-
Notifications
You must be signed in to change notification settings - Fork 1
/
kmeans.py
101 lines (87 loc) · 3.98 KB
/
kmeans.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
"""A module to perform kmeans clustering to determine anchor boxes
Performs kmeans-clustering on 1D datasets
"""
import os
import model.utils
import numpy
import tqdm
def euclidean_1d(width, mean):
distance = abs(mean - width)
return distance
def iou_distance(box, mean_box):
pass
def load_bounding_domains(label_dir_path: os.PathLike):
labels = []
for filename in tqdm.tqdm(os.listdir(label_dir_path)):
if filename.endswith(".csv"):
file_path = os.path.join(label_dir_path, filename)
labels.extend(numpy.genfromtxt(fname=file_path,
delimiter=" ",
comments="#",
ndmin=2).tolist())
labels = numpy.array(labels)
bounding_domains = labels[:, 0:2] # throw away label classes, it is not required
return bounding_domains.T
def k_means_clustering(domains: numpy.ndarray,
k: int,
n_trials: int=10,
max_iterations: int=300):
num_domains = domains.shape[-1]
widths = domains[1]
cluster_means_and_variances = \
numpy.zeros((n_trials, k, 2), dtype=float)
for trial_idx in range(n_trials):
# choose random intialization
last_cluster_means = numpy.random.choice(widths, size=(k,), replace=False)
variances = None
for iter_idx in range(max_iterations):
# determine distances from means
distances = numpy.zeros((k, num_domains), dtype=float)
for cluster_idx in range(k):
distances[cluster_idx, :] = abs(widths - last_cluster_means[cluster_idx])
# calculate cluster index of minimum distances
cluster_indices = numpy.argmin(distances, axis=0)
# assign clusters
clusters = [[] for _ in range(k)]
for domain_idx, cluster_idx in enumerate(cluster_indices):
clusters[cluster_idx].append(widths[domain_idx])
# recalculate means for each cluster
new_cluster_means = numpy.array([numpy.mean(cluster) for cluster in clusters])
variances = numpy.array([numpy.var(cluster) for cluster in clusters])
# convergence determined when means of this iter same as means of last iter
if numpy.isclose(new_cluster_means, last_cluster_means, rtol=1e-4).all():
break
elif (iter_idx + 1) == max_iterations:
raise UserWarning("KMeans clustering did not converge on"
f" trial {trial_idx} after {max_iterations}"
" iterations.")
else:
last_cluster_means = new_cluster_means
cluster_means_and_variances[trial_idx, :, 0] = new_cluster_means
cluster_means_and_variances[trial_idx, :, 1] = variances
# the best clusters are the ones with the minimum sum of variances
min_varsum_cluster_means_idx = numpy.argmin(numpy.sum(
cluster_means_and_variances[:, :, 1], axis=1))
# finally, return the optimal cluster means and variances
return cluster_means_and_variances[min_varsum_cluster_means_idx, :, :]
def elbow_plot(domains):
print("calculating kmeans...")
variances = []
for k in tqdm.tqdm(range(1, 9+1)):
means_and_vars = k_means_clustering(domains, k, n_trials=10, max_iterations=300)
sumvar = numpy.sum(means_and_vars[:, 1])
variances.append(sumvar)
import matplotlib.pyplot as plt
plt.plot(range(2, 9+1), numpy.diff(variances))
plt.show()
def calculate_optimal_domain_anchors(domains, k=6):
means_and_vars = k_means_clustering(domains, k, n_trials=10, max_iterations=300)
means = means_and_vars[:, 0]
print(numpy.sort(means))
def main():
print("loading domains...")
domains = load_bounding_domains(os.path.join("data", "test_data", "labels"))
#elbow_plot(domains)
calculate_optimal_domain_anchors(domains)
if __name__ == "__main__":
main()