-
Notifications
You must be signed in to change notification settings - Fork 4
/
features.py
263 lines (211 loc) · 8.04 KB
/
features.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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
import sys
import time
import numpy as np
np.seterr(divide='ignore', invalid='ignore')
import pandas as pd
from scipy.spatial import distance_matrix
zlist = [1.0, 6.0, 7.0, 8.0, 9.0, 16.0]
charge_to_index = { 0.0 : 100000,
1.0 : 0,
6.0 : 1,
7.0 : 2,
8.0 : 3,
9.0 : 4,
16.0 : 5,
}
def cutoff(D, rc1=0.0, rc2=8.0):
"""
Cutoff function used to ensure locality of features.
Args:
D: distance matrix with dimensions [N x N]
rc1 : min cutoff distance
rc2 : max cutoff distance
Returns:
cutoff matrix with same shape as D
"""
C = (np.cos(np.pi * (D - rc1) / (rc2 - rc1)) + 1.0) / 2.0
C[D >= rc2] = 0.0
C[D <= rc1] = 1.0
np.fill_diagonal(C, 0.0)
return C
def cos_theta(R):
"""
Calculates the cosine between all possible angles given a set of points.
Args:
R: cartesian coordinates of a monomer or dimer [NATOM x 3]
Returns three index tensor ct, where ct[i,j,k] is cosine(theta_jik).
The first index (i) is the center point of the angle.
i, j, and k index atoms in R.
"""
ct = np.zeros((R.shape[0], R.shape[0], R.shape[0]))
for i in range(R.shape[0]):
dRxyz = R - R[i]
dR1 = np.linalg.norm(dRxyz, axis=1)
num = np.inner(dRxyz, dRxyz)
denom = np.outer(dR1, dR1)
ct[i,:,:] = (num / denom)
return np.nan_to_num(ct)
def cos_theta_im(RA, RB):
"""
Calculates the cosine between all possible intermolecular angles given set of points in two monomers.
Args:
RA: cartesian coordinates of monomer A
RB: cartesian coordinates of monomer B
Returns three index tensor ct, where ct[i,j,k] is cosine(theta_jik).
The first index (i) is the center point of the angle.
i and k index atoms in RA while j indexes atoms in RB.
"""
ct = np.zeros((RA.shape[0], RB.shape[0], RA.shape[0]))
for i in range(RA.shape[0]):
dRAxyz = RA - RA[i]
dRBxyz = RB - RA[i]
dRA1 = np.linalg.norm(dRAxyz, axis=1)
dRB1 = np.linalg.norm(dRBxyz, axis=1)
num = np.inner(dRBxyz, dRAxyz)
denom = np.outer(dRB1, dRA1)
ct[i,:,:] = (num / denom)
return np.nan_to_num(ct)
def acsfs(Z, R, mus, eta=100.0, rc1=0.0, rc2=8.0):
"""
Calculates the radial atom centered symmetry functions of all atoms in a monomer
Args:
Z: atom type array with dimensions [NATOM]
R: atom coordinate array with dimensions [NATOM x 3]
mus: ACSF descriptor shift parameter array with dimensions [NMU]
eta: ACSF descriptor gaussian width parameter
rc1: cutoff function first parameter
rc2: cutoff function second parameter
Returns a tensor containing the radial ACSFs for all atoms in a monomer with dimensions [NATOM x NMU X NZ]
"""
natom = len(R)
nmu = len(mus)
ntype = len(zlist)
zindex = [charge_to_index[z] for z in Z]
Dxyz = distance_matrix(R, R)
C = cutoff(Dxyz, rc1, rc2)
G = np.zeros((natom, nmu, ntype))
for ind1 in range(natom):
Dxyz_atom = Dxyz[ind1]
for ind2, d in enumerate(Dxyz_atom):
if d >= rc2 or zindex[ind1] > 100 or zindex[ind2] > 100:
continue
G[ind1,:,zindex[ind2]] += np.exp(-1.0 * np.square(d - mus) * eta) * C[ind1, ind2]
return G
def apsfs(ZA, RA, ZB, RB, mus, eta=100.0, rc1=0.0, rc2=8.0):
"""
Calculates the angular atom pair symmetry functions of all pairs atoms in two monomers
Args:
ZA: monomer A atom type array with dimensions [NATOMA]
RA: monomer A atom coordinate array with dimensions [NATOM x 3]
ZB: monomer B atom type array with dimensions [NATOMB]
RB: monomer B atom coordinate array with dimensions [NATOM x 3]
