notebooks/robustness.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ac0f944f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append(\"../\")\n",
    "import re\n",
    "from os.path import join\n",
    "from IPython.display import display\n",
    "import os\n",
    "import inspect\n",
    "\n",
    "import matplotlib\n",
    "matplotlib.rcParams['pdf.fonttype'] = 42\n",
    "matplotlib.rcParams['ps.fonttype'] = 42\n",
    "matplotlib.rc('text', usetex=False)\n",
    "\n",
    "import scipy\n",
    "import faiss\n",
    "import matplotlib.pyplot as plt\n",
    "import joblib\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import torch\n",
    "from tqdm.notebook import tqdm\n",
    "\n",
    "from lolip.models.torch_utils.archs import *\n",
    "from lolip.variables import auto_var\n",
    "from lolip.tools.nn_utils import get_faiss_index\n",
    "from params import *\n",
    "from utils import params_to_dataframe\n",
    "from notebook_utils import get_model_name, get_arch, get_ds_name\n",
    "\n",
    "fontsize=18\n",
    "\n",
    "# for auto-reloading external modules \n",
    "# see http://stackoverflow.com/questions/1907993/autoreload-of-modules-in-ipython \n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc91e4b2",
   "metadata": {},
   "source": [
    "# new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5caca76c",
   "metadata": {},
   "outputs": [],
   "source": [
    "from lolip.models.torch_utils import archs\n",
    "\n",
    "def get_dataset_info(dataset):\n",
    "    if dataset not in dataset_infos:\n",
    "        trnX, trny, tstX, _, ood = auto_var.get_var_with_argument(\"dataset\", dataset)\n",
    "        oodX = np.concatenate((ood[0], ood[1]), axis=0)\n",
    "        \n",
    "        index = get_faiss_index(int(np.prod(trnX.shape[1:])), 2)\n",
    "        \n",
    "        if (\"calced\" not in dataset) and (\"aug10-imgnet\" in dataset):\n",
    "            from torchvision import transforms\n",
    "            normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],\n",
    "                                    std=[0.229, 0.224, 0.225])\n",
    "            def np_normalize(X):\n",
    "                X = X.transpose(0, 3, 1, 2)\n",
    "                return normalize(torch.from_numpy(X)).numpy().transpose(0, 2, 3, 1)\n",
    "            index.add(np_normalize(trnX).astype(np.float32).reshape(len(trnX), -1))\n",
    "            D, I = index.search(np_normalize(oodX).astype(np.float32).reshape(len(oodX), -1), k=1)\n",
    "        else:\n",
    "            index.add(trnX.astype(np.float32).reshape(len(trnX), -1))\n",
    "            D, I = index.search(oodX.astype(np.float32).reshape(len(oodX), -1), k=1)\n",
    "        D, I = D.reshape(-1), I.reshape(-1)\n",
    "        D = np.sqrt(D)\n",
    "        oody = trny[I]\n",
    "        dataset_infos[dataset] = (trnX, trny, tstX, oodX, I, D, oody)\n",
    "    return dataset_infos[dataset]\n",
    "\n",
    "def get_preds(model, dset, batch_size=64, device=\"cuda\", verbose=False):\n",
    "    model.eval().to(device)\n",
    "    if isinstance(dset, np.ndarray):\n",
    "        if len(dset.shape) == 4:\n",
    "            dset = dset.transpose(0, 3, 1, 2)\n",
    "        dset = torch.utils.data.TensorDataset(\n",
    "            #torch.from_numpy(dset).float(),\n",
    "            torch.from_numpy(dset),\n",
    "            torch.ones(len(dset))\n",
    "        )\n",
    "\n",
    "    loader = torch.utils.data.DataLoader(dset, batch_size=batch_size, shuffle=False, num_workers=24)\n",
    "    \n",
    "    ret = []\n",
    "    for (x, _) in tqdm(loader, disable=(not verbose)):\n",
    "        #x.requires_grad_(False)\n",
    "        #output = F.softmax(model(x.to(device)).detach(), dim=1)\n",
    "        with torch.no_grad():\n",
    "            output = F.softmax(model(x.to(device)), dim=1)\n",
    "        ret.append(output.cpu().numpy())\n",
    "    del loader\n",
    "    return np.concatenate(ret, axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "249efc4e",
   "metadata": {},
   "outputs": [],
   "source": [
    "results = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "f2955d79",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_infos = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "d5368d3f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_hist(adv_dists, ood_dists, figure_name, show=True, with_legend=True):\n",
    "    ttt = np.log(adv_dists)\n",
    "    if abs(ttt.max() - ttt.min()) < 0.1:\n",
    "        y = [len(adv_dists)]\n",
    "        x = [(ttt.max() - ttt.min()) / 2]\n",
    "        matplotlib.pyplot.bar(x, y, width=0.1, color=\"#ff7f0e\")\n",
