debug set and locking

src/paddle_ocr_raytune_rest.ipynb

@@ -7,263 +7,81 @@
    "source": [
     "# PaddleOCR Hyperparameter Optimization via REST API\n",
     "\n",
-    "This notebook runs Ray Tune hyperparameter search calling the PaddleOCR REST API (Docker container).\n",
+    "Uses Ray Tune + Optuna to find optimal PaddleOCR parameters.\n",
     "\n",
-    "**Benefits:**\n",
-    "- No model reload per trial - Model stays loaded in Docker container\n",
-    "- Faster trials - Skip ~10s model load time per trial\n",
-    "- Cleaner code - REST API replaces subprocess + CLI arg parsing"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "prereq",
-   "metadata": {},
-   "source": [
     "## Prerequisites\n",
     "\n",
-    "Start 2 PaddleOCR workers for parallel hyperparameter tuning:\n",
-    "\n",
     "```bash\n",
     "cd src/paddle_ocr\n",
-    "docker compose -f docker-compose.workers.yml up\n",
-    "```\n",
-    "\n",
-    "This starts 2 GPU workers on ports 8001-8002, allowing 2 concurrent trials.\n",
-    "\n",
-    "For CPU-only systems:\n",
-    "```bash\n",
-    "docker compose -f docker-compose.workers.yml --profile cpu up\n",
+    "docker compose -f docker-compose.workers.yml up  # GPU workers on 8001-8002\n",
+    "# or: docker compose -f docker-compose.workers.yml --profile cpu up\n",
     "```"
    ]
   },
   {
-   "cell_type": "markdown",
-   "id": "3ob9fsoilc4",
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "deps",
    "metadata": {},
+   "outputs": [],
    "source": [
-    "## 0. Dependencies"
+    "%pip install -q -U \"ray[tune]\" optuna requests pandas"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "wyr2nsoj7",
+   "id": "setup",
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Install dependencies (run once)\n",
-    "%pip install -U \"ray[tune]\"\n",
-    "%pip install optuna\n",
-    "%pip install requests pandas"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "imports-header",
-   "metadata": {},
-   "source": [
-    "## 1. Imports & Setup"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "imports",
-   "metadata": {},
-   "outputs": [],
-   "source": "import os\nfrom datetime import datetime\n\nimport requests\nimport pandas as pd\n\nimport ray\nfrom ray import tune, train\nfrom ray.tune.search.optuna import OptunaSearch"
-  },
-  {
-   "cell_type": "markdown",
-   "id": "config-header",
-   "metadata": {},
-   "source": [
-    "## 2. API Configuration"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "config",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# PaddleOCR REST API endpoints - 2 workers for parallel trials\n",
-    "# Start workers with: cd src/paddle_ocr && docker compose -f docker-compose.workers.yml up\n",
-    "WORKER_PORTS = [8001, 8002]\n",
-    "WORKER_URLS = [f\"http://localhost:{port}\" for port in WORKER_PORTS]\n",
+    "from raytune_ocr import (\n",
+    "    check_workers, create_trainable, run_tuner, analyze_results, correlation_analysis,\n",
+    "    paddle_ocr_payload, PADDLE_OCR_SEARCH_SPACE, PADDLE_OCR_CONFIG_KEYS,\n",
+    ")\n",
     "\n",
-    "# Output folder for results\n",
-    "OUTPUT_FOLDER = \"results\"\n",
-    "os.makedirs(OUTPUT_FOLDER, exist_ok=True)\n",
+    "# Worker ports\n",
+    "PORTS = [8001, 8002]\n",
     "\n",
-    "# Number of concurrent trials = number of workers\n",
-    "NUM_WORKERS = len(WORKER_URLS)"
+    "# Check workers are running\n",
+    "healthy = check_workers(PORTS, \"PaddleOCR\")"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "health-check",
+   "id": "tune",
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Verify all workers are running\n",
-    "healthy_workers = []\n",
-    "for url in WORKER_URLS:\n",
-    "    try:\n",
-    "        health = requests.get(f\"{url}/health\", timeout=10).json()\n",
-    "        if health['status'] == 'ok' and health['model_loaded']:\n",
