Update demo/modules/init_model.py
Browse files- demo/modules/init_model.py +117 -117
demo/modules/init_model.py
CHANGED
@@ -1,118 +1,118 @@
|
|
1 |
-
import faiss
|
2 |
-
import numpy as np
|
3 |
-
import pandas as pd
|
4 |
-
import os
|
5 |
-
import yaml
|
6 |
-
import glob
|
7 |
-
|
8 |
-
from easydict import EasyDict
|
9 |
-
from utils.constants import sequence_level
|
10 |
-
from model.ProTrek.protrek_trimodal_model import ProTrekTrimodalModel
|
11 |
-
from tqdm import tqdm
|
12 |
-
|
13 |
-
|
14 |
-
def load_model():
|
15 |
-
model_config = {
|
16 |
-
"protein_config": glob.glob(f"{config.model_dir}/esm2_*")[0],
|
17 |
-
"text_config": f"{config.model_dir}/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext",
|
18 |
-
"structure_config": glob.glob(f"{config.model_dir}/foldseek_*")[0],
|
19 |
-
"load_protein_pretrained": False,
|
20 |
-
"load_text_pretrained": False,
|
21 |
-
"from_checkpoint": glob.glob(f"{config.model_dir}/*.pt")[0]
|
22 |
-
}
|
23 |
-
|
24 |
-
model = ProTrekTrimodalModel(**model_config)
|
25 |
-
model.eval()
|
26 |
-
return model
|
27 |
-
|
28 |
-
|
29 |
-
def load_faiss_index(index_path: str):
|
30 |
-
if config.faiss_config.IO_FLAG_MMAP:
|
31 |
-
index = faiss.read_index(index_path, faiss.IO_FLAG_MMAP)
|
32 |
-
else:
|
33 |
-
index = faiss.read_index(index_path)
|
34 |
-
|
35 |
-
index.metric_type = faiss.METRIC_INNER_PRODUCT
|
36 |
-
return index
|
37 |
-
|
38 |
-
|
39 |
-
def load_index():
|
40 |
-
all_index = {}
|
41 |
-
|
42 |
-
# Load protein sequence index
|
43 |
-
all_index["sequence"] = {}
|
44 |
-
for db in tqdm(config.sequence_index_dir, desc="Loading sequence index..."):
|
45 |
-
db_name = db["name"]
|
46 |
-
index_dir = db["index_dir"]
|
47 |
-
|
48 |
-
index_path = f"{index_dir}/sequence.index"
|
49 |
-
sequence_index = load_faiss_index(index_path)
|
50 |
-
|
51 |
-
id_path = f"{index_dir}/ids.tsv"
|
52 |
-
uniprot_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
53 |
-
|
54 |
-
all_index["sequence"][db_name] = {"index": sequence_index, "ids": uniprot_ids}
|
55 |
-
|
56 |
-
# Load protein structure index
|
57 |
-
print("Loading structure index...")
|
58 |
-
all_index["structure"] = {}
|
59 |
-
for db in tqdm(config.structure_index_dir, desc="Loading structure index..."):
|
60 |
-
db_name = db["name"]
|
61 |
-
index_dir = db["index_dir"]
|
62 |
-
|
63 |
-
index_path = f"{index_dir}/structure.index"
|
64 |
-
structure_index = load_faiss_index(index_path)
|
65 |
-
|
66 |
-
id_path = f"{index_dir}/ids.tsv"
|
67 |
-
uniprot_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
68 |
-
|
69 |
-
all_index["structure"][db_name] = {"index": structure_index, "ids": uniprot_ids}
|
70 |
-
|
71 |
-
# Load text index
|
72 |
-
all_index["text"] = {}
|
73 |
-
valid_subsections = {}
|
74 |
-
for db in tqdm(config.text_index_dir, desc="Loading text index..."):
|
75 |
-
db_name = db["name"]
|
76 |
-
index_dir = db["index_dir"]
|
77 |
-
all_index["text"][db_name] = {}
|
78 |
-
text_dir = f"{index_dir}/subsections"
