File size: 7,734 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 |
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "bfd5cfe5-ea7e-49d8-9ef3-5d43bef5a0cf",
"metadata": {},
"outputs": [],
"source": [
"\n",
"import json\n",
"import re\n",
"import random\n",
"import time\n",
"from pathlib import Path\n",
"from tqdm.notebook import tqdm\n",
"from mlx_lm import load, generate\n",
"from collections import Counter\n",
"\n",
"\n",
"# Cell 11: Load parsed emails from cache\n",
"\n",
"cache_path = PROJECT / \"data/parsed/emails.json\"\n",
"\n",
"with open(cache_path, 'r', encoding='utf-8') as f:\n",
" parsed_emails = json.load(f)\n",
"\n",
"print(f\"✅ Loaded {len(parsed_emails):,} emails from cache\")\n",
"\n",
"# Cell 12: Random sampling\n",
"random.seed(42)\n",
"\n",
"# Pick 500 random emails\n",
"sample_size = 500\n",
"sample_emails = random.sample(parsed_emails,sample_size)\n",
"\n",
"print(f\"Total emails: {len(parsed_emails):,}\")\n",
"print(f\"Sample size: {len(sample_emails)}\")\n",
"\n",
"# Preview one sample\n",
"print(f\"\\n=== SAMPLE EMAIL #1 ===\")\n",
"print(f\"Subject: {sample_emails[0]['subject']}\")\n",
"print(f\"Sender: {sample_emails[0]['sender']}\")\n",
"print(f\"Body: {sample_emails[0]['body'][:300]}...\")\n",
"\n",
"# Cell 13: Classification prompt template\n",
"\n",
"CLASSIFICATION_PROMPT = \"\"\"You are an email classifier. Analyze this email and categorize it.\n",
"\n",
"EMAIL:\n",
"Subject: {subject}\n",
"From: {sender}\n",
"Body: {body}\n",
"\n",
"TASK:\n",
"Classify this email into exactly ONE category.\n",
"\n",
"CATEGORIES:\n",
"- finance: Banks, payments, transactions, investments, credit cards, loans, UPI, wallets\n",
"- shopping: Orders, deliveries, purchases, e-commerce\n",
"- social: Social networks, personal messages, invitations\n",
"- work: Job-related, recruitment, office, meetings, projects\n",
"- newsletter: Digests, subscriptions, blogs, articles\n",
"- promotional: Marketing, offers, discounts, advertisements\n",
"- other: Anything that doesn't fit above\n",
"\n",
"OUTPUT FORMAT (JSON only, no other text):\n",
"{{\"category\": \"<category>\", \"confidence\": \"<high/medium/low>\", \"reason\": \"<brief 5-10 word reason>\"}}\n",
"\"\"\"\n",
"\n",
"def build_prompt(email_data):\n",
" \"\"\"Build classification prompt for one email.\"\"\"\n",
" return CLASSIFICATION_PROMPT.format(\n",
" subject=email_data['subject'][:200],\n",
" sender=email_data['sender'][:100],\n",
" body=email_data['body'][:2000]\n",
" )\n",
"\n",
"# Test: See what prompt looks like\n",
"test_prompt = build_prompt(sample_emails[0])\n",
"print(f\"Prompt length: {len(test_prompt)} characters\")\n",
"print(f\"\\n=== PROMPT PREVIEW ===\\n{test_prompt[:1000]}...\")\n",
"\n",
"# Cell 14: Load Phi-3 model\n",
"model_path = str(PROJECT / \"models/base/phi3-mini\")\n",
"\n",
"print(\"Loading Phi-3 model...\")\n",
"model, tokenizer = load(model_path)\n",
"print(\"✅ Model loaded\")\n",
"\n",
"# Cell 15: Test classification on one email\n",
"test_email = sample_emails[0]\n",
"\n",
"# Build prompt\n",
"prompt = build_prompt(test_email)\n",
"\n",
"# Send to Phi-3\n",
