-
Notifications
You must be signed in to change notification settings - Fork 0
/
generation_utils.py
332 lines (243 loc) · 17.1 KB
/
generation_utils.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
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
import os
import random
import numpy as np
import torch
import encoding_utils as eutils
import analysis_utils as autils
import VAE
def obtain_domain(input_data, model):
model.eval()
iter = input_data.shape[0] // 1000
if iter > 2:
for i in range(1, iter):
z, _ = model.input_to_latent(input_data[1000*(i-1): 1000*i])
z_batch = z.detach().cpu().numpy()
if i == 1:
z_input = z_batch
else:
z_input = np.vstack((z_input,z_batch))
z, _ = model.input_to_latent(input_data[1000*(iter - 1):])
z_batch = z.detach().cpu().numpy()
z_input = np.vstack((z_input,z_batch))
else:
z, _ = model.input_to_latent(input_data)
z_batch = z.detach().cpu().numpy()
z_input = np.vstack((z_input,z_batch))
z_all_max = np.max(z_input, axis = 0)
z_all_min = np.min(z_input, axis = 0)
z_domain = list(zip(list(z_all_min),list(z_all_max)))
return z_domain
import GPyOpt
from scipy.stats import multivariate_normal
import json
def multi_bo(batch = 20, start_x = None, start_y = None,
model = None, bo_domain = None,
dataset = None,
index_to_smile = None, ordinalenc =None,
log_file = None, use_filter = True):
maxiter = 50
p_weight = 0.0025
mvn = multivariate_normal(mean=np.zeros(128), cov=np.identity(128))
with open(log_file, 'a') as bo_log:
bo_log.write("max iter for gaussian process: " + str(maxiter))
for i in range(0, batch):
while True:
model.eval()
def opt_func(z_opt):
z_opt_cuda = torch.from_numpy(np.array(z_opt)).type(torch.FloatTensor).cuda()
pred_log = torch.log(model.latent_to_prob(z_opt_cuda))
pred_cpu = pred_log.detach().cpu().numpy()
prior = mvn.logpdf(z_opt + 1e-9).reshape(-1, 1)
return pred_cpu - p_weight * prior
# 1. gaussian process
bounds = [{'name': 'var_' + str(i), 'type': 'continuous', 'domain': bo_domain[i-1]} for i in range(1, 129)]
opt_latent = GPyOpt.methods.BayesianOptimization(f= opt_func, domain=bounds)
opt_latent.X = start_x
opt_latent.Y = start_y
opt_latent.run_optimization(max_iter = maxiter)
# translate to molecules
z_gauss = torch.from_numpy(np.array(opt_latent.x_opt)).type(torch.FloatTensor).unsqueeze(0).cuda()
gauss_results = VAE.latent_to_mol(model = model, z_input = z_gauss, batch_size = 1, index_to_smile = index_to_smile, ordinal_encoder = ordinalenc)
# filter
if use_filter:
validity, idx_valid, validity_list = autils.validity_smiles([gauss_results['bb2_sk'][0]])
if validity == 1:
if eutils.is_bb(gauss_results['bb2_sk'][0], bb_type = "bb2"):
novelty = autils.novelty_cages([gauss_results['bb2_sk'][0]], [gauss_results['bb1_sk'][0]], [gauss_results['reaction type'][0]],
idx_valid,
dataset['bb2_skeleton'].tolist(), dataset['bb1_skeleton'].tolist(), dataset['reaction'].tolist())
if novelty == 1:
try:
bb1_string, bb1_factory = eutils.construct_bb(gauss_results['bb1_sk'][0], gauss_results['reaction type'][0])
bb2_string, bb2_factory = eutils.construct_bb(gauss_results['bb2_sk'][0], gauss_results['reaction type'][0])
if autils.point_group_symmetry(gauss_results['bb2_sk'][0]): # or autils.graph_canonical_labeling_symmetry
if autils.wrong_reactionsite_detect(bb2_string, gauss_results['reaction type'][0]):
break
else:
continue
else:
continue
except:
continue
else:
continue
else:
continue
else:
continue
else:
pass
with open(log_file, 'a') as bo_log:
