Matthew Franglen commited on
Commit
59da9af
1 Parent(s): c986f65

Create an entrypoint and split code up

Browse files

The aste file that is produced by this matches the notebook output

Files changed (5) hide show
  1. src/alignment.py +165 -0
  2. src/convert.py +11 -0
  3. src/main.py +44 -0
  4. src/sentiment.py +9 -0
  5. src/types.py +23 -3
src/alignment.py ADDED
@@ -0,0 +1,165 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from dataclasses import asdict
2
+ from typing import Optional
3
+
4
+ import Levenshtein
5
+ import pandas as pd
6
+
7
+ from .types import CharacterIndices, Triplet, WordSpans
8
+
9
+
10
+ def find_closest_text(
11
+ *,
12
+ original: pd.Series,
13
+ replacement: pd.Series,
14
+ ) -> pd.Series:
15
+ # Returns a series of the replacement values aligned to the original values
16
+ no_space_replacements = {text.replace(" ", ""): text for text in replacement}
17
+ result = original.str.replace(" ", "").map(no_space_replacements)
18
+ non_perfect_matches = result.isna().sum()
19
+
20
+ assert non_perfect_matches / len(original) <= 0.05, (
21
+ "Poor alignment with replacement text. "
22
+ f"{non_perfect_matches:,} of {len(original),} rows did not match well"
23
+ )
24
+
25
+ def closest(text: str) -> str:
26
+ distances = replacement.apply(
27
+ lambda comparison: Levenshtein.distance(text, comparison)
28
+ )
29
+ return replacement.iloc[distances.argmin()]
30
+
31
+ result.loc[result.isna()] = result[result.isna()].apply(closest)
32
+ return result
33
+
34
+
35
+ def to_character_indices_series(row: pd.Series) -> pd.Series:
36
+ result = to_character_indices(triplet=row.triples, text=row.text)
37
+ return pd.Series(asdict(result))
38
+
39
+
40
+ def to_character_indices(
41
+ *,
42
+ triplet: Triplet,
43
+ text: str,
44
+ ) -> CharacterIndices:
45
+ aspect_span, opinion_span, _ = triplet
46
+ assert _is_sequential(aspect_span), f"aspect span not sequential: {aspect_span}"
47
+ assert _is_sequential(opinion_span), f"opinion span not sequential: {opinion_span}"
48
+
49
+ spans = WordSpans.make(text)
50
+
51
+ aspect_start_index, aspect_end_index = spans.to_indices(aspect_span)
52
+ aspect_term = text[aspect_start_index : aspect_end_index + 1]
53
+ opinion_start_index, opinion_end_index = spans.to_indices(opinion_span)
54
+ opinion_term = text[opinion_start_index : opinion_end_index + 1]
55
+
56
+ return CharacterIndices(
57
+ aspect_start_index=aspect_start_index,
58
+ aspect_end_index=aspect_end_index,
59
+ aspect_term=aspect_term,
60
+ opinion_start_index=opinion_start_index,
61
+ opinion_end_index=opinion_end_index,
62
+ opinion_term=opinion_term,
63
+ )
64
+
65
+
66
+ def to_aligned_character_indices(
67
+ *,
68
+ original: str,
69
+ replacement: str,
70
+ original_indices: CharacterIndices,
71
+ ) -> CharacterIndices:
72
+ indices = _aligned_character_indices(original=original, replacement=replacement)
73
+
74
+ aspect_start_index = _aligned_start_index(
75
+ text=replacement,
76
+ original_index=original_indices.aspect_start_index,
77
+ indices=indices,
78
+ )
79
+ aspect_end_index = _aligned_end_index(
80
+ text=replacement,
81
