-
Notifications
You must be signed in to change notification settings - Fork 0
/
neural_tagger.py
167 lines (129 loc) · 7.58 KB
/
neural_tagger.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
from argparse import Namespace
from json import dump
from os import mkdir, path
from random import seed
from typing import Any, Optional, TextIO, Union
from cltk.tokenizers.lat.lat import LatinWordTokenizer
from torch import cuda, load, manual_seed, save
from utils.cli.interfaces import add_common_optional_arguments, add_common_training_arguments, setup_parser_divisions
from utils.cli.messages import NeuralMessage
from utils.data.interface import compute_kneser_ney_estimation, define_data, define_vocabulary_structures, \
get_lemmatizer, ParallelismDataset
from utils.models.constants import EMBEDDINGS, ENCODERS
from utils.models.interface import build_model
from utils.training.helper_functions import define_file_args, define_training_loop_args
from utils.training.optimizers import define_optimizer_args
from utils.visualizations.visualizers import visualize_training_outputs
from utils.training.evaluation_loop import evaluate
from utils.training.training_loop import train
if __name__ == "__main__":
main_parser, train_parser, evaluate_parser = setup_parser_divisions()
for subparser in (train_parser, evaluate_parser):
subparser_required_group = subparser.add_argument_group("Required Arguments")
subparser_required_group.add_argument("embedding", type=str, choices=EMBEDDINGS, help=NeuralMessage.EMBEDDING)
subparser_required_group.add_argument("encoder", type=str, choices=ENCODERS, help=NeuralMessage.ENCODER)
subparser_common_group = subparser.add_argument_group("Common Optional Arguments")
add_common_optional_arguments(subparser_common_group)
training_specific_group = train_parser.add_argument_group("Training-Specific Arguments")
add_common_training_arguments(training_specific_group)
args: Namespace = main_parser.parse_args()
kwargs: dict[str, Any] = vars(args)
# We provide a random seed to make computations deterministic.
if args.print_style != "none":
print(f"Running with random seed {args.random_seed} ...", flush=True)
seed(args.random_seed)
manual_seed(args.random_seed)
# We get the appropriate device.
device: str = "cuda" if cuda.is_available() else "cpu"
# Preliminary steps ...
if args.mode == "evaluate" and (args.model_location is None or not path.exists(args.model_location)):
raise ValueError("A valid directory is required for a model in order to evaluate it. Please try again.")
elif args.model_location is not None and path.exists(args.model_location) and not path.isdir(args.model_location):
raise ValueError("The given filepath for results exists, but it is not a valid directory. Please try again.")
required_partitions: list[str] = [args.evaluation_partition]
if args.mode == "train":
required_partitions.insert(0, args.training_partition)
dataset_directory, dataset_loader = args.dataset
tagging_kwargs: dict[str, Union[int, str]] = {
"link": args.link,
"stratum_count": args.stratum_count,
"tagset": args.tagset
}
loading_kwargs: dict[str, Any] = {
"collection_format": args.collection_format,
"tagging_kwargs": tagging_kwargs,
"tokenizer": LatinWordTokenizer()
}
current_dataset: ParallelismDataset = \
define_data(dataset_directory, dataset_loader, args.data_splits, required_partitions, loading_kwargs)
evaluation_kwargs: dict[str, Any] = {
"evaluation_partition": args.evaluation_partition,
"print_style": args.print_style,
"result_display_count": args.result_display_count,
"scoring_mode": args.scoring_mode,
"tagging_kwargs": tagging_kwargs
}
file_args: dict[str, Union[str, TextIO, None]] = define_file_args(kwargs)
if args.mode == "train":
if args.print_style != "none":
print("Setting up training...")
# We define the optimizer's arguments.
optimizer_args: dict[str, Any] = define_optimizer_args(kwargs["optimizer"], kwargs)
# We define the arguments for the training loop.
training_loop_args: dict[str, Any] = define_training_loop_args(kwargs)
vocabulary_kwargs: dict[str, Any] = {"embedding_filepath": args.embedding_filepath}
lemmatizer: Optional = get_lemmatizer(args.lemmatization)
vocabularies: dict[str, dict] = define_vocabulary_structures(current_dataset, lemmatizer, **vocabulary_kwargs)
kwargs["lemmatizer"] = lemmatizer
if args.replacement_probability == "kneser-ney":
kwargs["replacement_probability"] = compute_kneser_ney_estimation(vocabularies)
components: dict[str, str] = {"blender": args.blender, "embedding": args.embedding, "encoder": args.encoder}
# We instantiate the model and define the location at which it will be saved.
model = build_model(components, vocabularies, **kwargs)
model.to(device)
# We train the model.
trained_model, training_outputs = train(
model, device, current_dataset, kwargs["optimizer"], optimizer_args, training_loop_args,
evaluation_kwargs, file_args
)
best_results_display: str = training_outputs["best_scoring_structure"].get_statistics_display()
best_results_output: str = f"Overall Training Results - Best Model Statistics:\n{best_results_display}\n"
if args.print_style != "none":
print(best_results_output)
if file_args["validation_file"] is not None:
file_args["validation_file"].write(best_results_output)
# We save the model deemed best by the training process.
if file_args["model_location"] is not None:
save(trained_model, file_args["model_location"])
model_output_file: TextIO = open(file_args["model_outputs_location"], encoding="utf-8", mode="w+")
del training_outputs["best_scoring_structure"]
dump(training_outputs, model_output_file, indent=1)
model_output_file.close()
# Finally, if there are any visualizations specified, we create and save them.
if len(args.visualize) > 0 and args.visualization_directory is not None:
visualization_directory_path: str = args.visualization_directory
if args.model_name is not None:
visualization_directory_path += f"/{args.model_name}"
if not path.exists(visualization_directory_path):
mkdir(visualization_directory_path, mode=711)
elif path.exists(visualization_directory_path) and not path.isdir(visualization_directory_path):
raise ValueError("Invalid path for saving visuals. Please try again.")
visualize_training_outputs(training_outputs, args.visualize, visualization_directory_path)
elif args.mode == "evaluate":
if args.print_style != "none":
print("Starting evaluation...")
model = load(file_args["model_location"], map_location=device)
evaluation_outputs: dict[str, Any] = evaluate(
model, device, current_dataset, evaluation_kwargs, file_args, "test_file", args.tqdm
)
overall_results_display: str = f"Overall Test Results:" \
f"\n\t* Precision: {evaluation_outputs['precision']}" \
f"\n\t* Recall: {evaluation_outputs['recall']}" \
f"\n\t* F1: {evaluation_outputs['f1']}"
if args.print_style != "none":
print(overall_results_display)
else:
raise ValueError(f"The mode <{args.mode}> is not supported.")
for (key, value) in file_args.items():
if isinstance(value, TextIO):
value.close()