Python Image Processing Cookbook

Clustering handwritten digit images with SOM

Go through the following steps to cluster the digits images from the scikit-image dataset:

1. Load the digits and scale:

digits = datasets.load_digits(n_class=10)data = digits.data # matrix where each row is a vector that represent a digit.data = scale(data)num = digits.target # num[i] is the digit represented by data[i]

2. Create a SOM grid with 900 (30 x 30) neurons. Initialize the neurons with PCA weights:

som = MiniSom(30, 30, 64, sigma=4, learning_rate=0.5, neighborhood_function='triangle')som.pca_weights_init(data)

3. Train the SOM by choosing 5,000 randomly chosen digits using the following line of code:

som.train_random(data, 5000) # random training

3. Finally, plot the distance map of the weights ...