-
Notifications
You must be signed in to change notification settings - Fork 26
/
inference.py
124 lines (101 loc) · 5.32 KB
/
inference.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
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--model_dir', type=str, required=True)
parser.add_argument('--ckpt', type=str, required=True)
parser.add_argument('--splits', type=str, required=True)
parser.add_argument('--split_key', type=str, default=None)
parser.add_argument('--inf_mols', type=int, default=1000)
parser.add_argument('--wandb', type=str, default=None)
parser.add_argument('--num_workers', type=int, default=None)
parser.add_argument('--ode', action='store_true', default=False)
parser.add_argument('--elbo', action='store_true', default=False)
parser.add_argument('--alpha', type=float, default=0)
parser.add_argument('--beta', type=float, default=1)
parser.add_argument('--num_samples', type=int, default=1)
parser.add_argument('--inf_step', type=float, default=0.5)
parser.add_argument('--elbo_step', type=float, default=0.2)
parser.add_argument('--inf_type', type=str,
choices=['entropy', 'rate'], default='rate')
parser.add_argument('--max_len', type=int, default=1024)
parser.add_argument('--inf_Hf', type=float, default=None)
parser.add_argument('--inf_kmin', type=int, default=None)
parser.add_argument('--inf_tmin', type=int, default=None)
parser.add_argument('--inf_cutoff', type=int, default=None)
parser.add_argument('--embeddings_dir', type=str, default=None)
parser.add_argument('--pdb_dir', type=str, default=None)
parser.add_argument('--embeddings_key', type=str, default=None, choices=['name', 'reference'])
inf_args = parser.parse_args()
import os, yaml, torch, wandb
import numpy as np
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
from utils.logging import get_logger
logger = get_logger(__name__)
from utils.dataset import get_loader
import pandas as pd
from model import get_model
from utils.inference import inference_epoch
with open(f'{inf_args.model_dir}/args.yaml') as f:
args = argparse.Namespace(**yaml.full_load(f))
args.wandb = inf_args.wandb
if inf_args.wandb:
wandb.login(key=os.environ['WANDB_API_KEY'])
wandb.init(
entity=os.environ['WANDB_ENTITY'],
settings=wandb.Settings(start_method="fork"),
project=args.wandb,
name=str(args.time),
config=args.__dict__ | inf_args.__dict__
)
args.splits = inf_args.splits
args.inf_mols = inf_args.inf_mols
args.inf_type = inf_args.inf_type
args.num_samples = inf_args.num_samples
args.inf_step = inf_args.inf_step
args.max_len = inf_args.max_len
args.ode = inf_args.ode
args.alpha, args.beta = inf_args.alpha, inf_args.beta
args.elbo_step = inf_args.elbo_step
if inf_args.num_workers is not None:
args.num_workers = inf_args.num_workers
if inf_args.inf_Hf:
if inf_args.inf_Hf < args.train_Hf: logger.warning(f'Out of bounds: inf_Hf {inf_args.inf_Hf} < train_Hf {args.train_Hf}')
args.train_Hf = inf_args.inf_Hf
if inf_args.inf_kmin:
if inf_args.inf_kmin < args.train_kmin: logger.warning(f'Out of bounds: inf_kmin {inf_args.inf_kmin} < train_kmin {args.train_kmin}')
args.train_kmin = inf_args.inf_kmin
if inf_args.inf_tmin:
if inf_args.inf_tmin < args.train_tmin: logger.warning(f'Out of bounds: inf_tmin {inf_args.inf_tmin} < train_tmin {args.train_tmin}')
args.train_tmin = inf_args.inf_tmin
if inf_args.inf_cutoff:
if inf_args.inf_cutoff > args.train_cutoff: logger.warning(f'Out of bounds: inf_cutoff {inf_args.inf_cutoff} > train_cutoff {args.train_cutoff}')
if inf_args.pdb_dir: args.pdb_dir = inf_args.pdb_dir
if inf_args.embeddings_dir: args.embeddings_dir = inf_args.embeddings_dir
if inf_args.embeddings_key: args.embeddings_key = inf_args.embeddings_key
args.inference_mode = True
def main():
logger.info(f'Loading splits {args.splits}')
try: splits = pd.read_csv(args.splits).set_index('path')
except: splits = pd.read_csv(args.splits).set_index('name')
logger.info("Constructing model")
model = get_model(args).to(device)
ckpt = os.path.join(inf_args.model_dir, inf_args.ckpt)
logger.info(f'Loading weights from {ckpt}')
state_dict = torch.load(ckpt, map_location=torch.device('cpu'))
model.load_state_dict(state_dict['model'], strict=True)
ep = state_dict['epoch']
val_loader = get_loader(args, None, splits, mode=inf_args.split_key, shuffle=False)
samples, log = inference_epoch(args, model, val_loader.dataset, device=device, pdbs=True, elbo=inf_args.elbo)
means = {key: np.mean(log[key]) for key in log if key != 'path'}
logger.info(f"Inference epoch {ep}: len {len(log['rmsd'])} MEANS {means}")
inf_name = f"{args.splits.split('/')[-1]}.ep{ep}.num{args.num_samples}.step{args.inf_step}.alpha{args.alpha}.beta{args.beta}"
if inf_args.inf_step != inf_args.elbo_step: inf_name += f".elbo{args.elbo_step}"
csv_path = os.path.join(inf_args.model_dir, f'{inf_name}.csv')
pd.DataFrame(log).set_index('path').to_csv(csv_path)
logger.info(f"Saved inf csv {csv_path}")
if not os.path.exists(os.path.join(inf_args.model_dir, inf_name)): os.mkdir(os.path.join(inf_args.model_dir, inf_name))
for samp in samples:
samp.pdb.write(os.path.join(inf_args.model_dir, inf_name, samp.path.split('/')[-1] + f".{samp.copy}.anim.pdb"), reverse=True)
samp.pdb.clear().add(samp.Y).write(
os.path.join(inf_args.model_dir, inf_name, samp.path.split('/')[-1] + f".{samp.copy}.pdb"))
if __name__ == '__main__':
main()