Untitled1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "failing-calibration",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import argparse\n",
    "import pdb\n",
    "import seaborn as sns\n",
    "from dataset_utils import build_dataset, get_mask\n",
    "import torch.nn.functional as F\n",
    "import torch    \n",
    "\n",
    "from layer import CompatibilityLayer\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "quality-tuning",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "For dataset Chameleon\n",
      "2277 nodes, 36101 edges\n"
     ]
    }
   ],
   "source": [
    "dataset_name = 'Chameleon'\n",
    "\n",
    "dataset = build_dataset(dataset_name, to_cuda=False)\n",
    "raw_adj = dataset['raw_adj'].numpy()\n",
    "y = dataset['labels'].numpy()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "fabulous-schema",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "For dataset Chameleon\n",
      "2277 nodes, 36101 edges\n"
     ]
    }
   ],
   "source": [
    "# ours cheb\n",
    "model_path = f'{dataset_name}_model.pth' # main branch\n",
    "\n",
    "model = torch.load(model_path,map_location=lambda storage, loc: storage)\n",
    "train_mask = dataset['train_masks'][0]\n",
    "\n",
    "model.eval()\n",
    "\n",
    "with torch.no_grad():\n",
    "    logits, node_vec, first_adj = model.forward_one(train_mask)\n",
    "    # first_adj = cur_normed_adj\n",
    "    last_logits = logits.detach()\n",
    "\n",
    "    for _ in range(10):\n",
    "        logits, node_vec = model.forward_two(node_vec, last_logits, train_mask, first_adj)\n",
    "\n",
    "    pred = logits\n",
    "    \n",
    "our_pred = pred.max(1)[1].numpy()\n",
    "\n",
    "# ours = draw(y, pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "dangerous-accordance",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "For dataset Chameleon\n",
      "2277 nodes, 36101 edges\n"
     ]
    }
   ],
   "source": [
    "# gcn\n",
    "model_path = 'Chameleon_baseline_gcn_model.pth'\n",
    "model = torch.load(model_path,map_location=lambda storage, loc: storage)\n",
    "dataset = build_dataset(dataset_name, to_cuda=False)\n",
    "raw_dataset = dataset['raw_dataset']\n",
    "data = raw_dataset[0]\n",
    "pred = model(data)\n",
    "gcn_pred = pred.max(1)[1].numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "finnish-sleeve",
   "metadata": {},
   "outputs": [],
   "source": [
    "# idgl\n",
    "# needs to checkout idgl branch\n",
    "model_path = 'Chameleon_idgl_model.pth'\n",
    "model = torch.load(model_path,map_location=lambda storage, loc: storage)\n",
    "model.eval()\n",
    "   \n",
    "with torch.no_grad():\n",
    "    logits, node_vec, cur_raw_adj, cur_normed_adj = model.forward_one()\n",
    "    first_adj = cur_normed_adj\n",
    "\n",
    "    for _ in range(10):\n",
    "        logits, node_vec, cur_raw_adj, cur_normed_adj = model.forward_two(node_vec, first_adj)\n",
    "\n",
    "    pred = logits\n",
    "            \n",
    "idgl_pred = pred.max(1)[1].numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "id": "green-animation",
   "metadata": {},
   "outputs": [],
   "source": [
    "def draw(y, pred, n_bin):\n",
    "    neis = dict()\n",
    "    n_ys = dict()\n",
    "    n = y.shape[0]\n",
    "    hs = [0 for i in range(n)]\n",
    "#     hs = []\n",
    "    \n",
    "    for i in range(n):\n",
    "        neis[i] = raw_adj[i].nonzero()[0]\n",
    "        n_ys[i] = y[neis[i]]\n",
    "        hs[i] = (n_ys[i] == y[i]).sum()/len(neis[i])\n",
    "\n",
    "    bins = plt.hist(hs, bins=n_bins)\n",
    "    plt.close()\n",
    "    cnts = bins[0]\n",
    "    edges = bins[1]\n",
    "\n",
    "    acc_cnts = [0 for i in range(len(cnts))]\n",
    "    for i in range(n):\n",
    "        neis[i] = raw_adj[i].nonzero()[0]\n",
    "        if len(neis[i]) <= 1:\n",
    "            continue\n",
    "        is_true = (pred[i] == y[i])\n",
    "        for j in range(len(cnts)):\n",
    "            if hs[k]>=edges[j] and hs[k] < edges[j+1]:\n",
    "                acc_cnts[j] += is_true\n",
    "\n",
    "    accs = np.array(acc_cnts) / cnts\n",
    "    \n",
    "\n",
    "#     plt.hist(hs, bins=n_bins, density=True)\n",
    "#     plt.plot((edges[:-1]+edges[1:])/2, accs*7)\n",
    "    return edges, hs, accs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "id": "flexible-scanner",
   "metadata": {},
   "outputs": [],
   "source": [
    "n_bins = 10\n",
    "ours = draw(y, our_pred, n_bins)\n",
    "gcns = draw(y, gcn_pred, n_bins)\n",
    "idgls = draw(y, idgl_pred, n_bins)\n",
    "edges, hs = ours[0], ours[1]\n",
    "# edges = np.array(edges)\n",
    "def mask_nan_to_one(x):\n",
    "    x[np.isnan(x)] = 1\n",
    "    return x\n",
    "# our_acc, gcn_acc, idgl_acc = mask_nan_to_one(ours[-1]), mask_nan_to_one(gcns[-1]), mask_nan_to_one(idgls[-1])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "id": "acquired-paris",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def smooth(x, k=3):\n",
    "    n = len(x)\n",
    "    y = np.zeros(len(x))\n",
    "    y[:n-k+1] += x[:n-k+1]\n",
    "    for i in range(k-1):\n",
    "        y[:n-k+1] += x[i+1:n-k+i+2]\n",
    "    y /= k\n",
    "\n",
    "    y[n-k+1:] += x[n-k+1:]\n",
    "    return y\n",
    "\n",
    "fig, ax1 = plt.subplots()\n",
    "new_hs = [x for x in hs if x < 0.9]\n",
    "\n",
    "ax1.set_ylim(0, 800)\n",
    "ax1.hist(hs, range=(0, 0.9), bins=n_bins-1)\n",
    "\n",
    "ax2 = plt.twinx(ax1)\n",
    "k=1\n",
    "ax2.set_ylim(0, 1.0)\n",
    "ax2.plot(((edges[:-1]+edges[1:])/2)[:-1], smooth(our_acc, k)[:-1], 'r', label='LGS', marker='s')\n",
    "ax2.plot(((edges[:-1]+edges[1:])/2)[:-1],  smooth(gcn_acc, k)[:-1], c='#fca311', label='GCN', marker='^', markersize=6)\n",
    "ax2.plot(((edges[:-1]+edges[1:])/2)[:-1],  smooth(idgl_acc, k)[:-1], 'bo-', label='IDGL')\n",
    "# plt.plot((edges[:-1]+edges[1:])/2, smooth(our_gcn_acc, k), 'b', label='Ours_gcn')\n",
    "ax1.set_xlabel('Homophily Ratio')\n",
    "ax1.set_ylabel('# node')\n",
    "ax2.set_ylabel('Accuracy')\n",
    "\n",
    "plt.legend(loc='best')\n",
    "# plt.show()\n",
    "plt.savefig('acc.pdf')\n",
    "# plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "introductory-validation",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.0, 1.0)"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ax2.set_ylim(0, 1.0)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}