bo_log.write("***Gaussian Process epoch [" + str(i + 1) + "/" + str(batch) +"] \n" +
"P weight:" + str(p_weight) + "\n" +
"Query shape persistency: " + "shape persistent" + "\n" +
"bb1 skeleton: " + str(gauss_results['bb1_sk'][0]) + "\n" +
"bb2 skeleton: " + str(gauss_results['bb2_sk'][0]) + "\n" +
"reaction type: " + str(gauss_results['reaction type'][0]) + "\n")
with open(log_file, 'a') as bo_log:
bo_log.write("Optimized latent vector X: " + "\n")
json.dump(opt_latent.x_opt.tolist(), bo_log)
bo_log.write("\n" + "Optimized Acquisition Function Value Y: " + str(opt_latent.fx_opt) + "\n\n")
if use_filter:
validity, idx_valid, validity_list = autils.validity_smiles([gauss_results['bb2_sk'][0]])
if validity == 0:
with open(log_file, 'a') as bo_log:
bo_log.write('validity?: No' + "\n")
else:
with open(log_file, 'a') as bo_log:
bo_log.write('validity?: Yes' + "\n")
novelty = autils.novelty_cages([gauss_results['bb2_sk'][0]], [gauss_results['bb1_sk'][0]], [gauss_results['reaction type'][0]], idx_valid,
dataset['bb2_skeleton'].tolist(),
dataset['bb1_skeleton'].tolist(),
dataset['reaction'].tolist())
if novelty == 0:
with open(log_file, 'a') as bo_log:
bo_log.write('novelty?: No' + "\n")
else:
with open(log_file, 'a') as bo_log:
bo_log.write('novelty?: Yes' + "\n")
try:
bb1_string, bb1_factory = eutils.construct_bb(gauss_results['bb1_sk'][0], gauss_results['reaction type'][0])
bb2_string, bb2_factory = eutils.construct_bb(gauss_results['bb2_sk'][0], gauss_results['reaction type'][0])
with open(log_file, 'a') as bo_log:
bo_log.write("bb1 string with reactive: " + str(bb1_string) + "\n" +
"bb2 string with reactive: " + str(bb2_string) + "\n" +
"bb1 factory: " + str(bb1_factory) + "\n" +
"bb2 factory: " + str(bb2_factory) + "\n"
)
except:
with open(log_file, 'a') as bo_log:
bo_log.write("problem construct cage" + "\n")
with open(log_file, 'a') as bo_log:
bo_log.write("=============================================================================" + "\n\n")
########## interpolation
def intp_sampling(lerp_type = "slerp-in", steps = 6, threshold = 0.8, model = None,
data_input = None,
nc_idx = None, c_idx = None,
index_to_smile = None, ordinalenc = None
):
index_i, index_2 = np.random.choice(nc_idx, size = 2)
index_3 = np.random.choice(c_idx)
index_f = random.choice([index_2, index_3])
mu_1, _ = model.input_to_latent(data_input[index_i])
mu_2, _ = model.input_to_latent(data_input[index_f])
intp_results, known_z = VAE.interpolation(mu_1.cpu().detach().numpy().T.reshape(-1,), mu_2.cpu().detach().numpy().T.reshape(-1,), steps, model, lerp_type, index_to_smile, ordinalenc) # 这里的 6 是 distance 是一个可以调整的因素
intp_results["probability"] = model.latent_to_prob(known_z).detach().cpu().numpy()
intp_results['min'] = intp_results.groupby('bb2_sk')['probability'].transform('min')
intp_results['max'] = intp_results.groupby('bb2_sk')['probability'].transform('max')
intp_results['gap'] = intp_results['max'] - intp_results['min']
intp_filtered = intp_results[intp_results['gap'] < 0.2]
intp_filtered = intp_results[intp_results['probability'] < (1-threshold)]
intp_deduplicated = intp_filtered.loc[intp_filtered.groupby('bb2_sk')['probability'].idxmin()]
intp_deduplicated.drop(['min', 'max', 'gap'], axis=1, inplace = True)
intp_deduplicated.reset_index(inplace = True)
return intp_deduplicated
def interpolation_generation(lerp_type = "slerp-in", steps = 6, threshold = 0.8, model = None,
data_input = None,
nc_idx = None, c_idx = None,
index_to_smile = None, ordinalenc = None,
dataset = None,
batch = None,
