/
bond_cluster.py
162 lines (144 loc) · 5.64 KB
/
bond_cluster.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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
from log import log
import math
import numpy as np
from progress.bar import Bar
from ase import Atoms
from itertools import combinations
from ase.db import connect
from ase.visualize import view
from sklearn.datasets.samples_generator import make_blobs
from sklearn import metrics
from sklearn.cluster import AffinityPropagation
# Plot result
import matplotlib.pyplot as plt
from itertools import cycle
from sklearn.cluster import KMeans
from sklearn.metrics import silhouette_score
import numpy as np
import random
seed = 1234
random.seed(seed)
np.random.seed(seed)
# tf.set_random_seed(seed)
print(__doc__)
# #############################################################################
# Generate sample data
# centers = [[1, 1], [-1, -1], [1, -1]]
# Y, labels_true = make_blobs(n_samples=300, centers=centers, cluster_std=0.5,
# random_state=0)
# print(Y, labels_true)
results = []
total_cluster_cnt = 0
def trim_axs(axs, N):
"""little helper to massage the axs list to have correct length..."""
axs = axs.flat
for ax in axs[N:]:
ax.remove()
return axs[:N]
def atoms_clustering(sym1, sym2, ax):
global total_cluster_cnt
log.logger.info('new round start ' + sym1 + '-' + sym2)
cluster_finding_distance = 5.0 ## must same with main!!!
X = [[i] for i in range(10000)]
Y = [0 for i in range(10000)] # 0.001A from 0A to 10A
db = connect('./qm9.db')
bar = Bar('finding '+sym1 + '-' + sym2 + ' bonds in db', max=db.count())
for row in db.select():
atoms = row.toatoms()
for i in range(len(atoms)):
for j in range(len(atoms)):
if(i != j and atoms[i].symbol == sym1 and atoms[j].symbol == sym2
and atoms.get_distance(i, j) < cluster_finding_distance):
Y[int(atoms.get_distance(i, j)*1000)] += 1
bar.next()
#if(row.id == 2000):
# break
bar.finish()
non_zero_X = [X[i] for i in range(10000) if Y[i] >= int(max(Y)/20)]
non_zero_Y = [Y[i] for i in range(10000) if Y[i] >= int(max(Y)/20)]
# https://www.codercto.com/a/23218.html
fit_scores = []
ret_found = False
if((sym1 == 'F' and sym2 == 'O') or (sym1 == 'F' and sym2 == 'F')):
pass
else:
for i in range(2, 20):
init_x = []
km = KMeans(n_clusters=i, init='random')
km.fit(X=non_zero_X, sample_weight=non_zero_Y)
#sc_x = []
#sc_y = []
lst_x = []
# sc don't have sample_weight, so ... brute force
for j in range(len(non_zero_X)):
#sc_x += [non_zero_X[j]] * non_zero_Y[j]
#sc_y += [km.labels_[j]] * non_zero_Y[j]
lst_x.append(non_zero_X[j][0])
# https://blog.wavelabs.ai/jenks-natural-breaks-optimization-finder/
# but jenks don't have weight in it. so write myself...
sdam = 0
for idx, x in enumerate(lst_x):
sdam += (x - np.array(lst_x).mean())**2 * non_zero_Y[idx]
sdcm = 0
for l in range(i):
sub_sum = sum([lst_x[k]*non_zero_Y[k] for k in range(len(lst_x)) if km.labels_[k]==l])
sub_cnt = sum([non_zero_Y[k] for k in range(len(lst_x)) if km.labels_[k]==l])
sub_mean = sub_sum/sub_cnt
sdcm += sum([(lst_x[k] - sub_mean)**2*non_zero_Y[k] for k in range(len(lst_x)) if km.labels_[k]==l])
gvf = 1-sdcm/sdam
fit_scores.append(gvf) # GVF > 99.5% we choose
if(gvf > 0.995 and ret_found == False):
ret_found = True
results.append([sym1, sym2, i, gvf, km.cluster_centers_])
total_cluster_cnt += i
# # sc can't figure out short distance C-C
# sc_score = silhouette_score(sc_x, sc_y, metric='euclidean', sample_size=20000)
# #sc_score = silhouette_score(non_zero_X, km.labels_, metric='euclidean')
# sc_scores.append(sc_score)
log.logger.info(str(i))
log.logger.info(str(km.cluster_centers_))
log.logger.info(str(results))
log.logger.info(str(fit_scores))
# plt.close('all')
# plt.figure(1)
# plt.clf()
'''
for ax, case in zip(axs, cases):
ax.set_title('markevery=%s' % str(case))
ax.plot(x, y, 'o', ls='-', ms=4, markevery=case)
'''
ax.set_title(sym1 + '-' + sym2)
ax.plot(list(np.array(non_zero_X)/1000.0), non_zero_Y)
# print(results[0][4])
if((sym1 == 'F' and sym2 == 'O') or (sym1 == 'F' and sym2 == 'F')):
pass
else:
for i in range(len(results[-1][4])):
ax.plot([results[-1][4][i][0]/1000, results[-1][4][i][0]/1000], [0, max(non_zero_Y)], linestyle=':')
# plt.savefig(sym1 + '-' + sym2 + ' distance distribution')
# plt.figure(2)
# x = [i for i in range(2, 20)]
# plt.plot(x, fit_scores)
# plt.savefig(sym1 + '-' + sym2 + 'sc_scores')
# plt.show()
if __name__ == '__main__':
symbols = ['C', 'H', 'O', 'N', 'F']
# symbols = ['C']
figsize = (10, 8)
cols = 5
rows = 3
fig1, axs = plt.subplots(rows, cols, figsize=figsize, constrained_layout=True)
axs = trim_axs(axs, cols*rows)
ax_idx = 0
for idx1, sym1 in enumerate(symbols):
for idx2, sym2 in enumerate(symbols):
if(idx1 < idx2):
# F cnt so small F-F only 18 so pass it
continue
else:
atoms_clustering(sym1, sym2, axs[ax_idx])
ax_idx += 1
axs[5].set_ylabel('pair count')
axs[12].set_xlabel('pair distance, ' + r'$\AA$')
log.logger.info('total_cluster_cnt: ' + str(total_cluster_cnt))
plt.show()