{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**RLeXplore allows you to combine multiple intrinsic rewards to explore the potential  assembly advantages.**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Load the libraries**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch as th\n",
    "\n",
    "from rllte.env import make_atari_env\n",
    "from rllte.xplore.reward import Fabric, RE3, ICM\n",
    "from rllte.agent import PPO"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Use the `Fabric` class to create a mixed intrinsic reward**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class TwoMixed(Fabric):\n",
    "    def __init__(self, m1, m2):\n",
    "        super().__init__(m1, m2)\n",
    "    \n",
    "    def compute(self, samples, sync):\n",
    "        rwd1, rwd2 = super().compute(samples, sync)\n",
    "\n",
    "        return rwd1 + rwd2\n",
    "\n",
    "class ThreeMixed(Fabric):\n",
    "    def __init__(self, m1, m2, m3):\n",
    "        super().__init__(m1, m2, m3)\n",
    "    \n",
    "    def compute(self, samples, sync):\n",
    "        rwd1, rwd2, rw3 = super().compute(samples, sync)\n",
    "\n",
    "        return (rwd1 + rwd2) * rw3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set the parameters\n",
    "device = 'cuda' if th.cuda.is_available() else 'cpu'\n",
    "# create the vectorized environments\n",
    "envs = make_atari_env(env_id='BreakoutNoFrameskip-v4', device=device, num_envs=8)\n",
    "# create two intrinsic reward functions\n",
    "irs1 = ICM(envs, device)\n",
    "irs2 = RE3(envs, device)\n",
    "# create the mixed intrinsic reward function\n",
    "irs = TwoMixed(irs1, irs2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create the PPO agent\n",
    "agent = PPO(envs, device=device)\n",
    "# set the intrinsic reward module\n",
    "agent.set(reward=irs)\n",
    "# train the agent\n",
    "agent.train(10000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "marllib",
   "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": "3.8.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}