    "    else:\n",
    "        bins = 20\n",
    "        plt.hist(ttt, bins=bins, label=\"empirical robust radius\")\n",
    "    ttt = np.log(ood_dists)\n",
    "    plt.hist(ttt, bins=20, label=\"OOD distance\")\n",
    "    \n",
    "\n",
    "    plt.yticks(fontsize=fontsize)\n",
    "    #plt.yticks([], [])\n",
    "    plt.ylabel('counts', fontsize=fontsize)\n",
    "    plt.xlabel('natural log $\\ell_2$ distance', fontsize=fontsize)\n",
    "    plt.xticks(fontsize=fontsize-2)\n",
    "    if with_legend:\n",
    "        plt.legend(fontsize=(fontsize-2))\n",
    "    plt.tight_layout()\n",
    "    \n",
    "    plt.savefig(f\"./figs/rob_histograms/{figure_name}\")\n",
    "    if show:\n",
    "        plt.show()\n",
    "    plt.close()\n",
    "\n",
    "def get_rob_results(dataset, dname, method_name, model_path, result_path, ood_result_paths, arch, with_legend=True):\n",
    "    res = joblib.load(result_path)\n",
    "    trnX, trny, tstX, oodX, I, D, oody = get_dataset_info(dataset)\n",
    "    n_classes = len(np.unique(trny))\n",
    "    \n",
    "    arch_fn = getattr(archs, arch)\n",
    "    n_channels = 1 if ('mnist' in dataset) else 3\n",
    "    arch_params = dict(n_classes=n_classes, n_channels=n_channels)\n",
    "    if 'n_features' in inspect.getfullargspec(arch_fn)[0]:\n",
    "        arch_params['n_features'] = (trnX.shape[1], )\n",
    "    model = arch_fn(**arch_params).eval()\n",
    "    model.load_state_dict(torch.load(model_path)['model_state_dict'])\n",
    "\n",
    "    ood_pred = np.concatenate((res['oos_trn_pred'].reshape(-1), res['oos_tst_pred'].reshape(-1)))\n",
    "    ncg_accs = (ood_pred == oody)\n",
    "\n",
    "    for i, ood_robust_path in enumerate(ood_result_paths):\n",
    "        if not os.path.exists(ood_robust_path):\n",
    "            print(f\"Missing {ood_robust_path}\")\n",
    "            continue\n",
    "        try:\n",
    "            rob_res = joblib.load(ood_robust_path)\n",
    "        except:\n",
    "            print(f\"Missing {ood_robust_path}\")\n",
    "            continue\n",
    "        trn_idx = rob_res['trn_idx']\n",
    "        adv_trnX = rob_res['adv_trnX']\n",
    "        \n",
    "        if i == 0:\n",
    "            n_samples = len(trn_idx)\n",
    "            adv_dists = np.ones(n_samples) * np.inf\n",
    "            failed_attack_ind = np.ones(n_samples)\n",
    "            \n",
    "        temp = np.linalg.norm((adv_trnX - trnX[trn_idx]).reshape(len(trn_idx), -1), ord=2, axis=1)\n",
    "        preds = get_preds(model, trnX[trn_idx]).argmax(1)\n",
    "        #if (preds != trny[trn_idx]).sum() != 0:\n",
    "        #    import ipdb; ipdb.set_trace()\n",
    "        assert (preds != trny[trn_idx]).sum() < 10, (preds == trny[trn_idx]).sum()\n",
    "        # numarical errors could cause prediction to change after loading the saved model ....\n",
    "        #failed_temp = np.logical_or((preds == trny[trn_idx]), temp == 0)\n",
    "        \n",
    "        if 'successes' in rob_res:\n",
    "            failed_temp = np.logical_not(rob_res['successes'])\n",
    "        elif 'adv_preds' in rob_res:\n",
    "            preds = rob_res['adv_preds']\n",
    "            failed_temp = (preds == trny[trn_idx])\n",
    "        else:\n",
    "            preds = get_preds(model, adv_trnX).argmax(1)\n",
    "            failed_temp = (preds == trny[trn_idx])\n",
    "        print(failed_temp.mean())\n",
    "        \n",
    "        adv_dists[np.logical_not(failed_temp)] = np.minimum(adv_dists[np.logical_not(failed_temp)], temp[np.logical_not(failed_temp)])\n",
    "        failed_attack_ind = np.logical_and(failed_temp, failed_attack_ind)\n",
    "            \n",
    "    ood_dists = []\n",
    "    ood_counts = []\n",
    "    ood_within_counts = []\n",
    "    ood_nn_idx = []\n",
    "    within_list = []\n",
    "    ncg_correctness = []\n",
    "    first = True\n",
    "    for k, i in enumerate(trn_idx):\n",
    "        ind = np.where(I == i)[0]\n",
    "        if failed_attack_ind[k] == 1:\n",
    "            print(\"Rare\")\n",
    "            continue\n",
    "        for j in ind:\n",
    "            ood_dists.append(D[j] + 1e-5)\n",
    "        if len(ind) > 0:\n",
    "            ood_counts.append(len(ind))\n",
    "            ood_within_counts.append((D[ind] < adv_dists[k]).sum())\n",
    "            ood_nn_idx.append(ind)\n",
    "            within_list.append((D[ind] < adv_dists[k]))\n",
    "            ncg_correctness.append(trny[I[ind]] == ood_pred[ind])\n",
    "        else:\n",
