How do I plot the sorted vs unsorted sequences as seen in the documentation?

Hi All,

I'm working with LSTM networks, and would like to illustrate the advantage of sorting the data with my own data sets as illustrated below, and mentioned here. However, although I know how to plot the sorted date, the blue part, I don't know how to add the padding part and the mini-batches and the splits, how do plot this?

Thanks in advance!

Answer 1

[Abhinav Aravindan]

Hi @Chao Yang,

To illustrate the advantage of sorting data with your own dataset, you may use the “barh” function to plot your sequences.

For instance, in the screenshot attached, the plot illustrates ordering data by sequence length and padding all the sequences in a mini-batch to the nearest multiple of a length that is greater than the longest sequence length in the mini-batch. You may refer to the below code snippet to generate a similar plot using “barh”.

% Load Data
[Xtrain,Ytrain] = japaneseVowelsTrainData;
seqLen = cellfun(@(X) size(X,2), Xtrain);

% Sorted Sequence
[sequenceLengthsOrd,idx] = sort(seqLen);
Xtrain = Xtrain(idx);

% Define parameters
batchSize = 27;
sequenceLength = 5;
numBatches = floor(length(Xtrain) / batchSize);

allOriginalLengths = [];
allPaddedLengths = [];
for i = 1:numBatches
    % Calculate indices and get mini-batch data
    startIdx = (i-1) * batchSize + 1;
    endIdx = i * batchSize;
    X_batch = Xtrain(startIdx:endIdx);
    
    % Determine the lengths of sequences in the mini-batch
    seqLen_Xbatch = cellfun(@(X) size(X,2), X_batch);
    maxSeqLength = max(seqLen_Xbatch);
    
    % Determine padding length 
    paddingLength = ceil(maxSeqLength / sequenceLength) * sequenceLength;
   
    allOriginalLengths = [allOriginalLengths; seqLen_Xbatch];
    allPaddedLengths = [allPaddedLengths; repmat(paddingLength, batchSize, 1)];
end

% Plot 
figure;
barh(allOriginalLengths,'BarWidth', 1); % Original lengths
hold on;
barh(allPaddedLengths, 'FaceColor', [0.7, 0.7, 0.9], 'FaceAlpha', 0.5,'BarWidth', 1); % Padded lengths

% Horizontal Lines
yValues = [0, 27, 27*2, 27*3, 27*4, 27*5, 27*6, 27*7, 27*8, 27*9, 27*10];
xRanges = [15, 15, 15, 15, 15, 20, 20, 20, 25, 30, 30];

for i = 1:length(yValues)
    plot([0, xRanges(i)], [yValues(i), yValues(i)], '-r', 'LineWidth', 1);
end

% Vertical Lines
lineData = {
    [15, 15], [0, 27*5];  
    [20, 20], [27*5, 27*8];
    [25, 25], [27*8, 27*9];
    [30, 30], [27*9, 27*10];
};

for i = 1:size(lineData, 1)
    plot(lineData{i, 1}, lineData{i, 2}, '-r', 'LineWidth', 1);
end

xline(5, '--r', 'LineWidth',1);
xline(10, '--r', 'LineWidth',1);
xline(15, '--r', 'LineWidth',1);
plot([20 20], [27*8 27*10], '--r', 'LineWidth', 1);
plot([250 25], [27*9 27*10], '--r', 'LineWidth', 1);

hold off;

title('Sorted Data');
xlabel('Length');
ylabel('Sequence');
xlim([0, max(allPaddedLengths)]);

Output:

Please refer to the below documentation for more details on the usage of "barh"

https://www.mathworks.com/help/matlab/ref/barh.html

I hope this helps!