Visualize Activations of LSTM Network

This example shows how to investigate and visualize the features learned by LSTM networks by extracting the activations.

Load pretrained network.JapaneseVowelsNetis a pretrained LSTM network trained on the Japanese Vowels dataset as described in [1] and [2]. It was trained on the sequences sorted by sequence length with a mini-batch size of 27.

loadJapaneseVowelsNet

View the network architecture.

net.Layers

ans = 5x1 Layer array with layers: 1 'sequenceinput' Sequence Input Sequence input with 12 dimensions 2 'lstm' LSTM LSTM with 100 hidden units 3 'fc' Fully Connected 9 fully connected layer 4 'softmax' Softmax softmax 5 'classoutput' Classification Output crossentropyex with '1' and 8 other classes

Load the test data.

[XTest,YTest] = japaneseVowelsTestData;

Visualize the first time series in a plot. Each line corresponds to a feature.

X = XTest{1}; figure plot(XTest{1}') xlabel("Time Step") title("Test Observation 1") numFeatures = size(XTest{1},1); legend("Feature "+ string(1:numFeatures),'Location','northeastoutside')

For each time step of the sequences, get the activations output by the LSTM layer (layer 2) for that time step and update the network state.

sequenceLength = size(X,2); idxLayer = 2; outputSize = net.Layers(idxLayer).NumHiddenUnits;fori = 1:sequenceLength features(:,i) = activations(net,X(:,i),idxLayer); [net, YPred(i)] = classifyAndUpdateState(net,X(:,i));end

Visualize the first 10 hidden units using a heatmap.

figure heatmap(features(1:10,:)); xlabel("Time Step") ylabel("Hidden Unit") title("LSTM Activations")

The heatmap shows how strongly each hidden unit activates and highlights how the activations change over time.

See Also

trainNetwork|trainingOptions|lstmLayer|bilstmLayer|sequenceInputLayer|activations

Related Topics

• Time Series Forecasting Using Deep Learning
• Sequence-to-Sequence Classification Using Deep Learning
• Sequence-to-Sequence Regression Using Deep Learning
• Sequence-to-One Regression Using Deep Learning
• Long Short-Term Memory Networks
• Deep Learning in MATLAB