-
Notifications
You must be signed in to change notification settings - Fork 1
/
Seed_Tuning.py
142 lines (120 loc) · 6.69 KB
/
Seed_Tuning.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
# -*- coding: utf-8 -*-
# @Time : 2022/4/3 10:54
# @Author : FAN FAN
# @Site :
# @File : test.py
# @Software: PyCharm
from src.feature.atom_featurizer import classic_atom_featurizer
from src.feature.bond_featurizer import classic_bond_featurizer
from src.feature.mol_featurizer import classic_mol_featurizer
from utils.mol2graph import smiles_2_bigraph
from utils.splitter import Splitter
from utils.metrics import Metrics
from utils.Earlystopping import EarlyStopping
from data.csv_dataset import MoleculeCSVDataset
from src.dgltools import collate_molgraphs
from data.dataloading import import_dataset
import torch
from torch.utils.data import DataLoader
import os
import time
import numpy as np
import pandas as pd
from tqdm import tqdm
from utils.count_parameters import count_parameters
from networks.DMPNN import DMPNNNet
from networks.MPNN import MPNNNet
from networks.AttentiveFP import AttentiveFPNet
from utils.piplines import train_epoch, evaluate
from utils.Set_Seed_Reproducibility import set_seed
params = {}
net_params = {}
#params['Dataset'] = 'HFUS'
params['init_lr'] = 10 ** -3.2189
params['min_lr'] = 1e-9
params['weight_decay'] = 0
params['lr_reduce_factor'] = 0.2
params['lr_schedule_patience'] = 30
params['earlystopping_patience'] = 50
params['max_epoch'] = 300
net_params['num_atom_type'] = 36
net_params['num_bond_type'] = 12
net_params['hidden_dim'] = 16
net_params['dropout'] = 0
net_params['depth'] = 2
net_params['layers'] = 2
net_params['residual'] = False
net_params['batch_norm'] = False
net_params['layer_norm'] = False
net_params['device'] = 'cuda'
dataset_list = ['ESOL', 'BMP']
for i in range(len(dataset_list)):
params['Dataset'] = dataset_list[i]
df, scaling = import_dataset(params)
cache_file_path = os.path.realpath('./cache')
if not os.path.exists(cache_file_path):
os.mkdir(cache_file_path)
cache_file = os.path.join(cache_file_path, params['Dataset'] + '_CCC')
error_path = os.path.realpath('./error_log')
if not os.path.exists(error_path):
os.mkdir(error_path)
error_log_path = os.path.join(error_path, '{}_{}'.format(params['Dataset'], time.strftime('%Y-%m-%d-%H-%M')) + '.csv')
dataset = MoleculeCSVDataset(df, smiles_2_bigraph, classic_atom_featurizer, classic_bond_featurizer, classic_mol_featurizer, cache_file, load=True
, error_log=error_log_path)
splitter = Splitter(dataset)
rows = []
file_path = os.path.realpath('./output')
if not os.path.exists(file_path):
os.mkdir(file_path)
save_file_path = os.path.join(file_path, '{}_{}_{}'.format(params['Dataset'], 'AFP', time.strftime('%Y-%m-%d-%H-%M')) + '.csv')
df = pd.DataFrame(columns=['seed', 'train_R2', 'val_R2', 'test_R2', 'all_R2', 'train_MAE', 'val_MAE', 'test_MAE', 'all_MAE',
'train_RMSE', 'val_RMSE', 'test_RMSE', 'all_RMSE'])
for i in range(0, 500):
seed = np.random.randint(1, 5000)
set_seed(seed=1000)
train_set, val_set, test_set, raw_set = splitter.Random_Splitter(seed=seed, frac_train=0.8, frac_val=0.1)
train_loader = DataLoader(train_set, collate_fn=collate_molgraphs, batch_size=len(train_set), shuffle=False)
val_loader = DataLoader(val_set, collate_fn=collate_molgraphs, batch_size=len(val_set), shuffle=False)
test_loader = DataLoader(test_set, collate_fn=collate_molgraphs, batch_size=len(test_set), shuffle=False)
raw_loader = DataLoader(raw_set, collate_fn=collate_molgraphs, batch_size=len(raw_set), shuffle=False)
model = AttentiveFPNet(net_params).to(device='cuda')
optimizer = torch.optim.Adam(model.parameters(), lr = params['init_lr'], weight_decay = params['weight_decay'])
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=params['lr_reduce_factor'],
patience=params['lr_schedule_patience'], verbose=False)
t0 = time.time()
per_epoch_time = []
early_stopping = EarlyStopping(patience=params['earlystopping_patience'])
with tqdm(range(params['max_epoch'])) as t:
n_param = count_parameters(model)
for epoch in t:
t.set_description('Epoch %d' % epoch)
start = time.time()
model, epoch_train_loss, epoch_train_metrics = train_epoch(model, optimizer, scaling, train_loader, n_param)
epoch_val_loss, epoch_val_metrics = evaluate(model, scaling, val_loader, n_param)
epoch_test_loss, epoch_test_metrics = evaluate(model, scaling, test_loader, n_param)
t.set_postfix({'time': time.time() - start, 'lr': optimizer.param_groups[0]['lr'],
'train_loss': epoch_train_loss, 'val_loss': epoch_val_loss, 'test_loss': epoch_test_loss,
'train_R2': epoch_train_metrics.R2, 'val_R2': epoch_val_metrics.R2, 'test_R2': epoch_test_metrics.R2})
per_epoch_time.append(time.time() - start)
scheduler.step(epoch_val_loss)
if optimizer.param_groups[0]['lr'] < params['min_lr']:
print('\n! LR equal to min LR set.')
break
early_stopping(epoch_val_loss, model)
if early_stopping.early_stop:
break
_, epoch_raw_metrics = evaluate(model, scaling, raw_loader, n_param)
#rows.append({'seed': seed, 'train_R2': epoch_train_metrics.R2, 'val_R2': epoch_val_metrics.R2,
#'test_R2': epoch_test_metrics.R2, 'all_R2': epoch_raw_metrics.R2,
#'train_MAE': epoch_train_metrics.MAE, 'val_MAE': epoch_val_metrics.MAE,
#'test_MAE': epoch_test_metrics.MAE, 'all_MAE': epoch_raw_metrics.MAE,
#'train_RMSE': epoch_train_metrics.RMSE, 'val_RMSE': epoch_val_metrics.RMSE,
#'test_RMSE': epoch_test_metrics.RMSE, 'all_RMSE': epoch_raw_metrics.RMSE})
row = pd.Series({'seed': seed, 'train_R2': epoch_train_metrics.R2, 'val_R2': epoch_val_metrics.R2,
'test_R2': epoch_test_metrics.R2, 'all_R2': epoch_raw_metrics.R2,
'train_MAE': epoch_train_metrics.MAE, 'val_MAE': epoch_val_metrics.MAE,
'test_MAE': epoch_test_metrics.MAE, 'all_MAE': epoch_raw_metrics.MAE,
'train_RMSE': epoch_train_metrics.RMSE, 'val_RMSE': epoch_val_metrics.RMSE,
'test_RMSE': epoch_test_metrics.RMSE, 'all_RMSE': epoch_raw_metrics.RMSE})
df = df.append(row, ignore_index=True)
df.to_csv(save_file_path)