-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathoptimize.py
202 lines (160 loc) · 7.54 KB
/
optimize.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
import argparse
import yaml
import importlib
import pandas as pd
import os
import torch
import json
from datetime import datetime
from lightning import seed_everything
from omegaconf import DictConfig, OmegaConf
from groot.common import DATA_DIR
from groot.common.utils import parse_module_name_from_path
from groot.eval import EvalRunner
from groot.models import CNNVAE
from groot.models.modules.predictor import DropoutPredictor
from groot.optimizers.optimizer import OptimizerInterface
def parse_args():
parser = argparse.ArgumentParser(description="Optimize sequences.")
parser.add_argument("config_file", type=str, help="Path to config module")
parser.add_argument("--model_ckpt_path",
type=str,
required=True,
help="Checkpoint of model.")
parser.add_argument("--dataset", type=str, choices=["AAV", "GFP"], required=True)
parser.add_argument("--level",
type=str,
choices=["easy", "medium", "hard", "harder1", "harder2", "harder3"],
required=True)
parser.add_argument("--optim_config_path", type=str, required=True)
parser.add_argument("--output_dir", type=str, required=True)
parser.add_argument("--batch_size", type=int, default=128)
parser.add_argument("--devices",
type=str,
default="-1",
help="Training devices separated by comma.")
parser.add_argument("--changes",
type=str,
nargs="+",
help="Specify any changes to the yaml file (optim only).")
args = parser.parse_args()
return args
def initialize_models(dec_type: str, model_ckpt: str, device: torch.device):
module = CNNVAE.load_from_checkpoint(model_ckpt,
map_location=device,
device=device)
module.eval()
return module
def sample_pool_sequences(
csv_file: str,
frac: float = None,
strategy: str = "random",
seed: int = 0
):
df = pd.read_csv(csv_file)
if strategy == "random":
pool_df = df.sample(frac=frac, random_state=seed)
elif strategy == "bottom":
n = int(frac * len(df))
pool_df = df.nsmallest(n, columns="target")
elif strategy == "quantile":
quantiles = [0.0, 0.2, 0.4, 0.6, 0.8, 1.0]
dfs = []
for i in range(len(quantiles) - 1):
lower_val = df.target.quantile(quantiles[i])
upper_val = df.target.quantile(quantiles[i + 1])
filtered_df = df[df.target.between(lower_val, upper_val)]
dfs.append(filtered_df)
pool_df = pd.concat(dfs, ignore_index=True)
return pool_df, df
def alter_argument(changes, config: DictConfig):
for change in changes:
conf, val = change.split("=")
subconf = conf.split(".")
if len(subconf) == 1:
if isinstance(config[subconf[0]], bool):
val = int(val)
config[subconf[0]] = type(config[subconf[0]])(val)
elif len(subconf) == 2:
if isinstance(config[subconf[0]][subconf[1]], bool):
val = int(val)
config[subconf[0]][subconf[1]] = type(config[subconf[0]][subconf[1]])(val)
else:
raise ValueError("Not valid alternatives.")
return config
def get_csv_files(rootdir: str, dataset: str, difficulty: str):
if difficulty == "easy":
csv_file = f"{rootdir}/{dataset}/mutations_0/percentile_0.5_0.6/base_seqs.csv"
elif difficulty == "medium":
csv_file = f"{rootdir}/{dataset}/mutations_6/percentile_0.2_0.4/base_seqs.csv"
elif difficulty == "hard":
csv_file = f"{rootdir}/{dataset}/mutations_7/percentile_0.0_0.3/base_seqs.csv"
else:
no_mut = 13 if dataset == "AAV" else "8"
if difficulty == "harder1":
csv_file = f"{rootdir}/{dataset}/mutations_{no_mut}/percentile_0.0_0.3/base_seqs.csv"
elif difficulty == "harder2":
csv_file = f"{rootdir}/{dataset}/mutations_{no_mut}/percentile_0.0_0.2/base_seqs.csv"
elif difficulty == "harder3":
csv_file = f"{rootdir}/{dataset}/mutations_{no_mut}/percentile_0.0_0.1/base_seqs.csv"
return csv_file
def init_everything(args):
curr_time = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
# Load configs
cfg = importlib.import_module(parse_module_name_from_path(
args.config_file))
with open(args.optim_config_path, "r") as file_:
optim_config = yaml.safe_load(file_)
optim_config = DictConfig(optim_config)
if len(args.changes) != 0:
optim_config = alter_argument(args.changes, optim_config)
with open(f"scripts/configs/evaluate/{args.dataset}.yaml", "r") as file_:
eval_config = yaml.safe_load(file_)
eval_config = DictConfig(eval_config)
print("Finish loading configurations.")
# Get data file
csv_file = get_csv_files(DATA_DIR, args.dataset, args.level)
torch.set_float32_matmul_precision(cfg.precision)
device = torch.device("cpu" if args.devices == "-1" else f"cuda:{args.devices}")
module = initialize_models(cfg.decoder_type, args.model_ckpt_path, device)
return curr_time, device, module, optim_config, eval_config, csv_file
def main(args, seed, curr_time, device, module, optim_config, eval_config, csv_file):
seed_everything(seed)
pool_df, df = sample_pool_sequences(csv_file,
optim_config.pool_frac,
optim_config.sample_strategy,
seed)
save_path = f"{args.output_dir}/optim_log/{args.dataset}/{args.level}/base_pool_{optim_config.pool_frac}_{seed}.csv"
os.makedirs(os.path.dirname(save_path), exist_ok=True)
pool_df.to_csv(save_path, index=False)
pool_seqs = pool_df.sequence.tolist()
predictor = DropoutPredictor(module.hparams.latent_dim, optim_config.pred_hidden_dim)
optimizer = OptimizerInterface(module, predictor, optim_config)
if optim_config.smooth:
optimizer.smooth_func(pool_df, args.batch_size, device)
optimized_seqs, pred_scores = optimizer.optimize(pool_seqs)
eval_config.base_pool_path = save_path
evaluator = EvalRunner(eval_config)
prefix = "smoothed" if optim_config.smooth else "unsmoothed"
results_df, metrics_df = evaluator.evaluate_sequences(pred_scores, optimized_seqs)
save_path = f"{args.output_dir}/log_results/{args.dataset}/{args.level}/{prefix}_{optim_config.algo_name}/run_{curr_time}"
os.makedirs(save_path, exist_ok=True)
results_df.to_csv(os.path.join(save_path, f"sequences_{seed}.csv"), index=False)
metrics_df.to_csv(os.path.join(save_path, f"metrics_{seed}.csv"), index=False)
print(results_df.head(5))
print(metrics_df)
return optim_config, eval_config, save_path
def merge_write_dictionaries(args, optim_config, eval_config, save_dir):
conf = vars(args)
conf.update(OmegaConf.to_container(optim_config, resolve=True))
conf.update(OmegaConf.to_container(eval_config, resolve=True))
with open(f"{save_dir}/config.json", "w") as f:
json.dump(dict(conf), f, indent=4)
if __name__ == "__main__":
args = parse_args()
curr_time, device, module, optim_config, eval_config, csv_file = init_everything(args)
for seed in range(5):
optim_config, eval_config, save_dir = main(
args, seed, curr_time, device, module, optim_config, eval_config, csv_file
)
merge_write_dictionaries(args, optim_config, eval_config, save_dir)