File size: 9,907 Bytes
dad49b3 | 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 | {
"cells": [
{
"cell_type": "markdown",
"id": "93b5a3cf",
"metadata": {},
"source": [
"# EmpathRAG — FAISS Index Builder\n",
"**Run on Colab Pro with A100 GPU**\n",
"\n",
"Encodes Reddit Mental Health corpus into 768-dim vectors and builds a FAISS index + SQLite sidecar.\n",
"\n",
"- Estimated time on A100: 30–60 minutes\n",
"- Output: `faiss_flat.index` + `metadata.db` saved to Google Drive\n",
"- **Do not run on local RTX 3060 laptop** — would take 6–12 hours\n",
"\n",
"### Before running:\n",
"1. Upload `data/raw/reddit_mental_health/` (108 CSVs, ~3.1GB) to Google Drive at `My Drive/empathrag/data/raw/reddit_mental_health/`\n",
"2. Set Runtime → Change runtime type → **A100 GPU**\n",
"3. Run all cells in order"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b9ab42f1",
"metadata": {},
"outputs": [],
"source": [
"# Cell 1: Install dependencies\n",
"!pip install sentence-transformers faiss-cpu tqdm transformers -q"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9129b2f3",
"metadata": {},
"outputs": [],
"source": [
"# Cell 2: Mount Google Drive\n",
"from google.colab import drive\n",
"drive.mount(\"/content/drive\")\n",
"\n",
"BASE = \"/content/drive/MyDrive/empathrag\"\n",
"REDDIT_DIR = f\"{BASE}/data/raw/reddit_mental_health\"\n",
"INDEX_PATH = f\"{BASE}/data/indexes/faiss_flat.index\"\n",
"DB_PATH = f\"{BASE}/data/indexes/metadata.db\"\n",
"CKPT_PATH = f\"{BASE}/data/indexes/embeddings_checkpoint.npy\"\n",
"\n",
"import os\n",
"os.makedirs(f\"{BASE}/data/indexes\", exist_ok=True)\n",
"print(\"Drive mounted. Paths configured.\")\n",
"print(f\"Reddit dir exists: {os.path.exists(REDDIT_DIR)}\")\n",
"files = [f for f in os.listdir(REDDIT_DIR) if f.endswith('.csv')] if os.path.exists(REDDIT_DIR) else []\n",
"print(f\"CSV files found: {len(files)}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "54ccc2ba",
"metadata": {},
"outputs": [],
"source": [
"# Cell 3: Helper functions\n",
"import re\n",
"\n",
"def clean_text(text: str) -> str:\n",
" text = re.sub(r\"u/\\w+\", \"\", text)\n",
" text = re.sub(r\"r/\\w+\", \"\", text)\n",
" text = re.sub(r\"http\\S+\", \"\", text)\n",
" text = re.sub(r\"\\[deleted\\]|\\[removed\\]\", \"\", text)\n",
" text = re.sub(r\"[^\\x00-\\x7F]+\", \" \", text)\n",
" text = re.sub(r\"\\s+\", \" \", text).strip()\n",
" return text\n",
"\n",
"def chunk_text(text, tokenizer, chunk_size=256, stride=32, max_chunks=8):\n",
" tokens = tokenizer.encode(text)\n",
" if len(tokens) < 64:\n",
" return [text]\n",
" chunks = []\n",
" start = 0\n",
" while start < len(tokens) and len(chunks) < max_chunks:\n",
" end = min(start + chunk_size, len(tokens))\n",
" chunks.append(tokenizer.decode(tokens[start:end], skip_special_tokens=True))\n",
" start += chunk_size - stride\n",
" return chunks\n",
"\n",
"print(\"Helper functions defined.\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "68b83384",
"metadata": {},
"outputs": [],
"source": [
"# Cell 4: Load and chunk Reddit posts\n",
"# This reads all 108 CSVs and chunks the post text.\n",
"# Expected: 1-5 million chunks depending on corpus size.\n",
"import pandas as pd\n",
"from transformers import AutoTokenizer\n",
"from tqdm import tqdm\n",
"\n",
"tok = AutoTokenizer.from_pretrained(\"roberta-base\")\n",
"\n",
"all_posts = []\n",
"csv_files = [f for f in os.listdir(REDDIT_DIR) if f.endswith('.csv')]\n",
"print(f\"Reading {len(csv_files)} CSV files...\")\n",
"\n",
"for fname in tqdm(csv_files, desc=\"Reading CSVs\"):\n",
" fpath = os.path.join(REDDIT_DIR, fname)\n",
" try:\n",
" df = pd.read_csv(fpath, on_bad_lines=\"skip\",\n",
" usecols=lambda c: c in [\"post\", \"body\", \"selftext\"])\n",
" for col in [\"post\", \"body\", \"selftext\"]:\n",
" if col in df.columns:\n",
" all_posts.extend(df[col].dropna().tolist())\n",
" break\n",
" except Exception as e:\n",
" print(f\" Skipping {fname}: {e}\")\n",
"\n",
"print(f\"Raw posts loaded: {len(all_posts):,}\")\n",
"\n",
"chunks = []\n",
"for post in tqdm(all_posts, desc=\"Chunking\"):\n",
" cleaned = clean_text(str(post))\n",
" if not cleaned:\n",
" continue\n",
" chunks.extend(chunk_text(cleaned, tok))\n",
"\n",
"print(f\"Total chunks: {len(chunks):,}\")\n",
"\n",
"# Save chunks list to Drive as checkpoint\n",
