def _kmeans(descriptors):
""" Map a set of brisk descriptors (each descriptor is 64 floats) into a
pre-training cluster space. Each descriptor maps to an integer (i.e.
the cluster id) so we transform the (n,64) descriptor matrix into
an n-vector. A k-histogram of the n-vector is returned. """
kmeans_model = KMeans(centroids=centroids)
cluster_ids = [kmeans_model.classify(d) for d in descriptors]
""" Take the multiplicity of each cluster id in the mapped descriptor space.
The n-vector of cluster ids becomes a histogram of cardinality 128 """
histogram = [0] * brisk_constants.N_CLUSTERS
for cluster_id in set(cluster_ids):
histogram[cluster_id] = cluster_ids.count(cluster_id)
return histogram
kmeans.convergence = 0.001 # 收敛误差
kmeans.setup() # 在开始训练前跑个设定
# 每迭代的回呼函式
def iteration_callback(iteration_times, groups):
print("iteration %r" % iteration_times)
for group in groups:
print("新旧群心 （%r） : % r -> %r" %
(group.tag, group.old_center, group.center))
# 完成训练时的回呼函式
def completion_callback(iteration_times, groups, sse):
print("completion %r and sse %r" % (iteration_times, sse))
for group in groups:
print("最终的群心 （%r）: %r, %r" % (group.tag, group.center, group.samples))
# 开始训练
kmeans.run(iteration_callback, completion_callback)
# 对新样本进行分类后的结果回呼函式
def classified_callback(point, group):
print("%r classified to %r" % (point, group.tag))
# 用已训练好的模型来对新样本进行分类
kmeans.classify([[3, 4], [5, 6]], classified_callback)
def main():
print("Clustering Fun\n")
verbose = False
k = None
plot = False
num_points = None
csv_path = None
lim_distance = None
argCount = len(sys.argv)
args = sys.argv
locations = None
for i, arg in enumerate(args):
if i == 0:
continue # ignore script name
if arg.startswith("-"):
if arg.startswith('-k='):
k = int(arg.split('=', 1)[1])
elif arg.startswith('-n='):
num_points = int(arg.split('=', 1)[1])
elif arg.startswith('-l='):
lim_distance = int(arg.split('=', 1)[1])
elif arg == "-v":
verbose = True
elif arg == "-p":
plot = True
else:
print(f"Unknown argument: {arg}")
print_usage()
exit(1)
elif not csv_path:
csv_path = arg
else:
print(f"Too many parameters!")
print_usage()
exit(1)
if not csv_path and not num_points or not k:
print(f"Too few parameters!")
print_usage()
exit(1)
if verbose:
print_options(verbose, csv_path, k, plot, lim_distance)
if csv_path:
locations = read_csv(csv_path)
if verbose:
print(f"\nRead {len(locations)} locations from {csv_path}")
else:
locations = generate_points_gauss(num_points, k)
print(f"\nSetting up K-Means Classifier:")
km = KMeans(locations, k, verbose, dist_limit=lim_distance)
print(f"Finding best initial center points with K-Means++")
centers = np.array(np.empty)
min_var = None
for i in range(8):
tmp_centers, variance = km.init_centers()
if verbose:
print(f" Iteration {i}, variance = {variance} km")
if min_var is None or variance < min_var:
min_var = variance
centers = np.copy(tmp_centers)
if plot and verbose:
plot_locations(
locations,
k,
centers,
img_tag=f"{'' if not csv_path else csv_path+'_'}initial_state")
centers, locations = km.classify(centers)
if plot:
plot_locations(locations,
k,
centers,
img_tag=f"{'' if not csv_path else csv_path+'_'}")
print(f"\nClass centers:")
for idx, c in enumerate(centers):
print(f' Class = {idx}:')
print(f' Center = {c}')
print(
f' Farthest point from center = {max([loc.class_distance for loc in locations if loc.classification == idx])} km'
)
print(
f' Attribute1 = {any(loc.attr1 for loc in locations if loc.classification == idx)} (Logical OR)'
)
print(
f' Attribute2 = {any(loc.attr2 for loc in locations if loc.classification == idx)} (Logical OR)'
)
print()
if lim_distance:
print(f"\nWith distance limit of {lim_distance} km:")
print(
f" {len([loc for loc in locations if loc.classification == None])} locations are left unclassified"
)