-    "            healthy_workers.append(url)\n",
-    "            print(f\"✓ {url}: {health['status']} (GPU: {health.get('gpu_name', 'N/A')})\")\n",
-    "        else:\n",
-    "            print(f\"✗ {url}: not ready yet\")\n",
-    "    except requests.exceptions.ConnectionError:\n",
-    "        print(f\"✗ {url}: not reachable\")\n",
+    "# Create trainable and run tuning\n",
+    "trainable = create_trainable(PORTS, paddle_ocr_payload)\n",
     "\n",
-    "if not healthy_workers:\n",
-    "    raise RuntimeError(\n",
-    "        \"No healthy workers found. Start them with:\\n\"\n",
-    "        \"  cd src/paddle_ocr && docker compose -f docker-compose.workers.yml up\"\n",
-    "    )\n",
+    "results = run_tuner(\n",
+    "    trainable=trainable,\n",
+    "    search_space=PADDLE_OCR_SEARCH_SPACE,\n",
+    "    num_samples=64,\n",
+    "    num_workers=len(healthy),\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "analysis",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Analyze results\n",
+    "df = analyze_results(\n",
+    "    results,\n",
+    "    prefix=\"raytune_paddle\",\n",
+    "    config_keys=PADDLE_OCR_CONFIG_KEYS,\n",
+    ")\n",
     "\n",
-    "print(f\"\\n{len(healthy_workers)}/{len(WORKER_URLS)} workers ready for parallel tuning\")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "search-space-header",
-   "metadata": {},
-   "source": [
-    "## 3. Search Space"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "search-space",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "search_space = {\n",
-    "    # Whether to use document image orientation classification\n",
-    "    \"use_doc_orientation_classify\": tune.choice([True, False]),\n",
-    "    # Whether to use text image unwarping\n",
-    "    \"use_doc_unwarping\": tune.choice([True, False]),\n",
-    "    # Whether to use text line orientation classification\n",
-    "    \"textline_orientation\": tune.choice([True, False]),\n",
-    "    # Detection pixel threshold (pixels > threshold are considered text)\n",
-    "    \"text_det_thresh\": tune.uniform(0.0, 0.7),\n",
-    "    # Detection box threshold (average score within border)\n",
-    "    \"text_det_box_thresh\": tune.uniform(0.0, 0.7),\n",
-    "    # Text detection expansion coefficient\n",
-    "    \"text_det_unclip_ratio\": tune.choice([0.0]),\n",
-    "    # Text recognition threshold (filter low confidence results)\n",
-    "    \"text_rec_score_thresh\": tune.uniform(0.0, 0.7),\n",
-    "}"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "trainable-header",
-   "metadata": {},
-   "source": [
-    "## 4. Trainable Function"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "trainable",
-   "metadata": {},
-   "outputs": [],
-   "source": "def trainable_paddle_ocr(config):\n    \"\"\"Call PaddleOCR REST API with the given hyperparameter config.\"\"\"\n    import random\n    import requests\n    from ray import train\n\n    # Worker URLs - random selection (load balances with 2 workers, 2 concurrent trials)\n    WORKER_PORTS = [8001, 8002]\n    api_url = f\"http://localhost:{random.choice(WORKER_PORTS)}\"\n\n    payload = {\n        \"pdf_folder\": \"/app/dataset\",\n        \"use_doc_orientation_classify\": config.get(\"use_doc_orientation_classify\", False),\n        \"use_doc_unwarping\": config.get(\"use_doc_unwarping\", False),\n        \"textline_orientation\": config.get(\"textline_orientation\", True),\n        \"text_det_thresh\": config.get(\"text_det_thresh\", 0.0),\n        \"text_det_box_thresh\": config.get(\"text_det_box_thresh\", 0.0),\n        \"text_det_unclip_ratio\": config.get(\"text_det_unclip_ratio\", 1.5),\n        \"text_rec_score_thresh\": config.get(\"text_rec_score_thresh\", 0.0),\n        \"start_page\": 5,\n        \"end_page\": 10,\n    }\n\n    try:\n        response = requests.post(f\"{api_url}/evaluate\", json=payload, timeout=None)\n        response.raise_for_status()\n        metrics = response.json()\n        metrics[\"worker\"] = api_url\n        train.report(metrics)\n    except Exception as e:\n        train.report({\n            \"CER\": 1.0,\n            \"WER\": 1.0,\n            \"TIME\": 0.0,\n            \"PAGES\": 0,\n            \"TIME_PER_PAGE\": 0,\n            \"worker\": api_url,\n            \"ERROR\": str(e)[:500]\n        })"
