Files
drc-ners-nlp/processing/ner/ner_data_builder.py
T

153 lines
5.6 KiB
Python

import ast
import json
import logging
from pathlib import Path
import pandas as pd
import spacy
from spacy.tokens import DocBin
from spacy.util import filter_spans
from core.config import PipelineConfig
from core.utils import get_data_file_path
from core.utils.data_loader import DataLoader
class NERDataBuilder:
def __init__(self, config: PipelineConfig):
self.config = config
self.data_loader = DataLoader(config)
@staticmethod
def _parse_entities(series: pd.Series) -> pd.Series:
"""Vectorized parse of entity strings."""
def _parse(entities_str):
if not entities_str or entities_str in ["[]", "", "nan"]:
return []
entities_str = str(entities_str).strip()
try:
if entities_str.startswith("[(") and entities_str.endswith(")]"):
return ast.literal_eval(entities_str)
elif entities_str.startswith("[[") and entities_str.endswith("]]"):
return [tuple(e) for e in ast.literal_eval(entities_str)]
elif entities_str.startswith("[{") and entities_str.endswith("}]"):
return [(e["start"], e["end"], e["label"]) for e in json.loads(entities_str)]
else:
parsed = ast.literal_eval(entities_str)
return [
tuple(e) for e in parsed if isinstance(e, (list, tuple)) and len(e) == 3
]
except (ValueError, SyntaxError, json.JSONDecodeError):
return []
return series.map(_parse)
@staticmethod
def _validate_entities(texts: pd.Series, entities_series: pd.Series) -> pd.Series:
"""Vectorized entity validation."""
def _validate(text, entities):
if not entities or not text:
return []
text = str(text).strip()
valid = []
for ent in entities:
if not isinstance(ent, (list, tuple)) or len(ent) != 3:
continue
start, end, label = ent
try:
start, end = int(start), int(end)
except (ValueError, TypeError):
continue
if not isinstance(label, str):
continue
if not (0 <= start < end <= len(text)):
continue
if not text[start:end].strip():
continue
valid.append((start, end, label))
if not valid:
return []
valid.sort(key=lambda x: (x[0], x[1]))
# remove overlaps
filtered, last_end = [], -1
for s, e, l in valid:
if s >= last_end:
filtered.append((s, e, l))
last_end = e
return filtered
return pd.Series(map(_validate, texts, entities_series), index=texts.index)
@staticmethod
def _create_docs(nlp, texts, entities):
"""Batch create spaCy Docs."""
docs = []
for text, ents in zip(texts, entities):
doc = nlp(text)
spans = []
for start, end, label in ents:
span = doc.char_span(
start, end, label=label, alignment_mode="contract"
) or doc.char_span(start, end, label=label, alignment_mode="strict")
if span:
spans.append(span)
doc.ents = filter_spans(spans)
docs.append(doc)
return docs
def build(self) -> int:
input_filepath = get_data_file_path(
self.config.data.output_files["engineered"], self.config
)
df = self.data_loader.load_csv_complete(input_filepath)
df = df[["name", "ner_tagged", "ner_entities"]]
# Filter early
ner_df = df.loc[df["ner_tagged"] == 1, ["name", "ner_entities"]]
if ner_df.empty:
logging.error("No NER tagged data found")
return 1
total_rows = len(df)
del df # No need to keep in memory
logging.info(f"Found {len(ner_df)} NER tagged entries")
nlp = spacy.blank("fr")
# Vectorized parsing + validation
parsed_entities = self._parse_entities(ner_df["ner_entities"])
validated_entities = self._validate_entities(ner_df["name"], parsed_entities)
# Drop rows with no valid entities
mask = validated_entities.map(bool)
ner_df = ner_df.loc[mask]
validated_entities = validated_entities.loc[mask]
if ner_df.empty:
logging.error("No valid training examples after validation")
return 1
# Prepare training data
training_data = list(
zip(ner_df["name"].tolist(), [{"entities": ents} for ents in validated_entities])
)
# Create spaCy DocBin in batch
docs = self._create_docs(nlp, ner_df["name"].tolist(), validated_entities.tolist())
doc_bin = DocBin(docs=docs)
# Save
json_path = Path(self.config.paths.data_dir) / self.config.data.output_files["ner_data"]
spacy_path = Path(self.config.paths.data_dir) / self.config.data.output_files["ner_spacy"]
with open(json_path, "w", encoding="utf-8") as f:
json.dump(training_data, f, ensure_ascii=False, separators=(",", ":"))
doc_bin.to_disk(spacy_path)
logging.info(f"Processed: {len(training_data)}, Skipped: {total_rows - len(training_data)}")
logging.info(f"Saved NER JSON to {json_path}")
logging.info(f"Saved NER spacy to {spacy_path}")
return 0