current_batch = None,
log_file = None, use_filter = True,
):
while True:
intp_deduplicated = intp_sampling(lerp_type = lerp_type, steps = steps, threshold = threshold, model = model,
data_input = data_input,
nc_idx = nc_idx, c_idx = c_idx,
index_to_smile = index_to_smile, ordinalenc = ordinalenc
)
num = 0
for i in range(0, intp_deduplicated.shape[0]):
if use_filter:
validity, idx_valid, validity_list = autils.validity_smiles([intp_deduplicated['bb2_sk'].iloc[i]])
if validity == 1:
if eutils.is_bb(intp_deduplicated['bb2_sk'].iloc[i], bb_type = "bb2"):
novelty = autils.novelty_cages([intp_deduplicated['bb2_sk'].iloc[i]], [intp_deduplicated['bb1_sk'].iloc[i]], [intp_deduplicated['reaction type'].iloc[i]],
idx_valid,
dataset['bb2_skeleton'].tolist(), dataset['bb1_skeleton'].tolist(), dataset['reaction'].tolist())
if novelty == 1:
try:
bb1_string, bb1_factory = eutils.construct_bb(intp_deduplicated['bb1_sk'].iloc[i], intp_deduplicated['reaction type'].iloc[i])
bb2_string, bb2_factory = eutils.construct_bb(intp_deduplicated['bb2_sk'].iloc[i], intp_deduplicated['reaction type'].iloc[i])
if autils.point_group_symmetry(intp_deduplicated['bb2_sk'].iloc[i]): # or autils.graph_canonical_labeling_symmetry
if autils.wrong_reactionsite_detect(bb2_string, intp_deduplicated['reaction type'].iloc[i]):
num = num + 1
if not os.path.exists('./log'):
os.makedirs('./log')
with open(log_file, 'a') as intp_log:
intp_log.write("***Interpolation epoch [" + str(current_batch) + "/" + str(batch) +"] \n" +
"number in this batch: " + str(num) + "\n" +
"Query shape persistency: " + "shape persistent" + "\n" +
"bb1 skeleton: " + str(intp_deduplicated['bb1_sk'].iloc[i]) + "\n" +
"bb2 skeleton: " + str(intp_deduplicated['bb2_sk'].iloc[i]) + "\n" +
"reaction type: " + str(intp_deduplicated['reaction type'].iloc[i]) + "\n")
if use_filter:
validity, idx_valid, validity_list = autils.validity_smiles([intp_deduplicated['bb2_sk'].iloc[i]])
if validity == 0:
with open(log_file, 'a') as intp_log:
intp_log.write('validity?: No' + "\n")
else:
with open(log_file, 'a') as intp_log:
intp_log.write('validity?: Yes' + "\n")
novelty = autils.novelty_cages([intp_deduplicated['bb2_sk'].iloc[i]], [intp_deduplicated['bb1_sk'].iloc[i]], [intp_deduplicated['reaction type'].iloc[i]], idx_valid,
dataset['bb2_skeleton'].tolist(),
dataset['bb1_skeleton'].tolist(),
dataset['reaction'].tolist())
if novelty == 0:
with open(log_file, 'a') as intp_logg:
intp_log.write('novelty?: No' + "\n")
else:
with open(log_file, 'a') as intp_log:
intp_log.write('novelty?: Yes' + "\n")
try:
bb1_string, bb1_factory = eutils.construct_bb(intp_deduplicated['bb1_sk'].iloc[i], intp_deduplicated['reaction type'].iloc[i])
bb2_string, bb2_factory = eutils.construct_bb(intp_deduplicated['bb2_sk'].iloc[i], intp_deduplicated['reaction type'].iloc[i])
with open(log_file, 'a') as intp_log:
intp_log.write("bb1 string with reactive: " + str(bb1_string) + "\n" +
"bb2 string with reactive: " + str(bb2_string) + "\n" +
"bb1 factory: " + str(bb1_factory) + "\n" +
"bb2 factory: " + str(bb2_factory) + "\n"
)
except:
with open(log_file, 'a') as intp_log:
intp_log.write("problem construct cage" + "\n")
with open(log_file, 'a') as intp_log:
intp_log.write("=============================================================================" + "\n\n")
else:
continue
else:
continue
except:
continue
else:
continue
else:
continue
else:
continue
else:
pass
current_batch = current_batch + num
if current_batch >= batch + 1:
break