File size: 13,163 Bytes
66242b8 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 | from __future__ import annotations
from dataclasses import dataclass
from typing import Any
import pandas as pd
from tqdm.auto import tqdm
from function_words import FUNCTION_WORDS
FUNCTION_WORD_SET = {word.lower() for word in FUNCTION_WORDS} # lowercase function words
PLACEHOLDER_RE = r"^<[^<>]+>$" # for masking placeholder
@dataclass(slots=True)
class Config:
verbose: bool = True
include_function_word_rate: bool = True
exclude_placeholders_from_avg_word_length: bool = True
phrase_role_dependency_labels: tuple[str, ...] = ("acl", "advcl", "ccomp", "pcomp", "relcl", "xcomp") # for clausal phrase signals
pos_roles: dict[str, tuple[str, ...]] = None
dep_roles: dict[str, tuple[str, ...]] = None
# covering tags for en_core_web_lg spacy model
def __post_init__(self) -> None:
if self.pos_roles is None:
self.pos_roles = {
"adjective": ("ADJ",),
"adposition": ("ADP",),
"adverb": ("ADV",),
"auxiliary": ("AUX",),
"conjunction": ("CONJ",),
"coordinating_conjunction": ("CCONJ",),
"determiner": ("DET",),
"interjection": ("INTJ",),
"noun": ("NOUN",),
"numeral": ("NUM",),
"particle": ("PART",),
"pronoun": ("PRON",),
"proper_noun": ("PROPN",),
"punctuation": ("PUNCT",),
"subordinating_conjunction": ("SCONJ",),
"symbol": ("SYM",),
"verb": ("VERB",),
"other": ("X",),
"space": ("SPACE",),
}
if self.dep_roles is None:
self.dep_roles = {
"root": ("ROOT",),
"adjectival_clause": ("acl",),
"adjectival_complement": ("acomp",),
"adverbial_clause": ("advcl",),
"adverbial_modifier": ("advmod",),
"agent": ("agent",),
"adjectival_modifier": ("amod",),
"apposition": ("appos",),
"attribute": ("attr",),
"auxiliary": ("aux",),
"passive_auxiliary": ("auxpass",),
"case_marker": ("case",),
"coordinating_conjunction": ("cc",),
"clausal_complement": ("ccomp",),
"compound": ("compound",),
"conjunct": ("conj",),
"clausal_subject": ("csubj",),
"passive_clausal_subject": ("csubjpass",),
"dative": ("dative",),
"dependency_unspecified": ("dep",),
"determiner": ("det",),
"direct_object": ("dobj",),
"expletive": ("expl",),
"indirect_object": ("iobj",),
"interjection": ("intj",),
"marker": ("mark",),
"meta": ("meta",),
"negation": ("neg",),
"nominal_modifier": ("nmod",),
"noun_phrase_adverbial_modifier": ("npadvmod",),
"nominal_subject": ("nsubj",),
"passive_nominal_subject": ("nsubjpass",),
"numeric_modifier": ("nummod",),
"object": ("obj",),
"object_predicate": ("oprd",),
"parataxis": ("parataxis",),
"prepositional_complement": ("pcomp",),
"object_of_preposition": ("pobj",),
"possessive_modifier": ("poss",),
"preconjunct": ("preconj",),
"predeterminer": ("predet",),
"prepositional_modifier": ("prep",),
"particle": ("prt",),
"punctuation": ("punct",),
"quantifier_modifier": ("quantmod",),
"relative_clause_modifier": ("relcl",),
"open_clausal_complement": ("xcomp",),
}
config = Config()
def _safe_mean(values: list[int]) -> float:
if not values:
return 0.0
return round(sum(values) / len(values), 3)
def _safe_rate(count: int, total: int) -> float:
if total == 0:
return 0.0
return round(count / total, 3)
# flagging non-word tokens including both punctuations and white-space
def _word_token_indices(record: dict[str, Any]) -> list[int]:
return [
index for index, (is_punct, is_space) in enumerate(zip(record["token_is_punct"], record["token_is_space"], strict=False))
if not is_punct and not is_space
]
# count average number of characters per word
def _avg_word_length(record: dict[str, Any], config: Config = config) -> float:
lengths: list[int] = []
for index, token_text in enumerate(record["tokens"]):
