{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "eb9fb4a1a68e633",
   "metadata": {},
   "source": [
    "# Tutorial AAAI 2026"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1ac309b575ed4764",
   "metadata": {},
   "source": [
    "AMLGym provides several benchmark domains from previous Internaional Planning Competitions (IPCs)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9bb5ad965be80b96",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T14:06:08.177367Z",
     "start_time": "2026-01-19T14:06:06.758340Z"
    }
   },
   "outputs": [],
   "source": [
    "!pip install amlgym > /dev/null 2>&1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "initial_id",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:47:19.567245Z",
     "start_time": "2026-01-19T08:47:19.560728Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import print_domains\n",
    "\n",
    "print_domains()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bf47f050c2091de8",
   "metadata": {},
   "source": [
    "The PDDL model of each domain can be inspected:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fd36b0d1fc890673",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:47:22.280376Z",
     "start_time": "2026-01-19T08:47:22.276892Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_domain_path\n",
    "\n",
    "domain_path = get_domain_path('blocksworld')\n",
    "print(domain_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9f69a6f17bb858d8",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:47:23.926085Z",
     "start_time": "2026-01-19T08:47:23.922732Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_domain\n",
    "\n",
    "domain_pddl = get_domain('blocksworld')\n",
    "print(domain_pddl)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a1140b080ac0052c",
   "metadata": {},
   "source": [
    "A set of 10 trajectories is provided for each domain:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2248b833a8b45659",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:47:29.441556Z",
     "start_time": "2026-01-19T08:47:29.435264Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_trajectories_path\n",
    "from pprint import pprint\n",
    "\n",
    "trajectory_paths = get_trajectories_path('blocksworld')\n",
    "pprint(trajectory_paths)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a9308a44e4025670",
   "metadata": {},
   "source": [
    "An example of trajectory for the `blocksworld` domain is:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e5bdac6da94e245d",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:47:34.386507Z",
     "start_time": "2026-01-19T08:47:34.380704Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_trajectories\n",
    "\n",
    "trajectory_idx = 0\n",
    "trajectory = get_trajectories('blocksworld')[trajectory_idx]\n",
    "print(trajectory)"
   ]
  },
  {
   "cell_type": "raw",
   "id": "fc68d4ccab3a98b3",
   "metadata": {},
   "source": [
    "The problem used for generating the trajectory can be obtained by:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5e42c531f4094aa",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:50:08.231726Z",
     "start_time": "2026-01-19T08:50:08.226442Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_problems_path\n",
    "\n",
    "problem_path = get_problems_path('blocksworld')[trajectory_idx]\n",
    "print(problem_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7bb6c8d57b7bf935",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:50:18.755776Z",
     "start_time": "2026-01-19T08:50:18.752924Z"
    }
   },
   "outputs": [],
   "source": [
    "with open(problem_path, 'r') as f:\n",
    "    print(f.read())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "80d1b88748df61ad",
   "metadata": {},
   "source": [
    "## Passive Algorithms"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9ac175afe650249f",
   "metadata": {},
   "source": [
    "AMLGym includes several algorithms for passive learning domain models from full/partial/noisy trajectories."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fe7fa775676f69e7",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:51:32.840730Z",
     "start_time": "2026-01-19T08:51:29.342303Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.algorithms import print_algorithms\n",
    "\n",
    "print_algorithms()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6765055c8a63d35",
   "metadata": {},
   "source": [
    "### Passive learning with full observability"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5aa808127d97a0aa",
   "metadata": {},
   "source": [
    "Instantiate an algorithm for passive learning domain models in fully observable environments (e.g., [SAM](https://proceedings.kr.org/2021/36/)):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a41e20826318a782",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:56:28.569991Z",
     "start_time": "2026-01-19T08:56:28.565182Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.algorithms import get_algorithm\n",
    "\n",
    "learner = get_algorithm('sam')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b1037c13855c660e",
   "metadata": {},
   "source": [
    "Get an empty IPC domain:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a99c933d94d01d00",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:56:37.649738Z",
     "start_time": "2026-01-19T08:56:37.495607Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.util.util import empty_domain\n",