    "            print(\"shouldn't happen\")\n",
    "            \n",
    "    ood_nn_idx = np.concatenate(ood_nn_idx)\n",
    "    within_list = np.concatenate(within_list)\n",
    "    ncg_correctness = np.concatenate(ncg_correctness)\n",
    "    in_rob_r_portion = np.sum(ood_within_counts) / np.sum(ood_counts)\n",
    "    \n",
    "    if with_legend:\n",
    "        figure_name = f\"{dname}_{method_name}.png\"\n",
    "    else:\n",
    "        figure_name = f\"{dname}_{method_name}_nolegend.png\"\n",
    "    plot_hist(adv_dists[~np.isinf(adv_dists)], ood_dists, figure_name, show=True, with_legend=with_legend)\n",
    "    \n",
    "    return adv_dists[~np.isinf(adv_dists)], ood_dists, in_rob_r_portion, ncg_accs.mean(), ood_nn_idx, within_list, ncg_correctness"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "c48c7430",
   "metadata": {
    "scrolled": true,
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('calcedrepr-mnistwo4', 'TRADES(4)')\n",
      "0.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "0.0\n",
      "0.0\n",
      "0.63\n",
      "0.96\n",
      "1.0\n",
      "0.04\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "0.0\n",
      "0.0\n",
      "1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('calcedrepr-mnistwo4', 'TRADES(8)')\n",
      "0.0\n",
      "1.0\n",
      "0.9966666666666667\n",
      "1.0\n",
      "0.0\n",
      "0.006666666666666667\n",
      "0.8333333333333334\n",
      "0.99\n",
      "0.9966666666666667\n",
      "0.39666666666666667\n",
      "0.99\n",
      "0.99\n",
      "0.99\n",
      "1.0\n",
      "1.0\n",
      "0.99\n",
      "0.0\n",
      "0.0\n",
      "0.99\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('calcedrepr-mnistwo9', 'TRADES(4)')\n",
      "0.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "0.0\n",
      "0.0\n",
      "0.6133333333333333\n",
      "0.9733333333333334\n",
      "1.0\n",
      "0.09333333333333334\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "0.9966666666666667\n",
      "0.0\n",
      "0.0\n",
      "0.9966666666666667\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('calcedrepr-mnistwo9', 'TRADES(8)')\n",
      "0.0\n",
      "1.0\n",
      "0.9966666666666667\n",
      "1.0\n",
      "0.0\n",
      "0.016666666666666666\n",
      "0.7666666666666667\n",
      "0.9966666666666667\n",
      "0.9966666666666667\n",
      "0.39666666666666667\n",
      "0.9966666666666667\n",
      "0.9966666666666667\n",
      "0.9966666666666667\n",
      "1.0\n",
      "1.0\n",
      "0.99\n",
      "0.0\n",
      "0.0\n",
      "0.99\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "all_settings = []\n",
    "\n",
    "method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(1)\", 'advce', 1.0)]\n",
    "\n",
    "#for i in [0, 4, 9]:\n",
    "#    for method_name, loss, eps in method_names:\n",
    "#        arch = \"LargeMLP\"\n",
    "#        all_settings.append(\n",
    "#            dict(\n",
    "#                dname = f\"calcedrepr-cifar10wo{i}\",\n",
    "#                dataset = f\"calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl\",\n",
    "#                model_path = f\"../models/out_of_sample/pgd-128-calcedrepr-cifar10wo{i}-pgd-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0-ep0070.pt\",\n",
    "#                result_path = f\"../results/out_of_sample/cwl2-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                ood_result_paths = [\n",
    "#                    f\"../results/ood_robustness_correct/signopt-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2_10000_1.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdlinf_10000_1.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_8.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_3.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_2.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_25000_2.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_25000_1.5-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbl2bba-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2_10000_2.0-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/cwl2-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/pgd_1000-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/mtv2-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/mtv2_50-128-calcedrepr-cifar10wo{i}-cwl2-64-cifar10wo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                ],\n",
    "#                method_name = method_name,\n",
    "#                arch = arch,\n",
    "#            )\n",
    "#        )\n",
    "#        \n",
    "##for i in [0, 4, 9]:\n",
    "#for i in [9]:\n",