|
79 |
-
|
80 |
-
# Remove "Taxonomic lineage" from sequence_level. This is a special case which we don't need to index.
|
81 |
-
valid_subsections[db_name] = set()
|
82 |
-
sequence_level.add("Global")
|
83 |
-
for subsection in tqdm(sequence_level):
|
84 |
-
index_path = f"{text_dir}/{subsection.replace(' ', '_')}.index"
|
85 |
-
if not os.path.exists(index_path):
|
86 |
-
continue
|
87 |
-
|
88 |
-
text_index = load_faiss_index(index_path)
|
89 |
-
|
90 |
-
id_path = f"{text_dir}/{subsection.replace(' ', '_')}_ids.tsv"
|
91 |
-
text_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
92 |
-
|
93 |
-
all_index["text"][db_name][subsection] = {"index": text_index, "ids": text_ids}
|
94 |
-
valid_subsections[db_name].add(subsection)
|
95 |
-
|
96 |
-
# Sort valid_subsections
|
97 |
-
for db_name in valid_subsections:
|
98 |
-
valid_subsections[db_name] = sorted(list(valid_subsections[db_name]))
|
99 |
-
|
100 |
-
return all_index, valid_subsections
|
101 |
-
|
102 |
-
|
103 |
-
# Load the config file
|
104 |
-
root_dir = __file__.rsplit("/", 3)[0]
|
105 |
-
config_path = f"{root_dir}/demo/config.yaml"
|
106 |
-
with open(config_path, 'r', encoding='utf-8') as r:
|
107 |
-
config = EasyDict(yaml.safe_load(r))
|
108 |
-
|
109 |
-
device = "cuda"
|
110 |
-
|
111 |
-
print("Loading model...")
|
112 |
-
model = load_model()
|
113 |
-
model.to(device)
|
114 |
-
|
115 |
-
all_index, valid_subsections = load_index()
|
116 |
-
print("Done...")
|
117 |
-
# model = None
|
118 |
# all_index, valid_subsections = {"text": {}, "sequence": {"UniRef50": None}, "structure": {"UniRef50": None}}, {}
|
|
|
1 |
+
import faiss
|
2 |
+
import numpy as np
|
3 |
+
import pandas as pd
|
4 |
+
import os
|
5 |
+
import yaml
|
6 |
+
import glob
|
7 |
+
|
8 |
+
from easydict import EasyDict
|
9 |
+
from utils.constants import sequence_level
|
10 |
+
from model.ProTrek.protrek_trimodal_model import ProTrekTrimodalModel
|
11 |
+
from tqdm import tqdm
|
12 |
+
|
13 |
+
|
14 |
+
def load_model():
|
15 |
+
model_config = {
|
16 |
+
"protein_config": glob.glob(f"{config.model_dir}/esm2_*")[0],
|
17 |
+
"text_config": f"{config.model_dir}/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext",
|
18 |
+
"structure_config": glob.glob(f"{config.model_dir}/foldseek_*")[0],
|
19 |
+
"load_protein_pretrained": False,
|
20 |
+
"load_text_pretrained": False,
|
21 |
+
"from_checkpoint": glob.glob(f"{config.model_dir}/*.pt")[0]
|
22 |
+
}
|
23 |
+
|
24 |
+
model = ProTrekTrimodalModel(**model_config)
|
25 |
+
model.eval()
|
26 |
+
return model
|
27 |
+
|
28 |
+
|
29 |
+
def load_faiss_index(index_path: str):
|
30 |
+
if config.faiss_config.IO_FLAG_MMAP:
|
31 |
+
index = faiss.read_index(index_path, faiss.IO_FLAG_MMAP)
|
32 |
+
else:
|
33 |
+
index = faiss.read_index(index_path)
|
34 |
+
|
35 |
+
index.metric_type = faiss.METRIC_INNER_PRODUCT
|
36 |
+
return index
|
37 |
+
|
38 |
+
|
39 |
+
def load_index():
|
40 |
+
all_index = {}
|
41 |
+
|
42 |
+
# Load protein sequence index
|
43 |
+
all_index["sequence"] = {}
|
44 |
+
for db in tqdm(config.sequence_index_dir, desc="Loading sequence index..."):
|
45 |
+
db_name = db["name"]
|
46 |
+
index_dir = db["index_dir"]
|
47 |
+
|
48 |
+
index_path = f"{index_dir}/sequence.index"
|
49 |
+
sequence_index = load_faiss_index(index_path)
|
50 |
+
|
51 |
+
id_path = f"{index_dir}/ids.tsv"
|
52 |
+
uniprot_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
53 |
+
|
54 |
+
all_index["sequence"][db_name] = {"index": sequence_index, "ids": uniprot_ids}
|
55 |
+
|
56 |
+
# Load protein structure index
|
57 |
+
print("Loading structure index...")
