-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathrun_embeddings.py
169 lines (145 loc) · 6.82 KB
/
run_embeddings.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
#!/usr/bin/env python3
import argparse
import json
import logging
import os
from semb import run_semb
from taskemb import run_taskemb
from textemb import run_textemb
from utils import DEFAULT_TASK_MAP, RUN_CONFIG_DIR
from itrain import DatasetArguments, RunArguments
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
def _restore_path(adapter_map, task_name, dataset_args):
template = adapter_map["source_path_format"]
run_id = adapter_map["adapters"][task_name]
# get the actual manager name
if dataset_args.task_name:
manager_name = f"{dataset_args.dataset_name}_{dataset_args.task_name}"
else:
manager_name = dataset_args.dataset_name
# HACK: the actual path to the adapter may have different names
path = os.path.expanduser(template.format(task_name, run_id, manager_name))
if not os.path.exists(path):
path = os.path.expanduser(template.format(manager_name, run_id, manager_name))
if not os.path.exists(path):
path = os.path.expanduser(template.format(task_name, run_id, task_name))
return path
def _restore_target_path(adapter_map, task_name, dataset_args):
template = adapter_map["target_path_format"]
# get the actual manager name
if dataset_args.task_name:
manager_name = f"{dataset_args.dataset_name}_{dataset_args.task_name}_n{dataset_args.train_subset_size}"
else:
manager_name = f"{dataset_args.dataset_name}_n{dataset_args.train_subset_size}"
# HACK: the actual path to the adapter may have different names
path = os.path.expanduser(template.format(task_name, dataset_args.train_subset_size, manager_name))
if not os.path.exists(path):
path = os.path.expanduser(template.format(manager_name, dataset_args.train_subset_size, manager_name))
if not os.path.exists(path):
path = os.path.expanduser(template.format(task_name, dataset_args.train_subset_size, task_name))
return path
def run(args, load_dir, data_args, run_config, overwrite=False, seed=42):
logger.info(f"***** Compute measures for {load_dir} *****\n")
output_base = args.output_dir or ("full_output" if args.full_model else "output")
if args.textemb:
run_textemb(
{
"model_name": args.model_name or "roberta-base",
"fast_tokenizer": run_config["model"].get("use_fast_tokenizer", False),
"output_dir": os.path.join(output_base, "textemb"),
"overwrite": overwrite,
},
data_args=data_args,
seed=seed,
)
if args.semb:
model_name = args.model_name or "sentence-transformers/roberta-base-nli-stsb-mean-tokens"
run_semb(
{
"model_name": model_name,
"fast_tokenizer": run_config["model"].get("use_fast_tokenizer", False),
"output_dir": os.path.join(output_base, "semb", model_name.split("/")[-1]),
"overwrite": overwrite,
},
data_args=data_args,
seed=seed,
)
if args.taskemb:
run_args = RunArguments(**run_config["training"])
run_args.num_train_epochs = 1
task_args = {
"model_type": args.model_type,
"model_name": args.model_name or "roberta-base",
"fast_tokenizer": run_config["model"].get("use_fast_tokenizer", False),
"output_dir": os.path.join(output_base, "taskemb"),
"overwrite": overwrite,
}
if args.full_model:
task_args["model_dir"] = load_dir
else:
task_args["adapter_dir"] = load_dir
run_taskemb(
task_args,
data_args=data_args,
run_args=run_args,
seed=seed,
)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--task_map", type=str, default=DEFAULT_TASK_MAP)
parser.add_argument("--taskemb", action="store_true")
parser.add_argument("--textemb", action="store_true")
parser.add_argument("--semb", action="store_true")
parser.add_argument("-o", "--output_dir", default=None, type=str)
parser.add_argument("--overwrite", action="store_true")
group = parser.add_mutually_exclusive_group()
group.add_argument("--target_tasks", action="store_true")
group.add_argument("--source_tasks", action="store_true")
group.add_argument("--tasks", default="", type=lambda s: s.split(","))
parser.add_argument("--dataset_sizes", default="0", type=lambda s: [int(item) for item in s.split(",")])
parser.add_argument("--model_type", default="roberta", type=str)
parser.add_argument("--model_name", default=None, type=str)
parser.add_argument("--full_model", action="store_true")
parser.add_argument("--seed", default=42, type=int)
args = parser.parse_args()
with open(os.path.expanduser(args.task_map), "r") as f:
task_map = json.load(f)
# source tasks
if args.source_tasks:
for dataset in task_map["from"]:
dataset_cleaned = dataset.replace("-", "")
run_config_file = os.path.join(RUN_CONFIG_DIR, f"{dataset_cleaned}.json")
with open(run_config_file, "r") as f:
run_config = json.load(f)
for train_size in args.dataset_sizes:
run_config["dataset"]["train_subset_size"] = train_size
data_args = DatasetArguments(**run_config["dataset"])
adapter_dir = _restore_path(task_map, dataset, data_args)
run(args, adapter_dir, data_args, run_config, overwrite=args.overwrite, seed=args.seed)
# user-specified tasks
if args.tasks:
for task in args.tasks:
if not task:
continue
run_config_file = os.path.join(RUN_CONFIG_DIR, f"{task}.json")
with open(run_config_file, "r") as f:
run_config = json.load(f)
for train_size in args.dataset_sizes:
run_config["dataset"]["train_subset_size"] = train_size
data_args = DatasetArguments(**run_config["dataset"])
adapter_dir = _restore_path(task_map, task, data_args)
run(args, adapter_dir, data_args, run_config, overwrite=args.overwrite, seed=args.seed)
# target tasks
if args.target_tasks:
for dataset in task_map["to"]:
run_config_file = os.path.join(RUN_CONFIG_DIR, f"{dataset}.json")
with open(run_config_file, "r") as f:
run_config = json.load(f)
for train_size in args.dataset_sizes:
run_config["dataset"]["train_subset_size"] = train_size
data_args = DatasetArguments(**run_config["dataset"])
adapter_dir = _restore_target_path(task_map, dataset, data_args)
run(args, adapter_dir, data_args, run_config, overwrite=args.overwrite, seed=args.seed)
if __name__ == "__main__":
main()