if record["token_is_punct"][index] or record["token_is_space"][index]: # excluding punctuations and white spaces
continue
if config.exclude_placeholders_from_avg_word_length and (token_text.startswith("<") and token_text.endswith(">")): # excluding <...> placeholder from previous masking
continue
lengths.append(len(token_text))
return _safe_mean(lengths)
# ========== SENTENCE STATISTICS ============
# defining a sentence
def _sentence_spans(record: dict[str, Any]) -> list[tuple[int, int]]:
spans = record["sentence_token_spans"]
# either compute from sentence_token_spans or number of tokens
if spans: return spans
if record["tokens"]: return [(0, len(record["tokens"]))]
return [] # else no sentence
def _sentence_word_lengths(record: dict[str, Any]) -> list[int]:
word_indices = set(_word_token_indices(record))
sentence_lengths: list[int] = []
for start, end in _sentence_spans(record):
count = sum(1 for index in range(start, end) if index in word_indices)
sentence_lengths.append(count)
return sentence_lengths
def _sentence_function_word_counts(record: dict[str, Any]) -> list[int]:
sentence_counts: list[int] = []
for start, end in _sentence_spans(record):
count = 0
for index in range(start, end):
if record["token_is_punct"][index] or record["token_is_space"][index]:
continue
if record["token_lower"][index] in FUNCTION_WORD_SET:
count += 1
sentence_counts.append(count)
return sentence_counts
# ============ PHRASE STATISTICS ==============
def _phrase_role_features(record: dict[str, Any], config: Config = config) -> dict[str, float]:
noun_phrase_count = len(record["noun_chunk_spans"]) # count noun phrases
dependency_labels = record["token_dep"]
prepositional_phrase_count = sum(1 for label in dependency_labels if label == "prep") # count preprositional phrases (approximated from "prep" label)
clausal_phrase_count = sum(
1 for label in dependency_labels if label in config.phrase_role_dependency_labels
) # count clausal phrases
phrase_counts = {
"phrase_noun_phrase_rate": noun_phrase_count,
"phrase_prepositional_phrase_rate": prepositional_phrase_count,
"phrase_clausal_phrase_rate": clausal_phrase_count,
}
total_phrase_units = sum(phrase_counts.values())
# returning the rate for each phrase group
return {
feature_name: _safe_rate(count, total_phrase_units)
for feature_name, count in phrase_counts.items()
}
# ============ POS TAGS STATISTICS ==============
def _pos_role_features(record: dict[str, Any], config: Config = config) -> dict[str, float]:
word_indices = _word_token_indices(record)
pos_counts = {name: 0 for name in config.pos_roles}
for index in word_indices:
token_pos = record["token_pos"][index]
for role_name, labels in config.pos_roles.items():
if token_pos in labels:
pos_counts[role_name] += 1
total_pos_units = sum(pos_counts.values())
return {
f"pos_{role_name}_rate": _safe_rate(count, total_pos_units)
for role_name, count in pos_counts.items()
}
# ============ DEPENDENCY LABEL STATISTICS ==============
def _dep_role_features(record: dict[str, Any], config: Config = config) -> dict[str, float]:
word_indices = _word_token_indices(record)
dep_counts = {name: 0 for name in config.dep_roles}
for index in word_indices:
token_dep = record["token_dep"][index]
for role_name, labels in config.dep_roles.items():
if token_dep in labels:
dep_counts[role_name] += 1
total_dep_units = sum(dep_counts.values())
return {
f"dep_{role_name}_rate": _safe_rate(count, total_dep_units)
for role_name, count in dep_counts.items()
}
# ======================================================
def extract_document_statistics(record: dict[str, Any], config: Config = config) -> dict[str, float]:
word_indices = _word_token_indices(record) # flag non-word tokens including both punctuations and white-space