|
58 |
+
all_index["structure"] = {}
|
59 |
+
for db in tqdm(config.structure_index_dir, desc="Loading structure index..."):
|
60 |
+
db_name = db["name"]
|
61 |
+
index_dir = db["index_dir"]
|
62 |
+
|
63 |
+
index_path = f"{index_dir}/structure.index"
|
64 |
+
structure_index = load_faiss_index(index_path)
|
65 |
+
|
66 |
+
id_path = f"{index_dir}/ids.tsv"
|
67 |
+
uniprot_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
68 |
+
|
69 |
+
all_index["structure"][db_name] = {"index": structure_index, "ids": uniprot_ids}
|
70 |
+
|
71 |
+
# Load text index
|
72 |
+
all_index["text"] = {}
|
73 |
+
valid_subsections = {}
|
74 |
+
for db in tqdm(config.text_index_dir, desc="Loading text index..."):
|
75 |
+
db_name = db["name"]
|
76 |
+
index_dir = db["index_dir"]
|
77 |
+
all_index["text"][db_name] = {}
|
78 |
+
text_dir = f"{index_dir}/subsections"
|
79 |
+
|
80 |
+
# Remove "Taxonomic lineage" from sequence_level. This is a special case which we don't need to index.
|
81 |
+
valid_subsections[db_name] = set()
|
82 |
+
sequence_level.add("Global")
|
83 |
+
for subsection in tqdm(sequence_level):
|
84 |
+
index_path = f"{text_dir}/{subsection.replace(' ', '_')}.index"
|
85 |
+
if not os.path.exists(index_path):
|
86 |
+
continue
|
87 |
+
|
88 |
+
text_index = load_faiss_index(index_path)
|
89 |
+
|
90 |
+
id_path = f"{text_dir}/{subsection.replace(' ', '_')}_ids.tsv"
|
91 |
+
text_ids = pd.read_csv(id_path, sep="\t", header=None).values.flatten()
|
92 |
+
|
93 |
+
all_index["text"][db_name][subsection] = {"index": text_index, "ids": text_ids}
|
94 |
+
valid_subsections[db_name].add(subsection)
|
95 |
+
|
96 |
+
# Sort valid_subsections
|
97 |
+
for db_name in valid_subsections:
|
98 |
+
valid_subsections[db_name] = sorted(list(valid_subsections[db_name]))
|
99 |
+
|
100 |
+
return all_index, valid_subsections
|
101 |
+
|
102 |
+
|
103 |
+
# Load the config file
|
104 |
+
root_dir = __file__.rsplit("/", 3)[0]
|
105 |
+
config_path = f"{root_dir}/demo/config.yaml"
|
106 |
+
with open(config_path, 'r', encoding='utf-8') as r:
|
107 |
+
config = EasyDict(yaml.safe_load(r))
|
108 |
+
|
109 |
+
device = "cuda"
|
110 |
+
|
111 |
+
print("Loading model...")
|
112 |
+
model = load_model()
|
113 |
+
# model.to(device)
|
114 |
+
|
115 |
+
all_index, valid_subsections = load_index()
|
116 |
+
print("Done...")
|
117 |
+
# model = None
|
118 |
# all_index, valid_subsections = {"text": {}, "sequence": {"UniRef50": None}, "structure": {"UniRef50": None}}, {}
|