-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpredict_retro.py
173 lines (145 loc) · 8.11 KB
/
predict_retro.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
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"] = "2"
import argparse
from utils.smiles_utils import *
from utils.translate_utils import translate_batch
from utils.build_utils import build_model, build_retro_iterator, load_checkpoint_downstream
from datetime import datetime
from utils.logging import init_logger, TensorboardLogger
import json
import copy
import pickle
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from tqdm import tqdm
import warnings
warnings.filterwarnings("ignore")
from rdkit import RDLogger
RDLogger.DisableLog('rdApp.*')
def arg_parse():
parser = argparse.ArgumentParser()
parser.add_argument('--device', type=str, default='cuda', help='device GPU/CPU')
parser.add_argument('--batch_size_trn', type=int, default=32, help='batch size')
parser.add_argument('--batch_size_val', type=int, default=32, help='batch size')
parser.add_argument('--beam_size', type=int, default=10, help='beam size')
parser.add_argument('--encoder_num_layers', type=int, default=4, help='number of layers of transformer')
parser.add_argument('--decoder_num_layers', type=int, default=4, help='number of layers of transformer')
parser.add_argument('--d_model', type=int, default=256, help='dimension of model representation')
parser.add_argument('--heads', type=int, default=8, help='number of heads of multi-head attention')
parser.add_argument('--d_ff', type=int, default=2048, help='')
parser.add_argument('--max_length', type=int, default=200)
parser.add_argument('--dropout', type=float, default=0.1, help='dropout rate')
parser.add_argument('--known_class', action="store_true", default=False)
parser.add_argument('--shared_vocab', action="store_true", default=False)
parser.add_argument('--shared_encoder', action="store_true", default=False)
parser.add_argument('--data_dir', type=str, default='./data/uspto_50k_typed', help='base directory')
parser.add_argument('--exp_dir', type=str, default='./result/uspto_50k_untyped/rebuttal', help='result directory')
parser.add_argument('--checkpoint', type=str, help='checkpoint model file')
# delta tuning params
parser.add_argument('--ft_mode', type=str, default='petl', choices=["petl", "full", "none"])
parser.add_argument('--ffn_mode', type=str, default='none', choices=["none", "adapter"])
parser.add_argument('--ffn_option', type=str, default="none", choices=["parallel", "sequential", "none"])
parser.add_argument('--ffn_bn', type=int, default=256)
parser.add_argument('--ffn_adapter_scalar', type=str, default="1") # learnable or fixed
parser.add_argument('--ffn_adapter_init_option', type=str, default="lora", choices=["bert", "lora"])
parser.add_argument('--ffn_adapter_layernorm_option', type=str, default="none", choices=["in", "out", "none"])
parser.add_argument('--attn_mode', type=str, default='none', choices=["none", "prefix", "lora", "adpater"])
parser.add_argument('--attn_bn', type=int, default=10)
parser.add_argument('--attn_dim', type=int, default=32)
parser.add_argument('--prompt', action="store_true", default=False)
parser.add_argument('--input_prompt_attn', action="store_true", default=False)
parser.add_argument('--proto_hierarchy', type=int, default=3, help='the number of hierarchy')
parser.add_argument('--proto_path', type=str, default='./result/ecreact')
parser.add_argument('--proto_version', type=str, default="top", choices=["bottom", "middle", "top", "hierarchy", "namerxn"])
parser.add_argument('--freeze_proto', action="store_true", default=False)
args = parser.parse_args()
return args
def translate(iterator, model, dataset):
ground_truths_src = []
ground_truths_tgt = []
generations = []