+ original_index=original_indices.aspect_end_index,
82
+ indices=indices,
83
+ )
84
+ aspect_term = replacement[aspect_start_index : aspect_end_index + 1]
85
+
86
+ opinion_start_index = _aligned_start_index(
87
+ text=replacement,
88
+ original_index=original_indices.opinion_start_index,
89
+ indices=indices,
90
+ )
91
+ opinion_end_index = _aligned_end_index(
92
+ text=replacement,
93
+ original_index=original_indices.opinion_end_index,
94
+ indices=indices,
95
+ )
96
+ opinion_term = replacement[opinion_start_index : opinion_end_index + 1]
97
+
98
+ return CharacterIndices(
99
+ aspect_start_index=aspect_start_index,
100
+ aspect_end_index=aspect_end_index,
101
+ aspect_term=aspect_term,
102
+ opinion_start_index=opinion_start_index,
103
+ opinion_end_index=opinion_end_index,
104
+ opinion_term=opinion_term,
105
+ )
106
+
107
+
108
+ def _is_sequential(span: tuple[int, ...]) -> bool:
109
+ return all(span[index + 1] - span[index] == 1 for index in range(len(span) - 1))
110
+
111
+
112
+ def _aligned_character_indices(original: str, replacement: str) -> list[Optional[int]]:
113
+ indices: list[Optional[int]] = list(range(len(original)))
114
+ for operation, _source_position, destination_position in Levenshtein.editops(
115
+ original, replacement
116
+ ):
117
+ if operation == "replace":
118
+ indices[destination_position] = None
119
+ elif operation == "insert":
120
+ indices.insert(destination_position, None)
121
+ elif operation == "delete":
122
+ del indices[destination_position]
123
+ return indices
124
+
125
+
126
+ def _aligned_start_index(
127
+ text: str, original_index: int, indices: list[Optional[int]]
128
+ ) -> int:
129
+ closest_after = min(
130
+ index for index in indices if index is not None and index >= original_index
131
+ )
132
+ index = indices.index(closest_after)
133
+
134
+ # Not every character in the original text is aligned to a character in the
135
+ # replacement text. The replacement text may have deleted it, or replaced
136
+ # it. Can step back through each letter until the word boundary is found or
137
+ # an aligned character is found.
138
+ while index > 0:
139
+ if indices[index - 1] is not None:
140
+ break
141
+ if text[index - 1] == " ":
142
+ break
143
+ index -= 1
144
+ return index
145
+
146
+
147
+ def _aligned_end_index(
148
+ text: str, original_index: int, indices: list[Optional[int]]
149
+ ) -> int:
150
+ closest_before = min(
151
+ index for index in indices if index is not None and index <= original_index
152
+ )
153
+ index = indices.index(closest_before)
154
+
155
+ # Not every character in the original text is aligned to a character in the
156
+ # replacement text. The replacement text may have deleted it, or replaced
157
+ # it. Can step back through each letter until the word boundary is found or
158
+ # an aligned character is found.
159
+ while index < len(indices) - 1:
160
+ if indices[index + 1] is not None:
161
+ break
162
+ if text[index + 1] == " ":
163
+ break
164
+ index += 1
165
+ return index
src/convert.py CHANGED
@@ -263,3 +263,14 @@ def aste_to_character_indices(
263
  opinion_term=opinion_term,
264
  sentiment=nice_sentiment,
265
  )
 