    "#    for method_name, loss, eps in method_names:\n",
    "#        arch = \"LargeMLP\"\n",
    "#        all_settings.append(\n",
    "#            dict(\n",
    "#                dname = f\"calcedrepr-cifar100coarsewo{i}\",\n",
    "#                dataset = f\"calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl\",\n",
    "#                model_path = f\"../models/out_of_sample/pgd-128-calcedrepr-cifar100coarsewo{i}-pgd-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0-ep0070.pt\",\n",
    "#                result_path = f\"../results/out_of_sample/cwl2-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                ood_result_paths = [\n",
    "#                    f\"../results/ood_robustness_correct/signopt-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2_10000_1.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdlinf_10000_1.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_8.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_3.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000_2.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_50000-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_25000_2.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary_25000_1.5-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbboundary-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbl2bba-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/fbpgdl2_10000_2.0-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/cwl2-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/pgd_1000-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/mtv2-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                    f\"../results/ood_robustness_correct/mtv2_50-128-calcedrepr-cifar100coarsewo{i}-cwl2-64-cifar100coarsewo{i}-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "#                ],\n",
    "#                method_name = method_name,\n",
    "#                arch = arch,\n",
    "#            )\n",
    "#        )\n",
    "#        \n",
    "#method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(.5)\", 'advce', 0.5)]\n",
    "method_names = [(\"AT(.5)\", 'advce', 0.5)]\n",
    "#for i in [0, 1, 2]:\n",
    "for i in [0, ]:\n",
    "    arch = \"LargeMLPv4\"\n",
    "    for method_name, loss, eps in method_names:\n",
    "        lr = 0.01 if method_name == \"TRADES(2)\" else 0.1\n",
    "        all_settings.append(\n",
    "            dict(\n",
    "                dname = f\"calcedrepr-aug10-imgnet100wo{i}\",\n",
    "                dataset = f\"calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl\",\n",
    "                model_path = f\"../models/out_of_sample/pgd-128-calcedrepr-aug10-imgnet100wo{i}-pgd-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0-ep0070.pt\",\n",
    "                result_path = f\"../results/out_of_sample/cwl2-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ood_result_paths = [\n",
    "                    f\"../results/ood_robustness_correct/cwl2-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/signopt-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_25000_1.0-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_50000-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/pgd_1000-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/mtv2_50-128-calcedrepr-aug10-imgnet100wo{i}-cwl2-128-aug10-imgnet100wo{i}-70-1.0-0.01-ce-vtor2-ResNet50Norm01-0.0-2-adam-0-0.0.pkl-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ],\n",
    "                method_name = method_name,\n",
    "                arch = arch,\n",
    "            )\n",
    "        )\n",
    "        \n",
    "method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(2)\", 'advce', 2.0)]\n",
    "method_names = [(\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0)]\n",
    "#for i in [0, 1, 4, 9]:\n",
    "for i in [4, 9]:\n",
    "    for method_name, loss, eps in method_names:\n",
    "        all_settings.append(\n",
    "            dict(\n",
    "                dname = f\"calcedrepr-mnistwo{i}\",\n",
    "                dataset = f\"calcedreprold-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl\",\n",
    "                model_path = f\"../models/out_of_sample/pgd-256-calcedrepr-mnistwo{i}-pgd-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0-ep0070.pt\",\n",
    "                result_path = f\"../results/out_of_sample/cwl2-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ood_result_paths = [\n",