invalid_token_indices = [dataset.tgt_stoi['<RX_{}>'.format(i)] for i in range(1, 11)]
invalid_token_indices += [dataset.tgt_stoi['<UNK>'], dataset.tgt_stoi['<unk>'], dataset.tgt_stoi['<mask>']]
invalid_token_indices += [dataset.tgt_stoi['<unused{}>'.format(i)] for i in range(1, 11)]
# Translate:
for batch in tqdm(iterator, total=len(iterator)):
src, tgt, _, _, _ = batch
#! Graph2Smiles: model.predict_step()
pred_tokens, pred_scores = translate_batch(model, batch, device=args.device, beam_size=args.beam_size,
invalid_token_indices=invalid_token_indices,
max_length=args.max_length, prompt=args.prompt)
for idx in range(batch[0].shape[1]): # batch_size
gt_src = ''.join(dataset.reconstruct_smi(src[:, idx], src=True))
gt_tgt = ''.join(dataset.reconstruct_smi(tgt[:, idx], src=False)) #! remove <pad>/<sos>/<eos>
# map id to tokens
hypos = np.array([''.join(dataset.reconstruct_smi(tokens, src=False)) for tokens in pred_tokens[idx]])
hypo_len = np.array([len(smi_tokenizer(ht)) for ht in hypos])
new_pred_score = copy.deepcopy(pred_scores[idx]).cpu().numpy() / hypo_len
ordering = np.argsort(new_pred_score)[::-1]
ground_truths_src.append(gt_src)
ground_truths_tgt.append(gt_tgt)
generations.append(hypos[ordering])
return ground_truths_src, ground_truths_tgt, generations
def main(args):
# Build Data Iterator:
train_iter, val_iter, dataset = build_retro_iterator(args, mode="train")
test_iter, _ = build_retro_iterator(args, mode="test")
# Load Checkpoint Model:
args.weight = torch.randn(13, 512)
model = build_model(args, dataset.src_itos, dataset.tgt_itos)
model = load_checkpoint_downstream(args.checkpoint, model)
model.to(args.device)
# Get Output Path:
exp_version = 'typed' if args.known_class == 'True' else 'untyped'
aug_version = '_augment' if 'augment' in args.checkpoint else ''
output_path = os.path.join(args.exp_dir,'bs_top{}_generation_{}{}.pk'.format(args.beam_size, exp_version, aug_version))
print('Output path: {}'.format(output_path))
# Begin Translating:
#! select train/val/test_iter
ground_truths_src, ground_truths_tgt, generations = translate(val_iter, model, dataset)
accuracy_matrix = np.zeros((len(ground_truths_tgt), args.beam_size))
for i in range(len(ground_truths_tgt)):
gt_i = canonical_smiles(ground_truths_tgt[i])
generation_i = [canonical_smiles(gen) for gen in generations[i]]
for j in range(args.beam_size):
if gt_i in generation_i[:j + 1]:
accuracy_matrix[i][j] = 1
with open(output_path, 'wb') as f:
pickle.dump((ground_truths_src, ground_truths_tgt, generations), f)
for j in range(args.beam_size):
logger.info('Top-{}: {}'.format(j + 1, round(np.mean(accuracy_matrix[:, j]), 4)))
return
def calculate_topk_acc(predicted_path):
with open(predicted_path, "rb") as f:
ground_truths, generations = pickle.load(f)
beam_size = len(generations[0])
# print(beam_size)
accuracy_matrix = np.zeros((len(ground_truths), beam_size))
for i in tqdm(range(len(ground_truths))):
gt_i = canonical_smiles(ground_truths[i])
generation_i = [canonical_smiles(gen) for gen in generations[i]]
for j in range(beam_size):
if gt_i in generation_i[:j + 1]:
accuracy_matrix[i][j] = 1
for j in range(beam_size):
print('Top-{}: {}'.format(j + 1, round(np.mean(accuracy_matrix[:, j]), 4)))
if __name__ == "__main__":
args = arg_parse()
dt = datetime.now()
args.exp_dir = os.path.join(args.exp_dir, '{}_{:02d}-{:02d}-{:02d}'.format(
dt.date(), dt.hour, dt.minute, dt.second))
os.makedirs(args.exp_dir, exist_ok=True)
args.data_file = None
args.shared_vocab = True
args.known_class = False
args.prompt = False
args.checkpoint = "./result/uspto_50k_untyped/rebuttal/model_39000_augment.pt"
logger = init_logger(os.path.join(args.exp_dir, "log_metrics.txt"))
with open(os.path.join(args.exp_dir, 'config_predict.json'), 'w') as f:
json.dump(vars(args), f, indent=4)
main(args)