{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "4a37d83e-7866-4006-b27b-06d82da2cbef",
   "metadata": {},
   "source": [
    "# Optional Example: Convert simulator backend to ONNX\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f6fd83b2-d529-4ba2-b22f-268e2637ba78",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "%config InlineBackend.figure_format = \"retina\"\n",
    "\n",
    "# Require:\n",
    "#%pip install onnxruntime\n",
    "\n",
    "import my_nb_path  # isort: split\n",
    "import a2rl as wi\n",
    "import numpy as np\n",
    "import torch\n",
    "from stable_baselines3 import A2C\n",
    "\n",
    "from flight_sales.flight_sales_gym import flight_sales_gym\n",
    "\n",
    "# Load artifacts\n",
    "m = torch.load(\"model-dyn-pricing/model.pt\")\n",
    "tokenizer = wi.utils.pickle_load(\"model-dyn-pricing/tokenizer.pt\")\n",
    "\n",
    "# From data frame, get an input sequence\n",
    "env = flight_sales_gym(f_reward=\"revenue_0_05\")\n",
    "env.reset()\n",
    "env.step(0.5)\n",
    "len_ar = len(tokenizer.df.actions) + len(tokenizer.df.rewards)\n",
    "field_tokenizer = tokenizer.field_tokenizer\n",
    "ctx = np.tile(\n",
    "    field_tokenizer.transform(env.context(tail=2, fillna=True)).values.ravel()[:-len_ar],\n",
    "    (1, 1),\n",
    ")\n",
    "ctx = torch.from_numpy(ctx)\n",
    "\n",
    "# Export as ONNX format.\n",
    "torch.onnx.export(\n",
    "    m,\n",
    "    ctx,\n",
    "    \"model-dyn-pricing/model.onnx\",\n",
    "    verbose=True,\n",
    "    input_names=[\"input\"],\n",
    "    output_names=[\"output\"],\n",
    "    export_params=True,\n",
    "    do_constant_folding=True,\n",
    ")\n",
    "\n",
    "# Optional next step: visualize with https://github.com/lutzroeder/netron."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3662804a-485e-4d01-9632-c5b8e3d8e637",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.10.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}