 
 
 
 
 
 
 
 
 
 
 
263
  opinion_term=opinion_term,
264
  sentiment=nice_sentiment,
265
  )
266
+
267
+
268
+ label_to_sentiment = {
269
+ "POS": "positive",
270
+ "NEG": "negative",
271
+ "NEU": "neutral",
272
+ }
273
+
274
+
275
+ def to_nice_sentiment(label: str) -> str:
276
+ return label_to_sentiment[sentiment]
src/main.py ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pathlib import Path
2
+ from typing import Annotated
3
+
4
+ import typer
5
+
6
+ from .alignment import to_character_indices_series
7
+ from .data import read_aste_file
8
+ from .sentiment import to_nice_sentiment
9
+
10
+ app = typer.Typer()
11
+
12
+
13
+ @app.command()
14
+ def aste(
15
+ aste_file: Annotated[Path, typer.Option()],
16
+ output_file: Annotated[Path, typer.Option()],
17
+ ) -> None:
18
+ df = read_aste_file(aste_file)
19
+ df = df.explode("triples")
20
+ df = df.reset_index(drop=False)
21
+ df = df.merge(
22
+ df.apply(to_character_indices_series, axis="columns"),
23
+ left_index=True,
24
+ right_index=True,
25
+ )
26
+ df["sentiment"] = df.triples.apply(lambda triple: to_nice_sentiment(triple[2]))
27
+ df = df.drop(columns=["triples"])
28
+
29
+ print(df.sample(3))
30
+
31
+ df.to_parquet(output_file, compression="gzip")
32
+
33
+
34
+ @app.command()
35
+ def sem_eval(
36
+ aste_file: Annotated[Path, typer.Option()],
37
+ sem_eval_file: Annotated[Path, typer.Option()],
38
+ output_file: Annotated[Path, typer.Option()],
39
+ ) -> None:
40
+ pass
41
+
42
+
43
+ if __name__ == "__main__":
44
+ app()
src/sentiment.py ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ label_to_sentiment = {
2
+ "POS": "positive",
3
+ "NEG": "negative",
4
+ "NEU": "neutral",
5
+ }
6
+
7
+
8
+ def to_nice_sentiment(label: str) -> str:
9
+ return label_to_sentiment[label]
src/types.py CHANGED
@@ -2,6 +2,12 @@ from __future__ import annotations
2
 
3
  import re
4
  from dataclasses import dataclass
 
 
 
 
 
 
5
 
6
  word_pattern = re.compile(r"\S+")
7
 
@@ -11,12 +17,27 @@ class WordSpan:
11
  start_index: int
12
  end_index: int # this is the letter after the end
13
 
 
 
 
 
 
14
  @staticmethod
15
- def to_spans(text: str) -> list[WordSpan]:
16
- return [
17
  WordSpan(start_index=match.start(), end_index=match.end())
18
  for match in word_pattern.finditer(text)
19
  ]
 
 
 
 
 
 
 
 
 
 
20
 
21
 
22
  @dataclass(frozen=True)
@@ -27,4 +48,3 @@ class CharacterIndices:
27
  opinion_start_index: int
28
  opinion_end_index: int
29
  opinion_term: str
30
- sentiment: str
 
2
 
3
  import re
4
  from dataclasses import dataclass
5
+ from typing import Literal
6
+
7
+ AspectWordIndices = tuple[int, ...]
8
+ OpinionWordIndices = tuple[int, ...]
9
+ Sentiment = Literal["NEG", "NEU", "POS"]
10
+ Triplet = tuple[AspectWordIndices, OpinionWordIndices, Sentiment]
11
 
12
  word_pattern = re.compile(r"\S+")
13
 
 
17
  start_index: int
18
  end_index: int # this is the letter after the end
19
 
20
+
21
+ @dataclass
22
+ class WordSpans:
23
+ spans: list[WordSpan]
24
+
25
  @staticmethod
26
+ def make(text: str) -> WordSpans:
27
+ spans = [
28
  WordSpan(start_index=match.start(), end_index=match.end())
29
  for match in word_pattern.finditer(text)
30
  ]
31
+ return WordSpans(spans)
32
+
33
+ def to_indices(self, span: tuple[int, ...]) -> tuple[int, int]:
34
+ word_start = span[0]
35
+ word_start_span = self.spans[word_start]
36
+
37
+ word_end = span[-1]
38
+ word_end_span = self.spans[word_end]
39
+
40
+ return word_start_span.start_index, word_end_span.end_index - 1
41
 
42
 
43
  @dataclass(frozen=True)
 
48
  opinion_start_index: int
49
  opinion_end_index: int
50
  opinion_term: str