| # Copyright 2016 The Vanadium Authors. All rights reserved. |
| # Use of this source code is governed by a BSD-style |
| # license that can be found in the LICENSE file. |
| |
| """Runs the k-means clustering algorithms on the encoded images.""" |
| |
| import matplotlib.pyplot as plt |
| import numpy as np |
| import os |
| |
| from keras.models import model_from_json |
| from sklearn.cluster import KMeans |
| from theano import theano |
| |
| |
| def encode_images(images): |
| """Encodes images using model saved to disk earlier.""" |
| |
| json_file = open('model.json', 'r') |
| loaded_model_json = json_file.read() |
| json_file.close() |
| model = model_from_json(loaded_model_json) |
| model.load_weights('model.h5') |
| model.compile(optimizer='adadelta', loss='binary_crossentropy') |
| get_activations = theano.function([model.layers[0].input], |
| model.layers[3].output, |
| allow_input_downcast=True) |
| data = get_activations(images) |
| np.save('encoded_imgs.npy', data) |
| |
| if __name__ == '__main__': |
| data = np.load('encoded_imgs.npy') |
| print data.shape |
| |
| k_values = [2**exp for exp in range(5, 11)] |
| |
| if not os.path.exists('clusters'): |
| os.makedirs('clusters') |
| |
| for k in k_values: |
| if k > data.shape[0]: |
| break |
| |
| model = KMeans(init='k-means++', n_clusters=k, max_iter=10000, verbose=0, |
| tol=0.000000001, n_jobs=4) |
| model.fit(data) |
| |
| sse = 0 # Sum of squared errors. |
| centroids = model.cluster_centers_ |
| labels = model.labels_ |
| for idx in range(data.shape[0]): |
| dist = np.linalg.norm(data[idx] - centroids[labels[idx]]) |
| sse += dist**2 |
| mse = sse/data.shape[0] # Mean squared error. |
| mse_sqrt = np.sqrt(mse) |
| print str(k) + ': ' + str(model.inertia_) + ': ' + str(mse_sqrt) |
| |
| np.save('clusters/centroids_' + str(k) + '.npy', centroids) |
| np.save('clusters/labels_' + str(k) + '.npy', labels) |
| |