-  },
-  {
-   "cell_type": "markdown",
-   "id": "tuner-header",
-   "metadata": {},
-   "source": [
-    "## 5. Run Tuner"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "ray-init",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "ray.init(ignore_reinit_error=True)\n",
-    "print(f\"Ray Tune ready (version: {ray.__version__})\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "tuner",
-   "metadata": {},
-   "outputs": [],
-   "source": "tuner = tune.Tuner(\n    trainable_paddle_ocr,\n    tune_config=tune.TuneConfig(\n        metric=\"CER\",\n        mode=\"min\",\n        search_alg=OptunaSearch(),\n        num_samples=64,\n        max_concurrent_trials=NUM_WORKERS,  # Run trials in parallel across workers\n    ),\n    param_space=search_space,\n)\n\nresults = tuner.fit()"
-  },
-  {
-   "cell_type": "markdown",
-   "id": "analysis-header",
-   "metadata": {},
-   "source": [
-    "## 6. Results Analysis"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "results-df",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "df = results.get_dataframe()\n",
     "df.describe()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "save-results",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# Save results to CSV\n",
-    "timestamp = datetime.now().strftime(\"%Y%m%d_%H%M%S\")\n",
-    "filename = f\"raytune_paddle_rest_results_{timestamp}.csv\"\n",
-    "filepath = os.path.join(OUTPUT_FOLDER, filename)\n",
-    "\n",
-    "df.to_csv(filepath, index=False)\n",
-    "print(f\"Results saved: {filepath}\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "best-config",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# Best configuration\n",
-    "best = df.loc[df[\"CER\"].idxmin()]\n",
-    "\n",
-    "print(f\"Best CER: {best['CER']:.6f}\")\n",
-    "print(f\"Best WER: {best['WER']:.6f}\")\n",
-    "print(f\"\\nOptimal Configuration:\")\n",
-    "print(f\"  textline_orientation: {best['config/textline_orientation']}\")\n",
-    "print(f\"  use_doc_orientation_classify: {best['config/use_doc_orientation_classify']}\")\n",
-    "print(f\"  use_doc_unwarping: {best['config/use_doc_unwarping']}\")\n",
-    "print(f\"  text_det_thresh: {best['config/text_det_thresh']:.4f}\")\n",
-    "print(f\"  text_det_box_thresh: {best['config/text_det_box_thresh']:.4f}\")\n",
-    "print(f\"  text_det_unclip_ratio: {best['config/text_det_unclip_ratio']}\")\n",
-    "print(f\"  text_rec_score_thresh: {best['config/text_rec_score_thresh']:.4f}\")"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -272,42 +90,21 @@
    "outputs": [],
    "source": [
     "# Correlation analysis\n",
-    "param_cols = [\n",
-    "    \"config/text_det_thresh\",\n",
-    "    \"config/text_det_box_thresh\",\n",
-    "    \"config/text_det_unclip_ratio\",\n",
-    "    \"config/text_rec_score_thresh\",\n",
-    "]\n",
-    "\n",
-    "corr_cer = df[param_cols + [\"CER\"]].corr()[\"CER\"].sort_values(ascending=False)\n",
-    "corr_wer = df[param_cols + [\"WER\"]].corr()[\"WER\"].sort_values(ascending=False)\n",
-    "\n",
-    "print(\"Correlation with CER:\")\n",
-    "print(corr_cer)\n",
-    "print(\"\\nCorrelation with WER:\")\n",
-    "print(corr_wer)"
+    "correlation_analysis(df, PADDLE_OCR_CONFIG_KEYS)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": ".venv",
+   "display_name": "Python 3",
    "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": "3.12.3"
+   "version": "3.10.0"
   }
  },
  "nbformat": 4,
  "nbformat_minor": 5
-}
+}