Upload filter.py
Browse files
filter.py
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import Levenshtein
|
3 |
+
|
4 |
+
# CSV ファイルのパス
|
5 |
+
file_path = r"C:\Users\user\Desktop\git\nekopara.csv"
|
6 |
+
|
7 |
+
# CSV 読み込み
|
8 |
+
df = pd.read_csv(file_path)
|
9 |
+
|
10 |
+
# 正解と文字起こしの列名 (適宜変更)
|
11 |
+
correct_column = "True"
|
12 |
+
transcribed_column = "ASR"
|
13 |
+
|
14 |
+
# 編集距離を計算して新しい列に追加
|
15 |
+
df["EditDistance"] = df.apply(
|
16 |
+
lambda row: Levenshtein.distance(
|
17 |
+
str(row[correct_column]), str(row[transcribed_column])
|
18 |
+
),
|
19 |
+
axis=1,
|
20 |
+
)
|
21 |
+
|
22 |
+
# 文字数の差を計算して新しい列に追加
|
23 |
+
df["LengthDifference"] = df.apply(
|
24 |
+
lambda row: abs(len(str(row[correct_column])) - len(str(row[transcribed_column]))),
|
25 |
+
axis=1,
|
26 |
+
)
|
27 |
+
|
28 |
+
# LengthDifference > 10 のみをフィルタリング
|
29 |
+
df_filtered = df[df["LengthDifference"] < 10]
|
30 |
+
df_filtered = df[df["EditDistance"] > 3]
|
31 |
+
|
32 |
+
# 編集距離と文字数の差で降順にソート
|
33 |
+
df_sorted = df_filtered.sort_values(
|
34 |
+
by=["EditDistance", "LengthDifference"], ascending=[False, False]
|
35 |
+
)
|
36 |
+
|
37 |
+
# 結果の表示
|
38 |
+
print(df_sorted)
|
39 |
+
|
40 |
+
# 結果を CSV に保存する場合 (オプション)
|
41 |
+
output_path = r"C:\Users\user\Desktop\git\nekopara_sorted.csv"
|
42 |
+
df_sorted.to_csv(output_path, index=False)
|