"import pickle\n",
"with open(f\"{BASE}/data/indexes/chunks_checkpoint.pkl\", \"wb\") as f:\n",
" pickle.dump(chunks, f)\n",
"print(\"Chunks saved to Drive checkpoint.\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1ee62270",
"metadata": {},
"outputs": [],
"source": [
"# Cell 5: Encode embeddings\n",
"# Uses A100 GPU. Saves progress every 100K chunks so Colab disconnects are recoverable.\n",
"import numpy as np\n",
"from sentence_transformers import SentenceTransformer\n",
"import torch\n",
"\n",
"BATCH_SIZE = 256 # Safe for A100 40GB. Reduce to 64 if OOM.\n",
"SAVE_EVERY = 100_000 # Save checkpoint every N chunks\n",
"\n",
"encoder = SentenceTransformer(\"sentence-transformers/all-mpnet-base-v2\")\n",
"print(f\"Encoding on: {encoder.device}\")\n",
"print(f\"Total chunks to encode: {len(chunks):,}\")\n",
"print(f\"Estimated time at batch_size={BATCH_SIZE}: ~{len(chunks) // (BATCH_SIZE * 50)} minutes\")\n",
"\n",
"# Check if partial checkpoint exists\n",
"all_embeddings = []\n",
"start_idx = 0\n",
"\n",
"if os.path.exists(CKPT_PATH):\n",
" print(f\"Resuming from checkpoint: {CKPT_PATH}\")\n",
" all_embeddings = list(np.load(CKPT_PATH))\n",
" start_idx = len(all_embeddings)\n",
" print(f\"Resuming from chunk {start_idx:,}\")\n",
"\n",
"# Encode in segments\n",
"for seg_start in tqdm(range(start_idx, len(chunks), SAVE_EVERY), desc=\"Encoding segments\"):\n",
" seg_end = min(seg_start + SAVE_EVERY, len(chunks))\n",
" seg = chunks[seg_start:seg_end]\n",
" emb = encoder.encode(seg, batch_size=BATCH_SIZE, show_progress_bar=False,\n",
" normalize_embeddings=True, convert_to_numpy=True)\n",
" all_embeddings.extend(emb)\n",
" # Save checkpoint\n",
" np.save(CKPT_PATH, np.array(all_embeddings, dtype=np.float32))\n",
" print(f\" Checkpoint saved at {seg_end:,}/{len(chunks):,} chunks\")\n",
"\n",
"embeddings = np.array(all_embeddings, dtype=np.float32)\n",
"print(f\"Embeddings shape: {embeddings.shape}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a761c6e2",
"metadata": {},
"outputs": [],
"source": [
"# Cell 6: Build FAISS index and SQLite sidecar\n",
"import faiss\n",
"import sqlite3\n",
"\n",
"dim = embeddings.shape[1] # 768\n",
"n = embeddings.shape[0]\n",
"\n",
"print(f\"Building FAISS index for {n:,} vectors of dim {dim}...\")\n",
"\n",
"if n > 100_000:\n",
" quantizer = faiss.IndexFlatL2(dim)\n",
" index = faiss.IndexIVFFlat(quantizer, dim, 100)\n",
" print(\"Training IVFFlat index (this takes a few minutes)...\")\n",
" index.train(embeddings)\n",
"else:\n",
" index = faiss.IndexFlatL2(dim)\n",
"\n",
"index.add(embeddings)\n",
"faiss.write_index(index, INDEX_PATH)\n",
"print(f\"FAISS index saved: {index.ntotal:,} vectors → {INDEX_PATH}\")\n",
"\n",
"# SQLite sidecar\n",
"conn = sqlite3.connect(DB_PATH)\n",
"c = conn.cursor()\n",
"c.execute(\"\"\"CREATE TABLE IF NOT EXISTS chunks (\n",
" id INTEGER PRIMARY KEY,\n",
" text TEXT,\n",
" emotion_label INTEGER DEFAULT -1,\n",
" safety_score REAL DEFAULT 0.7,\n",
" source TEXT\n",
")\"\"\")\n",
"\n",
"BATCH = 10_000\n",
"for i in tqdm(range(0, len(chunks), BATCH), desc=\"Writing SQLite\"):\n",
" batch = chunks[i:i+BATCH]\n",
" c.executemany(\n",
" \"INSERT OR REPLACE INTO chunks VALUES (?,?,?,?,?)\",\n",
" [(i+j, text, -1, 0.7, \"reddit\") for j, text in enumerate(batch)]\n",
" )\n",
" conn.commit()\n",
"conn.close()\n",
"print(f\"SQLite DB saved: {len(chunks):,} rows → {DB_PATH}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "431f2949",
"metadata": {},
"outputs": [],
"source": [
"# Cell 7: Verify outputs\n",
"import faiss, sqlite3\n",
"\n",
"idx = faiss.read_index(INDEX_PATH)\n",
"print(f\"FAISS index: {idx.ntotal:,} vectors\")\n",
"\n",
"conn = sqlite3.connect(DB_PATH)\n",
"row_count = conn.execute(\"SELECT COUNT(*) FROM chunks\").fetchone()[0]\n",
"sample = conn.execute(\"SELECT text FROM chunks LIMIT 3\").fetchall()\n",
"conn.close()\n",
"print(f\"SQLite rows: {row_count:,}\")\n",
"print(\"Sample chunks:\")\n",
"for i, (t,) in enumerate(sample):\n",
" print(f\" [{i}] {t[:100]}...\")\n",
"\n",
"print()\n",
"print(\"=== BUILD COMPLETE ===\")\n",
"print(f\"Download these two files from Drive:\")\n",
"print(f\" {INDEX_PATH}\")\n",
"print(f\" {DB_PATH}\")\n",
"print(\"Place them in your local data/indexes/ folder.\")"
]
}
],
"metadata": {},
"nbformat": 4,
"nbformat_minor": 5
}
|