matlab.io.datastore.Partitionable - Add parallelization support to datastore - MATLAB (original) (raw)

Namespace: matlab.io.datastore

Add parallelization support to datastore

Description

matlab.io.datastore.Partitionable is an abstract mixin class that adds parallelization support to your custom datastore for use with Parallel Computing Toolbox™ and MATLAB® Parallel Server™.

To use this mixin class, you must inherit frommatlab.io.datastore.Partitionable class, in addition to inheriting from the matlab.io.Datastore base class. Type the following syntax as the first line of your class definition file:

classdef MyDatastore < matlab.io.Datastore & ... matlab.io.datastore.Partitionable ... end

To add support for parallel processing to your custom datastore, you must:

Inherit from the additional class matlab.io.datastore.Partitionable.
Define these additional methods: maxpartitions and partition.

For more details and steps to create your custom datastore with parallel processing support, see Develop Custom Datastore.

Methods

Examples

collapse all

Build Datastore with Parallel Processing Support

Build a datastore with parallel processing support and use it to bring your custom or proprietary data into MATLAB®. Then, process the data in a parallel pool.

Create a .m class definition file that contains the code implementing your custom datastore. You must save this file in your working folder or in a folder that is on the MATLAB® path. The name of the .m file must be the same as the name of your object constructor function. For example, if you want your constructor function to have the name MyDatastorePar, then the name of the .m file must be MyDatastorePar.m. The .m class definition file must contain the following steps:

Step 1: Inherit from the datastore classes.
Step 2: Define the constructor and the required methods.
Step 3: Define your custom file reading function.

In addition to these steps, define any other properties or methods that you need to process and analyze your data.

%% STEP 1: INHERIT FROM DATASTORE CLASSES classdef MyDatastorePar < matlab.io.Datastore & ... matlab.io.datastore.Partitionable

properties(Access = private)
    CurrentFileIndex double
    FileSet matlab.io.datastore.DsFileSet
end

% Property to support saving, loading, and processing of
% datastore on different file system machines or clusters.
% In addition, define the methods get.AlternateFileSystemRoots()
% and set.AlternateFileSystemRoots() in the methods section. 
properties(Dependent)
    AlternateFileSystemRoots
end

%% STEP 2: DEFINE THE CONSTRUCTOR AND THE REQUIRED METHODS methods % Define your datastore constructor function myds = MyDatastorePar(location,altRoots) myds.FileSet = matlab.io.datastore.DsFileSet(location,... 'FileExtensions','.bin', ... 'FileSplitSize',8*1024); myds.CurrentFileIndex = 1;

        if nargin == 2
             myds.AlternateFileSystemRoots = altRoots;
        end
        
        reset(myds);
    end
    
    % Define the hasdata method
    function tf = hasdata(myds)
        % Return true if more data is available
        tf = hasfile(myds.FileSet);
    end
    
    % Define the read method
    function [data,info] = read(myds)
        % Read data and information about the extracted data
        % See also: MyFileReader()
        if ~hasdata(myds)
            msgII = ['Use the reset method to reset the datastore ',... 
                     'to the start of the data.']; 
            msgIII = ['Before calling the read method, ',...
                      'check if data is available to read ',...
                      'by using the hasdata method.'];
            error('No more data to read.\n%s\n%s',msgII,msgIII);
        end
        
        fileInfoTbl = nextfile(myds.FileSet);
        data = MyFileReader(fileInfoTbl);
        info.Size = size(data);
        info.FileName = fileInfoTbl.FileName;
        info.Offset = fileInfoTbl.Offset;
        
        % Update CurrentFileIndex for tracking progress
        if fileInfoTbl.Offset + fileInfoTbl.SplitSize >= ...
                fileInfoTbl.FileSize
            myds.CurrentFileIndex = myds.CurrentFileIndex + 1 ;
        end
    end
    
    % Define the reset method
    function reset(myds)
        % Reset to the start of the data
        reset(myds.FileSet);
        myds.CurrentFileIndex = 1;
    end

    % Define the partition method
    function subds = partition(myds,n,ii)
        subds = copy(myds);
        subds.FileSet = partition(myds.FileSet,n,ii);
        reset(subds);
    end
    
    % Getter for AlternateFileSystemRoots property
    function altRoots = get.AlternateFileSystemRoots(myds)
        altRoots = myds.FileSet.AlternateFileSystemRoots;
    end

    % Setter for AlternateFileSystemRoots property
    function set.AlternateFileSystemRoots(myds,altRoots)
        try
          % The DsFileSet object manages AlternateFileSystemRoots
          % for your datastore
          myds.FileSet.AlternateFileSystemRoots = altRoots;

