Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Grammar 리팩토링 #78

Open
wants to merge 2 commits into
base: develop
Choose a base branch
from
Open

Grammar 리팩토링 #78

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f579155
fix: Grammar 모델 negative score 오류 수정
9ooDa Apr 3, 2024
f2fec41
refactor: 중복 오류 수정 및 output에 index 추가
9ooDa Jul 16, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 16 additions & 21 deletions Models/grammar_api.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,14 +1,13 @@
import json
import os
import json

import nltk
import torch
import uvicorn
from fastapi import FastAPI, Form
from gector.gector import GECToR, load_verb_dict, predict_verbose
from nltk.tokenize import sent_tokenize
from transformers import AutoTokenizer
from typing_extensions import Annotated

from gector.gector import GECToR, load_verb_dict, predict_verbose
from utils import gram_metrics, gram_out_json, gram_visualizer_json

app = FastAPI()
Expand All @@ -24,7 +23,6 @@
# }

model_path = "gotutiyan/gector-roberta-large-5k"
# input_path = os.path.join(gector_path, "input", "macominatya.json")

model = GECToR.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)
Expand All @@ -40,9 +38,7 @@ async def upload_json(
gector_path = "./gector"
verb_path = "./gector/data/verb-form-vocab.txt"

nltk.download("punkt")
print(text)
srcs = sent_tokenize(text)
srcs = gram_visualizer_json.process_input_text(text)
encode, decode = load_verb_dict(verb_path)

transcription = text
Expand Down Expand Up @@ -70,7 +66,8 @@ async def upload_json(
# Grammar Error Correction Process
final_corrected_sents, iteration_log = predict_verbose(**predict_args)
checker_data = gram_visualizer_json.visualizer_json(
iteration_log, final_corrected_sents
iteration_log=iteration_log,
out_sentence=final_corrected_sents
)

# dump visualized checker .json file
Expand All @@ -79,25 +76,23 @@ async def upload_json(
with open(check, "w", encoding="utf-8") as c:
json.dump(checker_data, c, indent="\t")

# Run this part if you want the metric json file
# metric = os.path.join(gector_path, "metric", "metric.json")
# with open(metric, "w", encoding="utf-8") as r:
# json.dump(metric_data, r, indent="\t")

# Final output
phase = "phase_2" # either "phase_1" or "phase_2"
score_type = "pwc" # "ec" or "psc" or "pwc"
error_count_type = "new" # "new" or "old"
out_path = os.path.join(gector_path, "real", f"grammar_{phase}.json")
score = gram_metrics.get_score(checker_data=checker_data, score_type=score_type)
token_path = os.path.join(gector_path, "data", "token_labels.txt")

score = gram_metrics.get_score(checker_data=checker_data, score_type=score_type,
error_count_type=error_count_type, token_path=token_path)

print(
gram_out_json.create_json(
phase=phase, out_path=out_path, score=score, check_data=checker_data
)
gram_out_json.create_json(phase=phase, out_path=out_path, score=score,
checker_data=checker_data, token_path=token_path)
)

return gram_out_json.create_json(
phase=phase, out_path=out_path, score=score, check_data=checker_data
)
return gram_out_json.create_json(phase=phase, out_path=out_path, score=score,
checker_data=checker_data, token_path=token_path)

except Exception as e:
return {"text": None, "status": f"Error: {str(e)}"}
Expand Down
97 changes: 68 additions & 29 deletions Models/utils/gram_metrics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,48 +1,80 @@
import string
from typing import Dict, Literal, get_args

from .gram_out_json import get_cleaned_token_list, get_scrs_tok
from gram_out_json import get_cleaned_token_list, get_scrs_tok, get_tag_grammar, get_phase_2_inner_data


# ec = error count, psc = per sentence count, pwc = per word count
_STYPES = Literal["ec", "psc", "pwc"]
_ETYPES = Literal["new", "old"]


def get_error_count(
checker_data: Dict,
error_count_type: _ETYPES = "new",
token_path: str = "../gector/data/token_labels.txt"
):
metric_data = {"main": {}}
ctl = get_cleaned_token_list()
for og_sent, inner_dict in checker_data.items():
if inner_dict["edited"] == True:
inner_data = get_scrs_tok(inner_dict, ctl)
metric_data["main"][og_sent] = inner_data
else:
metric_data["main"][og_sent] = {
"edited": False,
"corrected_sentence": inner_dict["fin_sentence"],
}

error_count = 0
for og in metric_data["main"]:
v = metric_data["main"][og]
if v["edited"] == True:
error_count += len(v["action_tag"])
ctl = get_cleaned_token_list(token_path)

