MastersThesis/src/doctr_raytune_rest.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "header",
   "metadata": {},
   "source": [
    "# DocTR Hyperparameter Optimization via REST API\n",
    "\n",
    "Uses Ray Tune + Optuna to find optimal DocTR parameters.\n",
    "\n",
    "## Prerequisites\n",
    "\n",
    "```bash\n",
    "cd src/doctr_service\n",
    "docker compose up ocr-cpu  # or ocr-gpu\n",
    "```\n",
    "\n",
    "Service runs on port 8003."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "deps",
   "metadata": {},
   "outputs": [],
   "source": "# Pin Ray version for API stability (tune.report takes dict, not kwargs in 2.x)\n%pip install -q \"ray[tune]==2.53.0\" optuna requests pandas"
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "setup",
   "metadata": {},
   "outputs": [],
   "source": [
    "from raytune_ocr import (\n",
    "    check_workers, create_trainable, run_tuner, analyze_results, correlation_analysis,\n",
    "    doctr_payload, DOCTR_SEARCH_SPACE, DOCTR_CONFIG_KEYS,\n",
    ")\n",
    "\n",
    "# Worker ports\n",
    "PORTS = [8003]\n",
    "\n",
    "# Check workers are running\n",
    "healthy = check_workers(PORTS, \"DocTR\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "tune",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create trainable and run tuning\n",
    "trainable = create_trainable(PORTS, doctr_payload)\n",
    "\n",
    "results = run_tuner(\n",
    "    trainable=trainable,\n",
    "    search_space=DOCTR_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_doctr\",\n",
    "    config_keys=DOCTR_CONFIG_KEYS,\n",
    ")\n",
    "\n",
    "df.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "correlation",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Correlation analysis\n",
    "correlation_analysis(df, DOCTR_CONFIG_KEYS)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.10.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}