args = parse_command_line_arguments()
datadir = args.datadir
k = args.k
idx = args.index
if idx == None:
train = MNISTReader("%s/train-images-idx3-ubyte" % datadir, "%s/train-labels-idx1-ubyte" % datadir, preload=True)
test = MNISTReader("%s/t10k-images-idx3-ubyte" % datadir, "%s/t10k-labels-idx1-ubyte" % datadir, preload=True)
else:
train = MNISTReader("%s/train-images-idx3-ubyte" % datadir, "%s/train-labels-idx1-ubyte" % datadir)
test = MNISTReader("%s/t10k-images-idx3-ubyte" % datadir, "%s/t10k-labels-idx1-ubyte" % datadir)
if idx == None:
i = 0
for p, known_label in test.images(as_array=True):
label, nearest = knn(p, train.images(as_array=True), k)
if label == known_label:
x=""
else:
x="XXX"
print "%i %i %i %s" % (i, known_label, label, x)
sys.stdout.flush()
i+=1
else:
img = test.get_image(idx)
label, nearest = knn(img.as_array(), train.images(as_array=True), k)
if idx == None:
train = MNISTReader("%s/train-images-idx3-ubyte" % datadir,
"%s/train-labels-idx1-ubyte" % datadir,
preload=True)
test = MNISTReader("%s/t10k-images-idx3-ubyte" % datadir,
"%s/t10k-labels-idx1-ubyte" % datadir,
preload=True)
else:
train = MNISTReader("%s/train-images-idx3-ubyte" % datadir,
"%s/train-labels-idx1-ubyte" % datadir)
test = MNISTReader("%s/t10k-images-idx3-ubyte" % datadir,
"%s/t10k-labels-idx1-ubyte" % datadir)
if idx == None:
i = 0
for p, known_label in test.images(as_array=True):
label, nearest = knn(p, train.images(as_array=True), k)
if label == known_label:
x = ""
else:
x = "XXX"
print "%i %i %i %s" % (i, known_label, label, x)
sys.stdout.flush()
i += 1
else:
img = test.get_image(idx)
label, nearest = knn(img.as_array(), train.images(as_array=True), k)
help="Converge once the centroids move less than this threshold.")
args = parser.parse_args()
return args
args = parse_command_line_arguments()
k = args.k
cutoff = args.cutoff
print "Loading data..."
test = MNISTReader("%s/t10k-images-idx3-ubyte" % args.datadir, "%s/t10k-labels-idx1-ubyte" % args.datadir)
points = None
for img in test.images():
if points is None:
points = numpy.array([img.imgdata])
else:
points = numpy.append(points, [img.imgdata], axis=0)
km = KMeans(points, k)
centroids = km.select_random_centroids()
iteration=1
while True:
print "Iteration #%i" % iteration
new_centroids, point_assignment = km.get_k_means(centroids)