# 중복 에러 없앤 버전
if error_count_type == "new":
og_list = list(checker_data.keys())
# corr_list = [v.get('fin_sentence') for _, v in checker_data.items()]
tag_grammar = get_tag_grammar(ctl)

for og_sent, inner_dict in checker_data.items():
gector_dict = get_scrs_tok(inner_dict, ctl)
sid = og_list.index(og_sent)
sent = og_sent
corr_sent = inner_dict["fin_sentence"]
edited = inner_dict["edited"]
ref_word_list = gector_dict["og_word"]
tag_list = gector_dict["full_tag"]
if edited == True:
inner = get_phase_2_inner_data(sid=sid, sent=sent, corr_sent=corr_sent, edited=edited, ref_word_list=ref_word_list,
tag_list=tag_list, tag_grammar=tag_grammar)
error_count += len(inner["ref_word"])

# 기존 버전
elif error_count_type == "old":
metric_data = {
"main": {}
}
for og_sent, inner_dict in checker_data.items():
if inner_dict["edited"] == True:
inner_data = get_scrs_tok(inner_dict, ctl)
metric_data["main"][og_sent] = inner_data
else:
metric_data["main"][og_sent] = {"edited": False, "corrected_sentence": inner_dict["fin_sentence"]}

for og in metric_data["main"]:
v = metric_data["main"][og]
if v["edited"] == True:
error_count += len(v["action_tag"])

return error_count


def get_error_rate_sen(
checker_data: Dict,
error_count_type: _ETYPES = "new",
token_path: str = "../gector/data/token_labels.txt"
):
og_list = list(checker_data.keys())
error_count = get_error_count(checker_data=checker_data)
error_count = get_error_count(checker_data=checker_data, error_count_type=error_count_type, token_path=token_path)
sentence_count = len(og_list)

return round(error_count / sentence_count, 2)


def get_error_rate_word(
checker_data: Dict,
error_count_type: _ETYPES = "new",
token_path: str = "../gector/data/token_labels.txt"
):
og_list = list(checker_data.keys())
error_count = get_error_count(checker_data=checker_data)
error_count = get_error_count(checker_data=checker_data, error_count_type=error_count_type, token_path=token_path)

word_count = 0
for sen in og_list:
Expand All @@ -54,18 +86,25 @@ def get_error_rate_word(
return round(result, 2)


# ec = error count, psc = per sentence count, pwc = per word count
_TYPES = Literal["ec", "psc", "pwc"]


def get_score(checker_data: Dict, score_type: _TYPES = "pwc"):
def get_score(
checker_data: Dict,
score_type: _STYPES = "pwc",
error_count_type: _ETYPES = "new",
token_path: str = "../gector/data/token_labels.txt"
):
# raise error if the score type is not in the _TYPE list
options = get_args(_TYPES)
assert score_type in options, f"'{score_type}' is not in {options}"
soptions = get_args(_STYPES)
eoptions = get_args(_ETYPES)
assert score_type in soptions, f"'{score_type}' is not in {soptions}"
assert error_count_type in eoptions, f"'{error_count_type}' is not in {eoptions}"

if score_type == "ec":
return get_error_count(checker_data=checker_data)
return get_error_count(checker_data=checker_data, error_count_type=error_count_type, token_path=token_path)
elif score_type == "psc":
return get_error_rate_sen(checker_data=checker_data)
return get_error_rate_sen(checker_data=checker_data, error_count_type=error_count_type, token_path=token_path)
elif score_type == "pwc":
return get_error_rate_word(checker_data=checker_data)
return get_error_rate_word(checker_data=checker_data, error_count_type=error_count_type, token_path=token_path)

if __name__ == "__main__":
token_path = "../gector/data/token_labels.txt"
get_cleaned_token_list(token_path)
Loading