix block ["+(row_offset+1)+":"+(row_offset+rows)+","+(col_offset+1)+":"+(col_offset+cols)+"] " +
"out of overall matrix range [1:"+rlen+",1:"+clen+"].");
}
//copy block to result
dest.add(new IndexedMatrixValue(new MatrixIndexes(key), new MatrixBlock(value)));
}
}
finally
{
IOUtilFunctions.closeSilently(reader);
}
}
}
}