-
Notifications
You must be signed in to change notification settings - Fork 0
/
gen_smiles.py
146 lines (94 loc) · 3.1 KB
/
gen_smiles.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
from bayes_opt import BayesianOptimization
from utils import utils
from bayes_opt.util import load_logs
from bayes_opt.logger import JSONLogger
from bayes_opt.event import Events
import torch
import numpy as np
import tensorflow as tf
import random
from bayes_opt import SequentialDomainReductionTransformer
import os
import pandas as pd
import moses
import torch
import pickle
from moses.models_storage import ModelsStorage
# from IPython.display import display, Markdown, HTML, clear_output
from utils import utils
MODELS = ModelsStorage()
model_config = torch.load("pretrained/latentgan_config.pt")
model_vocab = torch.load("pretrained/latentgan_vocab.pt")
model_state = torch.load("pretrained/latentgan_model.pt")
model = MODELS.get_model_class("latentgan")(model_vocab, model_config)
model.load_state_dict(model_state)
model = model.cuda()
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('-model_param', help='Iteration of Bayesian Optimzation',default = 'bayesian_result/output.json')
parser.add_argument('-sample', help='Iteration of Bayesian Optimzation',default = 3072)
parser.add_argument('-singular_size', help='Number of Singular Values of Optimized Model ',default = 5)
parser.add_argument('-output', help='Number of Singular Values ',default = 'generated_result/smilesx.txt')
args = parser.parse_args()
model_param = args.model_param
sample = int(args.sample)
singular_size = int(args.singular_size)
output = args.output
df = pd.read_json(model_param, lines=True)
args = parser.parse_args()
singular_size = 5
vector = []
layers = []
for c in model.Generator.model:
if "Linear" in str(type(c)):
v, s, u = utils.svdNeural(c)
vector += [float(e) for e in list(torch.diag(s.weight, 0)[0:singular_size])]
layers.append(c)
# convert df to list
def df_to_list(df):
l = []
for i in range(len(df)):
l.append(df.iloc[i])
return l
df = df_to_list(df)
# get highest score param of df
def get_highest_score(df):
max_score = 0
max_param = {}
for i in range(len(df)):
if df[i]["target"] > max_score:
max_score = df[i]["target"]
max_param = df[i]["params"]
return max_param
max_param = get_highest_score(df)
from rdkit import Chem
from moses.metrics.SA_Score import sascorer
def calculateSA(smi):
try:
return sascorer.calculateScore(smi)
except Exception as e:
print(e)
return 10
def samp(**v):
# utils.clear_tmp()
global singular_size
global sample
num = sample
batch_size = 256
v = {int(k): v[k] for k in v}
vec = utils.dict_to_list(v)
vec = torch.cuda.FloatTensor(vec)
tmp = utils.replaceLayers(vec, layers, singular_size)
result = []
for i in range(num // batch_size):
print(i, "/", num // batch_size)
s = utils.latentGanSample(model, batch_size)
result += s.copy()
# write smiles to files
with open(output.format(i), "a") as f:
for j in range(len(s)):
f.write(
s[j] + "\n"
)
return
samp(**max_param)