def streaming_kmeans(points, k=10, num_iters=10, num_ballkmeans_runs=2, trim_factor=0.9,
test_probability=0.1, correct_weight=False):
'''
clustering data points using streaming kmeans method.
Args:
points(DistArray): data points to be clustered.
k(int): the final number of clusters.
num_iters(int): the number of iterations to run in each ball kmeans run.
num_ballkmeans_runs(int): the number of ball kmeans to run.
trim_factor(float): the ball kmeans parameter to separate the nearest points and distant points.
test_probability(float): the percentage of points to be chosen as test set.
correct_weights(bool): whether to correct the weights of the centroids.
'''
centroids = expr.tile_operation(points,
_streaming_mapper,
kw={'k': k}).evaluate()
new_centroids = []
for tile_result in centroids.values():
for centroids_list in tile_result:
new_centroids.extend(centroids_list)
centriods = ball_kmeans(new_centroids, k, num_iters, num_ballkmeans_runs, trim_factor,
test_probability, correct_weight)
centers = np.zeros((k, points.shape[1]))
for i in range(k):
centers[i] = centriods[i].get_center()
return expr.shuffle(points, _cluster_mapper,
kw={'centers': centers}, shape_hint=(points.shape[0],))
def canopy_cluster(points, t1=0.1, t2=0.1, cf=1):
'''
A simple implementation of canopy clustering method.
Args:
points(Expr or DistArray): the input data points matrix.
t1(float): distance threshold between center point and the points within a canopy.
t2(float): distance threshold between center point and the points within a canopy.
cf(int): the minimum canopy size.
'''
new_points = expr.tile_operation(points, _canopy_mapper, kw={'t1': t1, 't2': t2, 'cf': cf}).evaluate()
centers = find_centers(new_points.values(), t1, t2, cf)
labels = expr.shuffle(points, _cluster_mapper, kw={'centers': centers}, shape_hint=(points.shape[0],))
return labels
def streaming_kmeans(points,
k=10,
num_iters=10,
num_ballkmeans_runs=2,
trim_factor=0.9,
test_probability=0.1,
correct_weight=False):
'''
clustering data points using streaming kmeans method.
Args:
points(DistArray): data points to be clustered.
k(int): the final number of clusters.
num_iters(int): the number of iterations to run in each ball kmeans run.
num_ballkmeans_runs(int): the number of ball kmeans to run.
trim_factor(float): the ball kmeans parameter to separate the nearest points and distant points.
test_probability(float): the percentage of points to be chosen as test set.
correct_weights(bool): whether to correct the weights of the centroids.
'''
centroids = expr.tile_operation(points, _streaming_mapper, kw={
'k': k
}).evaluate()
new_centroids = []
for tile_result in centroids.values():
for centroids_list in tile_result:
new_centroids.extend(centroids_list)
centriods = ball_kmeans(new_centroids, k, num_iters, num_ballkmeans_runs,
trim_factor, test_probability, correct_weight)
centers = np.zeros((k, points.shape[1]))
for i in range(k):
centers[i] = centriods[i].get_center()
return expr.shuffle(points,
_cluster_mapper,
kw={'centers': centers},
shape_hint=(points.shape[0], ))
def canopy_cluster(points, t1=0.1, t2=0.1, cf=1):
'''
A simple implementation of canopy clustering method.
Args:
points(Expr or DistArray): the input data points matrix.
t1(float): distance threshold between center point and the points within a canopy.
t2(float): distance threshold between center point and the points within a canopy.
cf(int): the minimum canopy size.
'''
new_points = expr.tile_operation(points,
_canopy_mapper,
kw={
't1': t1,
't2': t2,
'cf': cf
}).force()
centers = find_centers(new_points.values(), t1, t2, cf)
labels = expr.shuffle(points,
_cluster_mapper,
kw={'centers': centers},
shape_hint=(points.shape[0], ))
return labels
