Python Image Processing Cookbook

Online dictionary learning

Here are a few steps for online learning and reconstruction:

1. Define the constants:

batch_size = 128 n_epochs = 2 n_batches = len(data) // batch_size

2. Instantiate the MiniBatchDictionaryLearning object, this time with an additional argument, batch_size, to the constructor:

dico = MiniBatchDictionaryLearning(n_components=256, alpha=1, \                n_iter=1, batch_size=batch_size)# Now extract noisy patches as before

3. Iteratively extract the current batch of patches and use partial_fit() to fit/update the dictionary from the current batch:

for epoch in range(n_epochs): for i in range(n_batches): batch = data[i * batch_size: (i + 1) * batch_size] dico.partial_fit(batch) V = dico.components_ n_updates += 1 # Now reconstruct ...