"print(\"Classifying email...\")\n",
"print(f\"Subject: {test_email['subject'][:80]}...\")\n",
"print(\"-\" * 50)\n",
"\n",
"response = generate(\n",
" model, \n",
" tokenizer, \n",
" prompt=prompt,\n",
" max_tokens=100,\n",
" verbose=False\n",
")\n",
"\n",
"print(f\"\\n=== PHI-3 RESPONSE ===\\n{response}\")\n",
"\n",
"# Cell 16: JSON extraction helper\n",
"\n",
"def extract_json(response):\n",
" \"\"\"Extract JSON object from LLM response.\"\"\"\n",
"\n",
" # Find JSON pattern in response\n",
" match = re.search(r'\\{[^{}]*\\}', response)\n",
"\n",
" if(match):\n",
" try:\n",
" return json.loads(match.group())\n",
" except json.JSONDecodeError:\n",
" return None\n",
" return None\n",
"\n",
"# Test on previous response\n",
"parsed = extract_json(response)\n",
"\n",
"print(\"=== EXTRACTED JSON ===\")\n",
"print(parsed)\n",
"print(f\"\\nCategory: {parsed['category']}\")\n",
"print(f\"Confidence: {parsed['confidence']}\")\n",
"print(f\"Reason: {parsed['reason']}\")\n",
"\n",
"# Cell 17: Classify all sample emails\n",
"results = []\n",
"failed = 0\n",
"\n",
"print(f\"Classifying {len(sample_emails)} emails...\")\n",
"print(\"Estimated time: ~5 minutes\\n\")\n",
"\n",
"start_time = time.time()\n",
"\n",
"for i, email_data in enumerate(tqdm(sample_emails, desc=\"Classifying\")):\n",
" try:\n",
" # Build prompt\n",
" prompt = build_prompt(email_data)\n",
" \n",
" # Get classification\n",
" response = generate(\n",
" model, \n",
" tokenizer, \n",
" prompt=prompt,\n",
" max_tokens=100,\n",
" verbose=False\n",
" )\n",
" \n",
" # Extract JSON\n",
" parsed = extract_json(response)\n",
" \n",
" if parsed:\n",
" results.append({\n",
" 'id': email_data.get('id', i),\n",
" 'subject': email_data['subject'],\n",
" 'sender': email_data['sender'],\n",
" 'category': parsed.get('category', 'other'),\n",
" 'confidence': parsed.get('confidence', 'low'),\n",
" 'reason': parsed.get('reason', '')\n",
" })\n",
" else:\n",
" failed += 1\n",
" \n",
" except Exception as e:\n",
" failed += 1\n",
" continue\n",
"\n",
"elapsed = time.time() - start_time\n",
"\n",
"print(f\"\\n✅ Classified: {len(results)}\")\n",
"print(f\"❌ Failed: {failed}\")\n",
"print(f\"⏱️ Time: {elapsed/60:.1f} minutes\")\n",
"print(f\"⚡ Speed: {len(results)/elapsed:.1f} emails/sec\")\n",
"\n",
"# Cell 18: Category distribution\n",
"\n",
"categories = Counter([r['category'] for r in results])\n",
"\n",
"print(\"=== CATEGORY DISTRIBUTION ===\\n\")\n",
"for category, count in categories.most_common():\n",
" pct = count / len(results) * 100\n",
" bar = \"█\" * int(pct / 2)\n",
" print(f\"{category:12} {count:4} ({pct:5.1f}%) {bar}\")\n",
"\n",
"print(f\"\\n📊 Total classified: {len(results)}\")\n",
"\n",
"# Cell 19: Save classification results\n",
"results_path = PROJECT / \"data/parsed/classification_results.json\"\n",
"\n",
"with open(results_path, 'w', encoding='utf-8') as f:\n",
" json.dump(results, f, ensure_ascii=False, indent=2)\n",
"\n",
"print(f\"✅ Saved {len(results)} results to {results_path}\")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.6"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|