    "                    f\"../results/ood_robustness_correct/cwl2-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/signoptrand-256-calcedreprold-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/signopt-256-calcedreprold-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/signopt-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2_10000_64.0-256-calcedreprold-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2_10000_32.0-256-calcedreprold-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_50000_16.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_50000_8.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_50000_4.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdlinf_10000_1.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary_50000-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbboundary-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbl2bba-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2_10000_1.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2_10000_2.0-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/mtv2_50-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/mtv2-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/pgd_1000-256-calcedrepr-mnistwo{i}-cwl2-128-mnistwo{i}-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-{eps}-0.01-{loss}-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ],\n",
    "                method_name = method_name,\n",
    "                arch = \"LargeMLP\",\n",
    "            )\n",
    "        )\n",
    "\n",
    "        \n",
    "# fraction of time the label goes from cat to dog\n",
    "\n",
    "for ent in all_settings:\n",
    "    #dataset, dname, method_name, model_path, result_path, ood_robust_paths, arch = ent['dataset'], ent['dname'], ent['method_name'], ent['model_path'], ent['result_path'], ent['ood_robust_paths'], ent['arch']\n",
    "    #adv_dists, ood_dists, in_rob_r_portion = get_rob_results(dataset, dname, method_name, model_path, result_path, ood_robust_paths, arch)\n",
    "    key = (ent['dname'], ent['method_name'])\n",
    "    print(key)\n",
    "    if key not in results:\n",
    "        adv_dists, ood_dists, in_rob_r_portion, ncg_acc, ood_nn_idx, within_list, ncg_correctness = get_rob_results(**ent)\n",
    "        results[key] = [np.mean(adv_dists), np.mean(ood_dists), in_rob_r_portion, ncg_acc, ood_nn_idx, within_list, ncg_correctness]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "34dfae32",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = {}\n",
    "for k, v in results.items():\n",
    "    ood_nn_idx, within_list, ncg_correctness = v[4], v[5], v[6]\n",
    "    data[k] = [v[0], v[1], v[2], v[3], np.logical_and(within_list, ncg_correctness).mean()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "aab52612",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>avg adv dist.</th>\n",
       "      <th>avg OOD dist.</th>\n",
       "      <th>XD</th>\n",
       "      <th>ncg acc</th>\n",
       "      <th>in radius portion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">calcedrepr-mnistwo4</th>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>16.436266</td>\n",
       "      <td>54.225028</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.814918</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>20.389073</td>\n",
       "      <td>53.803331</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.859613</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">calcedrepr-mnistwo9</th>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>16.430955</td>\n",
       "      <td>50.695964</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.741880</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>20.465122</td>\n",
       "      <td>51.481997</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.798362</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                               avg adv dist.  avg OOD dist.   XD   ncg acc  \\\n",
       "calcedrepr-mnistwo4 TRADES(4)      16.436266      54.225028  0.0  0.814918   \n",
       "                    TRADES(8)      20.389073      53.803331  0.0  0.859613   \n",
       "calcedrepr-mnistwo9 TRADES(4)      16.430955      50.695964  0.0  0.741880   \n",
       "                    TRADES(8)      20.465122      51.481997  0.0  0.798362   \n",
       "\n",
       "                               in radius portion  \n",
       "calcedrepr-mnistwo4 TRADES(4)                0.0  \n",
       "                    TRADES(8)                0.0  \n",
       "calcedrepr-mnistwo9 TRADES(4)                0.0  \n",
       "                    TRADES(8)                0.0  "
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame.from_dict(data, orient=\"index\",\n",
    "                            columns=[\"avg adv dist.\", \"avg OOD dist.\", \"XD\", \"ncg acc\", \"in radius portion\"])\n",