mus: APSF descriptor shift parameter array with dimensions [NMU]
eta: APSF descriptor gaussian width parameter
rc1: cutoff function first parameter
rc2: cutoff function second parameter
Returns a tensor containing the radial ACSFs for all atoms in a monomer with dimensions [NATOM x NMU X NZ]
"""
natomA = len(RA)
natomB = len(RB)
nmu = len(mus)
ntype = len(zlist)
zindexA = [charge_to_index[z] for z in ZA]
zindexB = [charge_to_index[z] for z in ZB]
DAA = distance_matrix(RA, RA)
CAA = cutoff(DAA, rc1, rc2)
DAB = distance_matrix(RA, RB)
#CAB = cutoff(DAB, rc1, rc2)
A = cos_theta_im(RA, RB) # angles
G = np.zeros((natomA, natomB, nmu, ntype))
for a_A in range(natomA):
e_A = zindexA[a_A]
if e_A > 100:
continue
for a_B, d_AB in enumerate(DAB[a_A]):
e_B = zindexB[a_B]
if d_AB > rc2 or e_B > 100:
continue
for a_A2, d_AA2 in enumerate(DAA[a_A]):
e_A2 = zindexA[a_A2]
if d_AA2 > rc2 or e_A2 > 100:
continue
c_th = A[a_A, a_B, a_A2]
G[a_A, a_B, :, e_A2] += np.exp(-1.0 * np.square(c_th - mus) * eta) * CAA[a_A, a_A2] #* C[a_j, a_k]
return G
#def asyms(Z, R, mus, eta=100.0, rc1=0.0, rc2=8.0):
# """
# Calculates the angular atom centered symmetry functions of all atoms in a monomer
# """
#
# natom = len(R)
# nmu = len(mus)
# ntype = len(zlist)
#
# zindex = [charge_to_index[z] for z in Z]
#
# D = distance_matrix(R, R)
# C = cutoff(D, rc1, rc2)
# A = cos_theta(R)
# G = np.zeros((natom, nmu, ntype, ntype))
#
# for a_j in range(natom):
# e_j = zindex[a_j]
# if e_j > 100:
# continue
# for a_i, d_i in enumerate(D[a_j]):
# e_i = zindex[a_i]
# if d_i > rc2 or e_i > 100:
# continue
# for a_k, d_k in enumerate(D[a_j]):
# e_k = zindex[a_k]
# if d_k > rc2 or e_k > 100:
# continue
# c_th = A[a_j, a_i, a_k]
# G[a_j,:,e_i,e_k] += np.exp(-1.0 * np.square(c_th - mus) * eta) * C[a_j, a_i] * C[a_j, a_k]
# if e_k != e_i:
# G[a_j,:,e_k,e_i] += np.exp(-1.0 * np.square(c_th - mus) * eta) * C[a_j, a_k] * C[a_j, a_i]
# return G * 0.5
def calculate(dataset, ACSF_nmu=43, APSF_nmu=21, ACSF_eta=100, APSF_eta=25):
"""
Calculate and save ACSF and APSF features for a given dataset
"""
df_dimers = pd.read_pickle(f'datasets/{dataset}/dimers.pkl')
N_dimer = len(df_dimers.index)
# get coordinates and atom types, catch common errors
try:
df_ACSF = df_dimers[['RA', 'RB', 'ZA', 'ZB']].copy(deep=True)
df_APSF = df_dimers[['RA', 'RB', 'ZA', 'ZB']].copy(deep=True)
except KeyError:
print('Error: Dataset needs fields "RA", "RB", "ZA", and "ZB" for every dimer\n')
raise
# TODO: experiment with these ranges
ACSF_mus = np.linspace(0.8, 5.0, ACSF_nmu)
APSF_mus = np.linspace(-1.0, 1.0, APSF_nmu)
ACSF_A = []
ACSF_B = []
APSF_A_B = []
APSF_B_A = []
counter = 0
start = time.time()
for index, row in df_dimers.iterrows():
RA = row['RA']
RB = row['RB']
ZA = row['ZA']
ZB = row['ZB']
ACSF_A.append(acsfs(ZA, RA, ACSF_mus, ACSF_eta, 0.0, 8.0))
ACSF_B.append(acsfs(ZB, RB, ACSF_mus, ACSF_eta, 0.0, 8.0))
APSF_A_B.append(apsfs(ZA, RA, ZB, RB, APSF_mus, APSF_eta, 0.0, 8.0))
APSF_B_A.append(np.transpose(apsfs(ZB, RB, ZA, RA, APSF_mus, APSF_eta, 0.0, 8.0), axes=[1,0,2,3]))
if (counter+1) % 500 == 0:
dt = time.time() - start
rate = (counter + 1) / dt
print(f'{(counter+1):5d} of {N_dimer:5d} dimers done in {int(dt):6d} s ({int(rate)} dimers / s)')
counter += 1
df_ACSF['ACSF_A'] = ACSF_A
df_ACSF['ACSF_B'] = ACSF_B
df_APSF['APSF_A_B'] = APSF_A_B
df_APSF['APSF_B_A'] = APSF_B_A
df_ACSF.to_pickle(f'datasets/{dataset}/ACSF_{ACSF_nmu}_{ACSF_eta}.pkl')
df_APSF.to_pickle(f'datasets/{dataset}/APSF_{APSF_nmu}_{APSF_eta}.pkl')
if __name__ == '__main__':
pass