{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Quick start\n",
    "## Import datasets\n",
    "At first, import our library and datasets from the given path.\n",
    "Under the given directory, there should be a `data` directory with different datasets. E.g. `/home/AGL/data/cora`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import autogl\n",
    "from autogl.datasets import build_dataset_from_name\n",
    "cora_dataset = build_dataset_from_name('cora', path = '/home/AGL/')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decide modules\n",
    "Then, you should decide which models to use.\n",
    "Here, we use `deepgl` to pre-process graph features, then we use two GNNs to learn the target task, e.g. `GCN` and `GAT`.\n",
    "We use Simulated annealing algorithm to tune the hyper-parameters of the two GNNs. \n",
    "After training, use voting method to ensemble the results of the two GNNs.\n",
    "Also, you can specify which device to run on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "device = torch.device('cuda:5' if torch.cuda.is_available() else 'cpu')\n",
    "from autogl.solver import AutoNodeClassifier\n",
    "solver = AutoNodeClassifier(\n",
    "    feature_module='deepgl',\n",
    "    graph_models=['gcn', 'gat'],\n",
    "    hpo_module='anneal',\n",
    "    ensemble_module='voting',\n",
    "    device=device\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Running\n",
    "Run the whole process with a certain time limit and show the leaderboard.\n",
    "You can also get the accuracy by evaluate the predictions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "solver.fit(cora_dataset, time_limit=3600)\n",
    "solver.get_leaderboard().show()\n",
    "\n",
    "from autogl.module.train import Acc\n",
    "predicted = solver.predict_proba()\n",
    "print('Test accuracy: ', Acc.evaluate(predicted, \n",
    "    cora_dataset.data.y[cora_dataset.data.test_mask].cpu().numpy()))"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}