    "df = df.sort_index()\n",
    "df.index = pd.MultiIndex.from_tuples(df.index)\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "b38101c2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\begin{tabular}{llrrrr}\n",
      "\\toprule\n",
      "                    &           &  empirical robust radius &  OOD dist. &  percentage covered &  NCG acc. \\\\\n",
      "\\midrule\n",
      "\\multirow{2}{*}{M-wo4} & TRADES(4) &          16.44 &          54.23 &               0.00 &     0.81 \\\\\n",
      "                    & TRADES(8) &          20.39 &          53.80 &               0.00 &     0.86 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{2}{*}{M-wo9} & TRADES(4) &          16.43 &          50.70 &               0.00 &     0.74 \\\\\n",
      "                    & TRADES(8) &          20.47 &          51.48 &               0.00 &     0.80 \\\\\n",
      "\\bottomrule\n",
      "\\end{tabular}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "text = df[[\"avg adv dist.\", \"avg OOD dist.\", \"in radius portion\", \"ncg acc\",]].to_latex(float_format=\"%.2f\", multirow=True)\n",
    "text = text.replace(\"avg adv dist.\", \"empirical robust radius\")\n",
    "text = text.replace(\"avg OOD dist.\", \"OOD dist.\")\n",
    "text = text.replace(\"in radius portion\", \"percentage covered\")\n",
    "text = text.replace(\"ncg acc\", \"NCG acc.\")\n",
    "text = text.replace(\"calcedrepr-aug10-imgnet100wo\", \"I-\")\n",
    "text = text.replace(\"calcedrepr-cifar100coarsewo\", \"C100-\")\n",
    "text = text.replace(\"calcedrepr-cifar10wo\", \"C10-\")\n",
    "text = text.replace(\"calcedrepr-mnist\", \"M-\")\n",
    "print(text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "41f72a22",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([30979, 17950, 10570, ...,  6101, 28907,  9213])"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_infos['calcedrepr-cifar100coarsewo4-cwl2-64-cifar100coarsewo4-70-1.0-0.01-ce-vtor2-WRN_40_10-0.0-2-adam-0-0.0.pkl'][-3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "id": "1976811d",
   "metadata": {},
   "outputs": [],
   "source": [
    "i = 9\n",
    "loss = \"advce\"\n",
    "eps = \"1.0\"\n",
    "arch = \"LargeMLP\"\n",
    "path = f\"../results/ood_robustness_correct/signopt-256-calcedrepr-mnistwo0-cwl2-128-mnistwo0-70-1.0-0.01-ce-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl-70-1.0-0.01-ce-vtor2-LargeMLP-0.9-2-sgd-0-0.0.pkl\"\n",
    "res = joblib.load(path)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "415aa1c8",
   "metadata": {},
   "source": [
    "## Pixel space"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ba21c580",
   "metadata": {
    "scrolled": true,
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('aug10-imgnet100wo0', 'natural')\n",
      "('aug10-imgnet100wo0', 'TRADES(2)')\n",
      "('aug10-imgnet100wo0', 'TRADES(4)')\n",
      "('aug10-imgnet100wo0', 'TRADES(8)')\n",
      "('aug10-imgnet100wo0', 'AT(2)')\n",
      "('aug10-imgnet100wo1', 'natural')\n",
      "('aug10-imgnet100wo1', 'TRADES(2)')\n",
      "('aug10-imgnet100wo1', 'TRADES(4)')\n",
      "('aug10-imgnet100wo1', 'TRADES(8)')\n",
      "('aug10-imgnet100wo1', 'AT(2)')\n",
      "('aug10-imgnet100wo2', 'natural')\n",
      "('aug10-imgnet100wo2', 'TRADES(2)')\n",
      "('aug10-imgnet100wo2', 'TRADES(4)')\n",
      "('aug10-imgnet100wo2', 'TRADES(8)')\n",
      "('aug10-imgnet100wo2', 'AT(2)')\n",
      "('mnistwo0', 'natural')\n",
      "('mnistwo0', 'TRADES(2)')\n",
      "('mnistwo0', 'TRADES(4)')\n",
      "('mnistwo0', 'TRADES(8)')\n",
      "('mnistwo0', 'AT(2)')\n",
      "('mnistwo1', 'natural')\n",
      "('mnistwo1', 'TRADES(2)')\n",
      "('mnistwo1', 'TRADES(4)')\n",
      "('mnistwo1', 'TRADES(8)')\n",
      "('mnistwo1', 'AT(2)')\n",
      "('mnistwo4', 'natural')\n",
      "('mnistwo4', 'TRADES(2)')\n",
      "('mnistwo4', 'TRADES(4)')\n",
      "('mnistwo4', 'TRADES(8)')\n",
      "('mnistwo4', 'AT(2)')\n",