total_word_tokens = len(word_indices) # number of word tokens
total_non_space_tokens = sum(1 for is_space in record["token_is_space"] if not is_space) # flag non-word tokens for only white-space
total_punct_tokens = sum(1 for is_punct in record["token_is_punct"] if is_punct) # count punctuations
total_function_words = sum(1 for index in word_indices if record["token_lower"][index] in FUNCTION_WORD_SET)
# count sentence statistics
sentence_lengths = _sentence_word_lengths(record)
sentence_function_word_counts = _sentence_function_word_counts(record)
features: dict[str, float] = {
"avg_sentence_length_words": _safe_mean(sentence_lengths),
"avg_function_words_per_sentence": _safe_mean(sentence_function_word_counts),
"punctuation_rate": _safe_rate(total_punct_tokens, total_non_space_tokens),
"avg_word_length": _avg_word_length(record, config=config),
}
if config.include_function_word_rate:
features["function_word_rate"] = _safe_rate(total_function_words, total_word_tokens)
features.update(_phrase_role_features(record, config=config)) # compute rate of phrases (from three groups)
features.update(_pos_role_features(record, config=config)) # compute rate for POS tags
features.update(_dep_role_features(record, config=config)) # compute rate for dependency features
return features
def extract_split_statistics(
df: pd.DataFrame,
split_cache: dict[str, list[dict[str, Any]]],
split_name: str = "",
config: Config = config,
) -> pd.DataFrame:
"""
Append statistical features for each configured text column in one split.
"""
result = df.copy()
for column in ["text1", "text2"]:
records = split_cache[column] # linguistic_cache must contain "text1" and "text2"
iterator = records
iterator = tqdm(
records,
total=len(records),
desc=f"Stat features [{split_name}:{column}]",
)
feature_rows = [extract_document_statistics(record, config=config) for record in iterator] # loop over rows
feature_df = pd.DataFrame(feature_rows).add_prefix(f"{column}_") # make each feature as one column
result = pd.concat([result.reset_index(drop=True), feature_df.reset_index(drop=True)], axis=1)
return result
def build_feature_summary(
dict_df: dict[str, pd.DataFrame],
config: Config = config,
) -> pd.DataFrame:
rows: list[dict[str, Any]] = []
summary_columns = [
"avg_sentence_length_words",
"avg_function_words_per_sentence",
"function_word_rate",
"punctuation_rate",
"avg_word_length",
]
for split, df in dict_df.items():
row: dict[str, Any] = {"split": split, "num_rows": len(df)}
for column in ["text1", "text2"]:
for feature_name in summary_columns:
prefixed_name = f"{column}_{feature_name}"
if prefixed_name in df.columns:
row[f"{prefixed_name}_mean"] = round(df[prefixed_name].mean(), 6)
rows.append(row)
return pd.DataFrame(rows)
def statistical_features_wrapper(
dict_df: dict[str, pd.DataFrame],
linguistic_cache: dict[str, dict[str, list[dict[str, Any]]]],
config: Config = config,
) -> tuple[dict[str, pd.DataFrame], pd.DataFrame]:
if config.verbose:
print("======= STATISTICAL FEATURES START =======")
statistical_dict_df: dict[str, pd.DataFrame] = {}
for split, df in dict_df.items():
if config.verbose:
print(f"\nProcessing statistical features for split='{split}' ({len(df):,} rows)")
statistical_dict_df[split] = extract_split_statistics(
df,
split_cache=linguistic_cache[split],
split_name=split,
config=config,
)
statistical_summary_df = build_feature_summary(statistical_dict_df, config=config)
if config.verbose:
print("\nStatistical feature summary:")
print(statistical_summary_df)
print("")
print("======= STATISTICAL FEATURES END =======")
print("")
return statistical_dict_df, statistical_summary_df
|