File size: 7,896 Bytes
dcc24f8 |
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 |
"""
Training Data Preparation Module
Converts parsed emails into training format for fine-tuning.
"""
import json
import random
from pathlib import Path
from typing import List, Dict, Tuple, Optional
from dataclasses import dataclass
# Import sibling modules
import sys
sys.path.insert(0, str(Path(__file__).parent.parent))
from data.extractor import EntityExtractor
@dataclass
class TrainingExample:
"""A single training example."""
prompt: str
completion: str
def to_dict(self) -> Dict:
return {"prompt": self.prompt, "completion": self.completion}
def to_jsonl(self) -> str:
return json.dumps(self.to_dict())
class TrainingDataPreparer:
"""
Prepare training data from parsed emails.
Converts emails into prompt-completion pairs for fine-tuning.
"""
# Prompt template for entity extraction task
EXTRACTION_PROMPT_TEMPLATE = """Extract financial entities from this email:
Subject: {subject}
Body: {body}"""
def __init__(
self,
emails_path: Path,
output_dir: Path,
extractor: Optional[EntityExtractor] = None
):
"""
Initialize preparer.
Args:
emails_path: Path to parsed emails JSON
output_dir: Directory to save training files
extractor: EntityExtractor instance (created if not provided)
"""
self.emails_path = Path(emails_path)
self.output_dir = Path(output_dir)
self.extractor = extractor or EntityExtractor()
# Load emails
with open(self.emails_path, 'r', encoding='utf-8') as f:
self.all_emails = json.load(f)
print(f"✅ Loaded {len(self.all_emails):,} emails")
def filter_transaction_emails(self) -> List[Dict]:
"""Filter emails that contain transaction data."""
transaction_emails = []
for email in self.all_emails:
body = email.get('body', '').lower()
subject = email.get('subject', '').lower()
combined = f"{subject} {body}"
# Must have transaction indicators
has_transaction = any(
kw in combined
for kw in ['debited', 'credited', 'payment of', 'transferred']
)
# Must have amount pattern
has_amount = 'rs' in combined or '₹' in combined
if has_transaction and has_amount:
transaction_emails.append(email)
print(f"📧 Transaction emails found: {len(transaction_emails):,}")
return transaction_emails
def create_training_examples(
self,
emails: Optional[List[Dict]] = None,
min_entities: int = 2,
max_body_length: int = 1500
) -> List[TrainingExample]:
"""
Convert emails to training examples.
Args:
emails: List of email dicts (uses transaction emails if None)
min_entities: Minimum entities required for valid example
max_body_length: Maximum body length in prompt
Returns:
List of TrainingExample objects
"""
if emails is None:
emails = self.filter_transaction_emails()
examples = []
skipped = 0
for email in emails:
# Extract entities
entities = self.extractor.extract_to_dict(email.get('body', ''))
# Skip if not enough entities
if len(entities) < min_entities:
skipped += 1
continue
# Create prompt
prompt = self.EXTRACTION_PROMPT_TEMPLATE.format(
subject=email.get('subject', '')[:200],
body=email.get('body', '')[:max_body_length]
)
# Create completion (JSON output)
completion = json.dumps(entities, indent=2)
examples.append(TrainingExample(prompt=prompt, completion=completion))
print(f"✅ Created {len(examples):,} training examples")
print(f"⏭️ Skipped {skipped:,} (insufficient entities)")
return examples
def split_data(
self,
examples: List[TrainingExample],
train_ratio: float = 0.9,
seed: int = 42
) -> Tuple[List[TrainingExample], List[TrainingExample]]:
"""
Split examples into train and validation sets.
Args:
examples: List of training examples
train_ratio: Ratio of examples for training (default 0.9)
seed: Random seed for reproducibility
Returns:
Tuple of (train_examples, valid_examples)
"""
random.seed(seed)
shuffled = examples.copy()
random.shuffle(shuffled)
split_idx = int(len(shuffled) * train_ratio)
train = shuffled[:split_idx]
valid = shuffled[split_idx:]
print(f"📊 Train: {len(train):,}, Validation: {len(valid):,}")
return train, valid
def save_jsonl(
self,
examples: List[TrainingExample],
filename: str
) -> Path:
"""Save examples to JSONL file."""
output_path = self.output_dir / filename
output_path.parent.mkdir(parents=True, exist_ok=True)
with open(output_path, 'w', encoding='utf-8') as f:
for example in examples:
f.write(example.to_jsonl() + '\n')
print(f"💾 Saved {len(examples):,} examples to {output_path}")
return output_path
def prepare(
self,
train_ratio: float = 0.9,
min_entities: int = 2,
seed: int = 42
) -> Tuple[Path, Path]:
"""
Full pipeline: filter → create examples → split → save.
Returns:
Tuple of (train_path, valid_path)
"""
print("\n🚀 Starting training data preparation...")
# Create examples
examples = self.create_training_examples(min_entities=min_entities)
if not examples:
raise ValueError("No training examples created!")
# Split data
train, valid = self.split_data(examples, train_ratio, seed)
# Save files
train_path = self.save_jsonl(train, "train.jsonl")
valid_path = self.save_jsonl(valid, "valid.jsonl")
# Summary
print("\n📋 Summary:")
print(f" Total examples: {len(examples):,}")
print(f" Train: {len(train):,}")
print(f" Valid: {len(valid):,}")
print(f" Output: {self.output_dir}")
return train_path, valid_path
def analyze_balance(self, examples: List[TrainingExample]) -> Dict[str, int]:
"""Analyze balance of transaction types in examples."""
debit_count = sum(
1 for e in examples
if '"type": "debit"' in e.completion
)
credit_count = sum(
1 for e in examples
if '"type": "credit"' in e.completion
)
return {
'debit': debit_count,
'credit': credit_count,
'other': len(examples) - debit_count - credit_count
}
if __name__ == "__main__":
from pathlib import Path
PROJECT = Path.home() / "llm-mail-trainer"
preparer = TrainingDataPreparer(
emails_path=PROJECT / "data/parsed/emails.json",
output_dir=PROJECT / "data/training"
)
train_path, valid_path = preparer.prepare()
# Check balance
examples = preparer.create_training_examples()
balance = preparer.analyze_balance(examples)
print(f"\n📊 Data Balance: {balance}")
|