          % Reset the datastore
          reset(myds);  
        catch ME
          throw(ME);
        end
    end
  
end

methods (Hidden = true)          
    % Define the progress method
    function frac = progress(myds)
        % Determine percentage of data read from datastore
        if hasdata(myds) 
           frac = (myds.CurrentFileIndex-1)/...
                         myds.FileSet.NumFiles; 
        else 
           frac = 1;  
        end 
    end
end

methods(Access = protected)
    % If you use the  FileSet property in the datastore,
    % then you must define the copyElement method. The
    % copyElement method allows methods such as readall
    % and preview to remain stateless 
    function dscopy = copyElement(ds)
        dscopy = copyElement@matlab.mixin.Copyable(ds);
        dscopy.FileSet = copy(ds.FileSet);
    end
    
    % Define the maxpartitions method
    function n = maxpartitions(myds)
        n = maxpartitions(myds.FileSet);
    end
end

end

%% STEP 3: IMPLEMENT YOUR CUSTOM FILE READING FUNCTION function data = MyFileReader(fileInfoTbl) % create a reader object using FileName reader = matlab.io.datastore.DsFileReader(fileInfoTbl.FileName);

% seek to the offset seek(reader,fileInfoTbl.Offset,'Origin','start-of-file');

% read fileInfoTbl.SplitSize amount of data data = read(reader,fileInfoTbl.SplitSize);

end

Your custom datastore is now ready. Use your custom datastore to read and process the data in a parallel pool.

Read Data Using Custom Datastore And Process in Parallel Pool

Use custom datastore to preview and read your proprietary data into MATLAB for parallel processing.

This example uses a simple data set to illustrate a workflow using your custom datastore. The data set is a collection of 15 binary (.bin) files where each file contains a column (1 variable) and 10000 rows (records) of unsigned integers.

binary_data01.bin binary_data02.bin binary_data03.bin binary_data04.bin binary_data05.bin binary_data06.bin binary_data07.bin binary_data08.bin binary_data09.bin binary_data10.bin binary_data11.bin binary_data12.bin binary_data13.bin binary_data14.bin binary_data15.bin

Create a datastore object using the MyDatastorePar function. For implementation details of MyDatastorePar, see the example Build Datastore with Parallel Processing Support.

folder = fullfile('*.bin'); ds = MyDatastorePar(folder);

Preview the data from the datastore.

ans = 8x1 uint8 column vector

113 180 251 91 29 66 254 214

Identify the number of partitions for your datastore. If you have Parallel Computing Toolbox (PCT), then you can use n = numpartitions(ds,myPool), where myPool is gcp or parpool.

Partition the datastore into n parts and n workers in a parallel pool.

parfor ii = 1:n subds = partition(ds,n,ii); while hasdata(subds) data = read(subds); % do something end end

Process Datastore on Different Platforms

To process your datastore with parallel and distributed computing that involves different platform cloud or cluster machines, you must pre-define 'AlternateFileSystemRoots' parameter. For instance, create a datastore on your local machine, and analyze a small portion of the data. Then, scale up your analysis to the entire dataset using Parallel Computing Toolbox and MATLAB Parallel Server.

Create a datastore using MyDatastorePar and assign a value to the 'AlternateFileSystemRoots' property. For implementation details of MyDatastorePar, see the exampleBuild Datastore with Parallel Processing Support.

To set the value for the 'AlternateFileSystemRoots' property, identify the root paths for your data on the different platforms. The root paths differ based on the machine or file system. For instance, if you access your data using these root paths:

"Z:\DataSet" from the Windows® machine.
"/nfs-bldg001/DataSet" from the MATLAB Parallel Server Linux® cluster.

Then, associate these root paths using theAlternateFileSystemRoots property.

altRoots = ["Z:\DataSet","/nfs-bldg001/DataSet"]; ds = MyDatastorePar('Z:\DataSet',altRoots);

Analyze a small portion of the data on your local machine. For instance, get a partitioned subset of the data and clean the data by removing any missing entries. Then, examine a plot of the variables.

tt = tall(partition(ds,100,1)); summary(tt); % analyze your data
tt = rmmissing(tt);
plot(tt.MyVar1,tt.MyVar2)

Scale up your analysis to the entire dataset by using MATLAB Parallel Server cluster (Linux cluster). For instance, start a worker pool using the cluster profile, and then perform analysis on the entire dataset by using parallel and distributed computing capabilities.

parpool('MyMjsProfile') tt = tall(ds);
summary(tt); % analyze your data tt = rmmissing(tt);
plot(tt.MyVar1,tt.MyVar2)

Tips

For your custom datastore implementation, best practice is not to implement the numpartitions method.

Version History

Introduced in R2017b