    "\n",
    "domain_path = get_domain_path('blocksworld')\n",
    "\n",
    "domain_empty_path = empty_domain(domain_path)\n",
    "with open(domain_empty_path, 'r') as f:\n",
    "    print(f.read())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ba076d560eb96056",
   "metadata": {},
   "source": [
    "Get a set of trajectories associated with the domain:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9e8ba83a6cd89a6c",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:57:34.573166Z",
     "start_time": "2026-01-19T08:57:34.569111Z"
    }
   },
   "outputs": [],
   "source": [
    "traj_paths = get_trajectories_path('blocksworld')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f7c9ddcdd8b5c4bd",
   "metadata": {},
   "source": [
    "Learn a domain model specified in PDDL:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a8dd91c9f6b33c9",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T08:57:39.535753Z",
     "start_time": "2026-01-19T08:57:39.464850Z"
    }
   },
   "outputs": [],
   "source": [
    "model = learner.learn(domain_empty_path, traj_paths)\n",
    "print(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ba8a6d10b270e5e5",
   "metadata": {},
   "source": [
    "Save the learned model to a file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bfe618e6c595c95c",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:03:05.603379Z",
     "start_time": "2026-01-19T10:03:05.600336Z"
    }
   },
   "outputs": [],
   "source": [
    "domain_learned_path = 'domain_learned.pddl'\n",
    "with open(domain_learned_path, 'w') as f:\n",
    "    f.write(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "22c327014bb9a7a",
   "metadata": {},
   "source": [
    "## Evaluating a domain model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "edc3c77f83382608",
   "metadata": {},
   "source": [
    "AMLGym provides several measures for evaluation an action model w.r.t. a reference model or simulator."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec7aed30828b4ef0",
   "metadata": {},
   "source": [
    "### Syntactic Similarity"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "663de3d2b7129332",
   "metadata": {},
   "source": [
    "[Syntactic similarity](https://amlgym.readthedocs.io/en/latest/metrics/syntactic.html) can be measured by means of precision and recall:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "637f44d49e748a27",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:03:09.961801Z",
     "start_time": "2026-01-19T10:03:09.929788Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.metrics import syntactic_precision\n",
    "from pprint import pprint\n",
    "\n",
    "domain_ref_path = get_domain_path('blocksworld')\n",
    "precision = syntactic_precision(domain_learned_path, domain_ref_path)\n",
    "pprint(precision)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c90516c3f71cf6be",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:03:29.839448Z",
     "start_time": "2026-01-19T10:03:29.814863Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.metrics import syntactic_recall\n",
    "\n",
    "recall = syntactic_recall(domain_learned_path, domain_ref_path)\n",
    "pprint(recall)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4483500d62c03de3",
   "metadata": {},
   "source": [
    "### Problem solving"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4c6846a8d23e5e2",
   "metadata": {},
   "source": [
    "Get the set of problems necessary to evaluate [problem solving](https://amlgym.readthedocs.io/en/latest/metrics/solving.html) metrics:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "77137504f589c228",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:05:54.095958Z",
     "start_time": "2026-01-19T10:05:54.089401Z"
    }
   },
   "outputs": [],
   "source": [
    "probs_paths = get_problems_path('blocksworld', kind='solving')\n",
    "pprint(probs_paths)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e1ce941e0a92ce4a",
   "metadata": {},
   "source": [
    "Solve the problems with the model to be evaluated and evaluate it in a simulator defined by a reference model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4e119069be726c39",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:06:13.408270Z",
     "start_time": "2026-01-19T10:06:09.429748Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.metrics import problem_solving\n",
    "import unified_planning\n",
    "unified_planning.shortcuts.get_environment().credits_stream = None\n",
    "\n",
    "metrics = problem_solving(domain_learned_path, domain_ref_path, probs_paths, timeout=60)\n",
    "pprint(metrics)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "70cd0c1944c83fe4",
   "metadata": {},
   "source": [
    "### Predictive power"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5a28745fed48d215",
   "metadata": {},
   "source": [
    "Get the test set of states necessary to evaluate [predicted applicability and predicted effects](https://amlgym.readthedocs.io/en/latest/metrics/predictive.html) metrics:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e6e398e152bb6a76",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:23:13.258873Z",
     "start_time": "2026-01-19T10:23:13.253931Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.benchmarks import get_test_states\n",
    "\n",
    "test_states = get_test_states('blocksworld')\n",
    "print(len(test_states.keys()))\n",
    "pprint(next(iter(test_states.keys())))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ad28f33d9911677e",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:23:13.691036Z",
     "start_time": "2026-01-19T10:23:13.687057Z"
    }
   },
   "outputs": [],
   "source": [
    "problem_paths = get_problems_path('blocksworld', kind='predictive_power')\n",
    "\n",
    "# get first problem file path\n",
    "problem_path = problem_paths[0]\n",
    "\n",
    "# get test set of state for a single problem\n",
    "test_states = test_states[problem_path.split('/')[-1]]\n",
    "pprint(test_states)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d9d972d7e58019e",
   "metadata": {},
   "source": [
    "Create a simulator defined by the learned domain model and an environment simulator defined by a reference domain model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "86e3b1cbe0502833",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:24:51.575594Z",
     "start_time": "2026-01-19T10:24:51.340424Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.modeling.UPEnv import UPEnv\n",
    "\n",
    "simulator_learned = UPEnv(domain_learned_path, problem_path)\n",
    "simulator_ref = UPEnv(domain_ref_path, problem_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f467a155ecd10cab",
   "metadata": {},
   "source": [
    "Evaluate the predicted applicability and predicted effects metrics with a learned domain simulator and environment simulator."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cbf0b2790848de21",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T10:25:04.164109Z",
     "start_time": "2026-01-19T10:25:04.131159Z"
    }
   },
   "outputs": [],
   "source": [
    "from amlgym.metrics import predictive_power\n",
    "\n",
    "predictive_metrics = predictive_power(simulator_learned,\n",
    "                                      simulator_ref,\n",
    "                                      test_states)\n",
    "pprint(predictive_metrics)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d5b5e855a93ea3f4",
   "metadata": {},
   "source": [
    "## Comparing passive learning performances with full observability"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4634e006b1f79796",
   "metadata": {},
   "source": [
    "**Question A: Who achieves higher syntactic precision between NOLAM and OffLAM in domain tpp?**\n",
    "\n",
    "Instantiate the learning algorithms:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8d1c9a35da3e900e",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T13:25:12.475726Z",
     "start_time": "2026-01-19T13:25:12.469055Z"
    }
   },
   "outputs": [],
   "source": [
    "offlam = get_algorithm('offlam')\n",
    "nolam = get_algorithm('nolam', noise=0.)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "806fe1518301a026",
   "metadata": {},
   "source": [
    "Get an input (empty IPC) domain for tpp and the associated set of learning trajectories:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c3789d8988dffcd3",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T13:52:04.825732Z",
     "start_time": "2026-01-19T13:52:04.772103Z"
    }
   },
   "outputs": [],
   "source": [
    "domain_path = get_domain_path('tpp')\n",
    "domain_empty_path = empty_domain(domain_path)\n",
    "traj_paths = get_trajectories_path('tpp')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67cff31da01c7339",
   "metadata": {},
   "source": [
    "Learn a domain model specified in PDDL:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "27f8b237da00f033",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T13:52:08.360669Z",
     "start_time": "2026-01-19T13:52:07.173974Z"
    }
   },
   "outputs": [],
   "source": [
    "# Learn a domain model using OffLAM\n",
    "domain_offlam = offlam.learn(domain_empty_path, traj_paths)\n",
    "\n",
    "# Learn a domain model using NOLAM\n",
    "domain_nolam = nolam.learn(domain_empty_path, traj_paths)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "acf4b2586fdeb029",
   "metadata": {},
   "source": [
    "Store the learned models for evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a2d745db5573c9ed",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T13:52:10.853608Z",
     "start_time": "2026-01-19T13:52:10.850453Z"
    }
   },
   "outputs": [],
   "source": [
    "domain_offlam_path = 'offlam.pddl'\n",
    "with open(domain_offlam_path, 'w') as f:\n",
    "    f.write(domain_offlam)\n",
    "    \n",
    "domain_nolam_path = 'nolam.pddl'\n",
    "with open(domain_nolam_path, 'w') as f:\n",
    "    f.write(domain_nolam)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1f470295bce9df89",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T15:32:38.331892Z",
     "start_time": "2026-01-19T15:32:38.329975Z"
    }
   },
   "outputs": [],
   "source": [
    "# ..."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4bbb9c9135dd36c5",
   "metadata": {},
   "source": [
    "**Question B: Who achieves the highest solving ratio among SAM and OffLAM in domain goldminer?**\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6795e73e13bea543",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2026-01-19T15:32:20.129393Z",
     "start_time": "2026-01-19T15:32:20.126921Z"
    }
   },
   "outputs": [],
   "source": [
    "# ..."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}