      "0.11\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEYCAYAAAAJeGK1AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAAAgR0lEQVR4nO3debgcVZnH8e8vLIkRZDFRIRAuggIyKiogqCCyCqiIyiiIEhBQFBRQERUxLCO4ggou4ALKwAiyZkAQBqI8BISgIAYkBAgMEZXIFohJWN7545weO52+93b1VnVvfp/n6adunzpV/VZX9337nDpVpYjAzMysasaUHYCZmVkzTlBmZlZJTlBmZlZJTlBmZlZJTlBmZlZJK5YdQD9NmDAhBgYGyg7DzMzq3HrrrfMjYmJj+XKVoAYGBpg5c2bZYZiZWR1JDzQrdxefmZlVkhOUmZlVkhOUmZlVkhOUmZlVkhOUmZlVkhOUmZlVkhOUmZlVkhOUmZlVkhOUmZlV0nJ1JQkzs0qbulqTsif6H0dFuAVlZmaV5ARlZmaV5ARlZmaV5GNQZma90uyYEizXx5WKcAvKzMwqyQnKzMwqyV18Zmb9NljXny3FLSgzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6skJygzM6uk0hOUpFUkfUHSHZIWSJovaYakKZLUUPeNkq7J9Z6UdKWkzUoK3czMeqjUq5lLGgP8CngTcDbwXWA8sDfwU2AT4HO57lbAdGAecGxexaHA9ZLeFBF39DV4MzPrqbJvt/FG4C3AqRFxRK1Q0veAPwMfJSco4DvAEmDbiJiX650P3AV8E9i5j3GbmVmPld3F96I8/Ut9YUQsAeYDTwNI2hDYAriglpxyvXnABcCOkl7Wl4jNzKwvyk5QNwOPA0dJ2kvSZEkbSzoJeAMwNdfbIk9vbLKOmwDl+mZmNkqU2sUXEY9JehfwI+D8ulkLgPdGxCX5+dp5Oo9l1comNXsNSQcDBwNMnjy505DNzKxPym5BATwF/An4BvAe4EBgDnCupJ1ynfF5urjJ8osa6iwlIs6IiM0jYvOJEyd2L2ozM+upskfxvRqYARwRET+oKz+PlLTOlLQBsDDPGttkNePydGGTeWZmNkKV3YI6gpRgLqgvjIiFwOXAesAA/xpE0awbr1bWrPvPzMxGqLITVC25rNBk3op101vy31s3qbcVEMCt3Q3NzMzKVHaCujNPp9QXSlod2AN4DJgTEXOAmcBektauq7c2sBdwbUT8tR8Bm5lZf5R9ou6pwIeBk/PxqBuANYGDgLWAT0TEc7nup4DrSFeO+G4uO4yUZD/dz6DNzKz3yh5m/oCkLUmXLtoB+ADwT+A24NMRcVFd3RmStgNOzI8gDbDYKyJu72/kZmbWa2W3oIiIe4H9Wqx7IymRmZnZKFf2MSgzM7OmnKDMzKySnKDMzKySnKDMzKySnKDMzKySnKDMzKySnKDMzKySSj8PysxsVJi6WtkRjDpuQZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSVVIkFJWlPSNyTNkbRI0iOSrpO0TUO9N0q6RtICSU9KulLSZiWFbWZmPbRi2QFIWg+YDqwC/BiYDawGvAaYVFdvq1xvHnBsLj4UuF7SmyLijv5FbWZmvVZ6ggLOIcXxmoh4eIh63wGWANtGxDwASecDdwHfBHbudaBmZtY/pSYoSdsCbwE+GREPS1oJWCkiFjbU2xDYAvhJLTkBRMQ8SRcA+0t6WUT8tZ/xm40WA0dfXqj+3JN371EkZv9S9jGo3fL0QUnTgH8CT0uaLWnfunpb5OmNTdZxEyDgDb0L08zM+q3sBLVRnp4JrAnsBxxA6sr7uaT98/y183Qey6qVTWoyz8zMRqiudfFJGhsRiwsutmqeLgDeFhFL8rouAe4DviLpbGB8rtds/YvydHyTeUg6GDgYYPLkyQXDMzOzshRqQUnaVdLUhrKPS3qS1DV3bj6O1Kp/5ul5teQEEBGPAZcBLyO1smrHpMY2Wce4PF3YZB4RcUZEbB4Rm0+cOLFAaGZmVqaiLajPAn+vPZG0CfBt4F7gfuD9wM3AqS2u76E8bTa4oTaibw3gL/nvZt14tbJm3X9mZqPT1NUGKX+iv3H0UNFjUJsAM+uev5/UCtoyInYFfkE6jtSqm/N0nSbzamV/B27Jf2/dpN5WQAC3FnhdMzOruKIJag1gft3zHYFrI+LJ/Hw6sH6B9V1COv60r6RVaoWS1gLeDcyOiDkRMYeUGPeStHZdvbWBvXIMHmJuZjaKFE1Q84H1ACStShr+fX3d/JWAFVpdWT7W9BlSN91Nko6UdDRp6PjKwGF11T9FOgZ1vaTDJR2eX3sM8OmC22FmZhVX9BjUjcDHJM0Cds3L/6pu/ob869hRSyLiDEnzgaOAE4Dn8+vsExE31NWbIWk74MT8CGAGsFdE3F5wO8zMrOKKJqgvA9cB5+fnZ0fEnQCSBOyZ5xcSERcBF7VQ70Zgh6LrNzOzkadQgoqIO/PIvTcDT0TEb+tmrw6cQhsJyszMrFHR86C2BVaIiGkNyal2POlcUqIyMzPrSNFBEtcBOw0xf3vcgjIzsy4omqA0zPwVSIMczMzMOtLOxWJjiHlvYunzpMzMzNoy7CAJSZ8inYNUc6qk/2hSdQ3gRcBPuhSbmZktx1oZxfc48ED+ewD4B/C3hjoB/Il0gu0pXYrNzMyWY8MmqIg4GzgbQNL9wNERcVmvAzMzs+Vb0fOgilxnz8zMrG1t37BQ0njgxTQZ2RcRD3YSlJmZWaEEJWkM6Zp5h5FuJjiYli8Ya2Y2ogx2H6bR8noVUrQFdTLp6uOzgAtJAybMzMy6rmiC2he4MiJ260UwZmZmNe3csPDSXgRiZmZWr2iCugNYqxeBmJmZ1SuaoI4j3bBw3V4EY2ZmVlP0GNQbSFeVuFPSxcD9wHMNdSIiTuhGcGZmtvwqmqCm1v297yB1gnTrdjMrycDRl5cdglnHiiYoX0nCzMz6ouiljh4YvpaZmVnn2rkflJmZWc8VvdRRK/d6ioj4SJvxmJmZAcWPQU1poU4ATlBmZtaRQl18ETGm8QGsBGwEnEm6YeEaPYjTzMyWMx0fg4qI5yLinoj4KOnisV/tPCwzM1vedXuQxJXAe7u8TjMzWw51O0GtCazS5XWamdlyqO076taTtDqwI3AEcGs31mlmZsu3osPMnyeN0ms6G3gUOLLToMzMzIq2oH7GsgkqSIlpNnBeRCzoRmBmZrZ8K3qpoyk9isPMzGwpvtSRmZlVUuEEJWmMpP0lXSbpT/lxmaQpkjpKeJLGS7pPUkg6rcn8jSRdIukxSU9Lul7S9p28ppmZVVPRQRIvAK4AtiUde3o4z9oN2B34sKTdImJRm/EcD0wc5LU3AGYAzwJfA54ADgKukrRrRFzT5muamVkFFW3xHAO8FfgmMDEi1o2IdYEJwDeA7YAvthOIpNcDhwNfHqTKScDqwC4RcVJEfA/YBvgLcLoktfO6ZmZWTUVH8b0fOD8ijqovjIjHgc9JWg/YG/hSkZVKWoF0Lb8rgYtICbB+/guBdwHTI+K2utd9StKPSC2vLYCbC26PmVlzU1crO4L2DBb31Cf6G0cXFG1BrQNMH2L+b3Kdoo4ANgYOHWT+a4CxwI1N5t2Up1u08bpmZlZRRRPU48CGQ8zfMNdpmaT1geOA4yNi7iDV1s7TeU3m1comDbL+gyXNlDTzkUceKRKamZmVqGiCuhr4hKRdGmdI2hk4BLiq4Dp/ANwHfGuIOuPzdHGTeYsa6iwlIs6IiM0jYvOJE5uOvzAzswoqegzqGGAX4ApJfwBm5fJNgdcB84FjW12ZpH2BnYBtI+KZIaouzNOxTeaNa6hjZmajQNErSTwgaXPSiLp3Aq/PsxYA5wFfiIgHW1mXpLGkVtMVwF8l1boOa111q+Wy+aSRevXz6tXKmnX/mZnZCFX4auY5AX0wD+uu9Zk9EhGDXUR2MC/Iy++eH432zY/PkroBFwNbN6m3VZ7OLPj6ZmZWYW3fbiMnpL938NpPA3s1KZ8IfI805PzHwB/zcPJpwHskvTYibgeQtApwIHAPHmJuo9jA0ZeXHYJZ3xW9ksQngD0jYsdB5v8auDAifjjcuvIxp182WcdA/vPeiKif/3lgB+DXkk4BniRdSWISsHsbLTgzM6uwoqP4ppBaK4OZDRzQdjRDiIg5wJtJ5z0dTbpyxdPA2yOi6MhBMzOruKJdfK8AfjrE/FnAPu2HA/lcqKaXLYqIu4A9Olm/mZmNDEVbUCvxr2HdzYwbZr6ZmVlLiiao2aTzlgazM3Bv++GYmZklRRPUecDOkk6QtHKtUNJKko4jJahzuxmgmZktn4oegzoF2JV0S41DJP05l28MrAlcT8OVyM3MzNpRqAWVh4bvTBpF9xDp8kavA/4XOArYMSKWdDtIMzNb/rRzJYlnSHe0/Vr3wzEzM0uKHoMyMzPrCycoMzOrJCcoMzOrJCcoMzOrJCcoMzOrJCcoMzOrpLbvB2VmNmpMXa3sCKwJt6DMzKySnKDMzKySnKDMzKySnKDMzKySPEjCzAobOPryQvXnnrx7jyKx0cwtKDMzqyQnKDMzqyQnKDMzqyQnKDMzqyQnKDMzqySP4jPrs6Ij4MyWV25BmZlZJTlBmZlZJTlBmZlZJTlBmZlZJTlBmZlZJTlBmZlZJTlBmZlZJTlBmZlZJZWaoCS9UtLxkm6S9IikBZJuk/RFSS9sUn8jSZdIekzS05Kul7R9GbGbmVlvld2COgA4ArgXOB74LHA3cCIwQ9ILahUlbQDMALYGvpbrrgJcJWnHPsdtZmY9Vvaljn4JnBQRT9SV/UDSPcAXgY8Ap+Xyk4DVgTdExG0Akn4GzAJOl7RxRES/Ajczs94qtQUVETMbklPNL/L03wByd9+7gOm15JSXfwr4EfBKYIveRmtmZv1UdhffYNbJ07/l6WuAscCNTerelKdOUGZmo0jZXXzLkLQC8CXgWeDcXLx2ns5rskitbNIg6zsYOBhg8uTJ3QvULPPVyc16o4otqFNJAyGOjYi7c9n4PF3cpP6ihjpLiYgzImLziNh84sSJXQ3UzMx6p1ItKEknAIcCZ0TESXWzFubp2CaLjWuoY2bW3NTVyo7ACqhMC0rSVOAY4KfAxxpm/yVPm3Xj1cqadf+ZmdkIVYkElZPTl4GzgQObDBe/g9S9t3WTxbfK05k9C9DMzPqu9C4+SceSktPPgQMi4vnGOhHxlKRpwHskvTYibs/LrgIcCNwD3NzHsM2sgKIDSeaevHuPIlmONevenNrsLJ/qKDVBSfoEcBzwIHANsI+k+ip/i4ir89+fB3YAfi3pFOBJ4CBSF9/uPknXzGx0KbsFVTt3aTKpe6/Rb4CrASJijqQ3AycDRwMrA78H3h4R1/QhVjMz66NSE1RETAGmFKh/F7BHr+IxM7PqqMQgCTMzs0ZOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVkllX+rIzKz7fN+nUcEtKDMzqyQnKDMzqyR38ZmZLa8G6wqtyH2i3IIyM7NKcoIyM7NKcoIyM7NKcoIyM7NKcoIyM7NKcoIyM7NK8jBzMzNbWkWGn7sFZWZmleQWlJlVzsDRlxdeZu7Ju/cgEiuTW1BmZlZJbkHZqNfOr3EzK59bUGZmVkluQZmZWWuaje7r4cg+t6DMzKySnKDMzKySnKDMzKySnKDMzKySPEjCzEaEueP2WaZsYNG5JURi/eIWlJmZVdKIakFJGgN8CvgoMAA8ApwPHBsRT5cYmpmVrP6E7LnjSgzEumaktaBOAb4F3AkcBlwAfBKYlpOXmZmNEiOmBSVpU1JSuigi3ltXfj/wHeADgDukzcxGiRGToIC9AQGnNpSfCZwM7EuPE1Q/rulW9IrMRWPq9RWf+xGPr603OjQb9ADFBj4Mtg4bHUZSt9gWwPPAzfWFEbEIuC3PNzOzUUIRUXYMLZF0B/CSiHhpk3nnA3sBYyNiScO8g4GD89ONgLt7HWsXTADmlx1ED4327YPRv42jffvA29hP60XExMbCkdTFNx5YPMi8RXV1lkpQEXEGcEYP4+o6STMjYvOy4+iV0b59MPq3cbRvH3gbq2AkdfEtBMYOMm9cXR0zMxsFRlKC+gswQVKzJDUJmN/YvWdmZiPXSEpQt5Di3bK+UNI4YDNgZgkx9cqI6pJsw2jfPhj92zjatw+8jaUbSYMkXg3cDlzccB7UYaTzoD4UEeeUFZ+ZmXXXiElQAJK+CxwKXAxcAWxCupLEDcD2EfF8ieGZmVkXjbQEtQJwOGnY+ABpeOQvSNfie6q8yMzMrNtG0jEoIuK5iPhmRGwUEWMjYlJEHDmSk5OkMZI+L2mupEWSbpf03uGXBElnSYomj1N7HPZg8awr6ZeSnpD0pKSLJE1ucdlxkr4u6WFJ/5R0o6Rtex1zER1uX7P9FJI263HYLZO0jqTv5vd+YY5voMVl2/4c91OH2zh3kH347t5G3TpJ75N0oaQH8vfobkknSVq1hWUr9x0cUQlqlDoBmAqcBuwK3ARcIGm3Fpd/BNi64XFK98McmqTxwLXAxsB+wIeAVwDXSXphC6v4MXAQcCzwDuBh4Kqq/APvwvYBnMWy+2p214Nt34bAvwOPAdcXXLbTz3G/dLKNAFex7D78Tdei69xngOeALwBvB74PHAJc3cIFtav3HYwIP0p6AC8hnXx8XEP5/wB/bGH5s4CHyt6OHMunSF+MDevK1geeBY4cZtnXAgHsX1e2IumqH5eVvW2dbl+uG8CJZW/HMDGOqfv7wBzzQAvLdfQ5HgnbmOvPBc4pexuGiXFik7IP5+3cfojlKvkddAuqXLsAKwONow/PAV4taf3+h9S2dwE3RcScWkFE3E8awLJHC8s+QzqeWFv2WeC/gF0GOfet3zrZvhEh2h9kNGI+xx1s44gQEY80Kb4lTycNsWglv4NOUOXalPTLc05D+aw8fVUL63iJpPmSnpU0W9Ln8mCSftsU+FOT8lkMvx2bAvdHROOVQGaR/vFt2Hl4Hetk+2oOkbQ4H/u4VtI23QuvVN34HI8U78z7b7Gkm6p0/GkIb83Tu4aoU8nv4Ei6Ft9otCbweOT2dJ1H6+YP5TbgVtKHaBywJ3AS6djIgd0LsyVrkvr1Gz0KrNHBsrX5Zetk+yC1Jv6bdEWU9YDPAtdK2ikipncryJJ0+jkeKaaRWiP3Ay8ln/IiqbLnYEqaBBwPXBMRQ13MoJLfQSeoLpK0I3B1C1V/ExHbdfp6EXFqQ9EVkp4CDpf01Yi4p9PXsO6IiA/VPb1e0qWkFtmJwFvKicqKiIjD6p9Lupg0GOQklu3eLJ2kVYBLScdJ9y85nLY4QXXXDNLJw8OpNaMfA1aXpIZfn7VfK49S3Hmkc8U2B/qZoB6jeUtisF9mjcuuN8iy0N770G2dbN8yImKBpMuBj3QaWAX04nNceRHxnKQLgK9KWisiHi47phpJLyC1+F4OvDUiHhpmkUp+B52guij33/65wCKzSFdo34Cl++9rffZ3dhJOB8u2YxapH7vRqxh+O2YBe0oa39AH/irS7VMaj22UoZPtG8rIOVN+cL38HI8UldmPklYCfkn6kbpTRNzRwmKV/A56kES5riSNnPlgQ/m+wJ/yKLGiPkj6stwyXMUuuwzYStLLawX5BMg353lDmQasRLrpZG3ZFYH3A7+OiMHuA9ZPnWzfMiS9iHSuyc3D1R0BevE5rry6z+iDEfHXsuOBdMI08J/A9sC7I+KmFhet5new7HH7y/sDOJl0w8Ujge1IJ9Y9D7yjod7/AHPqnq8H/Bb4OLAz8E7gJ3nZ75ewHS8k/cq6gzTs+l2ki/veB6zSEPezpMtT1S//X6RuhgOBHUi/ABcBry97H3W6faSTJ88E9sn7eL+8niXANmVvW8N2vi8/vk/6oXNIfv7WujrPAj9u53NchUc72wjsnT+jHwbeBnyAdKJvAB8oe5vq4qxt04nAVg2PdQb7jObyyn0HS39Dl/cHsAJwDPAAaajuH4H3Nak3HZhb93xN4JK83CLSca3fk0YWjelX/A0xTgYuBJ4EFuT4BhrqDOQv0NSG8hcA3wL+mrfnd8B2Ze+fbmwf6cfDDaRrRz4D/IPU6tqy7G1qso0xyGN6Q52z2vkcV+HRzjbmf/DXAn/L+/Bx4Bpgl7K3p2Hb5g6xfVMH+4zm8sp9B0fUxWLNzGz54WNQZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZmZWSU5QZk1I2k5SSJrSzbqjhaS5kqY3lC1374P1lhOUlU7SZpKm5ouvWgckbS1plqR/SBoRd7L1/rfBOEFZFWwGfJl0jTBrU77Nwi+Ai3PR6/scwm9J13P7ecHlNsP735pwgrJRTdIKksaXHUefvAN4CemK1quRbi/fNxHxfEQsiojn+vm6Nno5QVnLJE3Jxxi2l/QZSfdKWixptqT9GuquKulESb+TND/XmyPp5PqEIWkq8NP89Lq8/pB0Vm1+fj7QJJ6ljoPUxbejpC9Jupd0VeZ/LxJTt0iaIOl0Sf8raUmeni7pxU3qDki6UNKT+XGppPWbHesZwp7Ab4D1SVervrUL27CupPMlPZHjmiZpg0HqLnMMStK4vA/vlrRQ0uOS7pD09Tx/KkPv/5b3WZHPZ90yK0s6StJtOb4nJM2UdGhDvbGSvpC7Txfl7Zgm6XVtvrXWAt9R19rxFVJXzg9Jt1Y4BDhL0pyIuCHXmUS6r8yFwLmk+8+8FTgKeB2wS653EbAWcHBe7125/N4O4vsG6eZrZ5JujXF3wZg6Jmk1YAawIek+Xb/Pr3EIsL2kLSNiQa77YtK9hV4K/ID0HmwDXEe6D1WrdiFt+07AjIh4osNtWJ3UbbdujutO0vt1HWn/t+J04ADgZ6RbOawIvIJ0Qz0Yfv+3s89a+XwiaWXgKtL9q34NnEP6QfNq4D3AabneSqSbMr6J1H15GqmFehBwg6RtI2Jmi++HFVH2/Uv8GDkPYArpl/kfgJXryieR/hGcV1e2MrBSk3WckNexZZP1btek/tQ8b6DJvLksfQ+f2nruBsY3qV8kpu1y2ZQW3pdl6gL/kcs+3lD3E7n8hLqyr+WyDzbUrZVPbyGGjXLdbUn/5Pfrwv7+Sl7n/g3lpzaLa5D34VHgihY/V832fzufo2E/n7n8qFz/K03WP6bu7yNyvV0a6rwIeLCV/eNHew938Vk7vhcRS2pPImIeMJv0y7hWtiQinoF062hJa0iaQLrJG8Abexjf9yNiYWNhn2PaE3gEOKOh/Ie5fM+6sncCDwPnNdT9RoHX24r0T3RV0s0szy8S7CDeTbpB388ayr9aYB1PAJtK+rd2Amhznw37+cw+SLqD7PFNXvf5uqf7An8Gbs3dthNyDCsDVwNvkdRqi9IKcIKydtzXpOwfwFLHViR9XNIfSb9eHyX9Y56eZ6/Rw/hmDzajjzGtD9wdEc/WF+bns4GXN9Sd0/BPkYj4O+nOra14FTAP+CipRfBP+P9jJ2dKuk/Sgnw85rAW1/ly4J5oGPQQEQ8XiOtw0vt6Rz4m9CNJe0hq+X9PG/uspc8nKWH9OSIWDRPCJsDG+XUbHweQ7iY8YdgNscJ8DMraMdgoLf3/H9KRwDdJffvfIY0oW0LqbjmL1n8cDXXL58E+v8u0nrocUxVNAMYCmwJ71ZWvSLqF986kf9yvAa6S9LeI6EYra0gRcanSAJfdSMeOdgQ+Alwvacf6lk4zbe6zYT+fBQm4AzhyiDqPtLluG4ITlPXKh0jHiHatbxlIenuTukMloUfzdM28vtp6xpEOrs/pUUydug/YSNKK9a0oSSsCr2TpX/lzgQ0ljWmI6yXA6gVecyLpeNHiWkFEPA18qa7ObZIuA97C8N2A9wGvkLRCfStK0lpF4oqIR0kDEM6RJOBk0vGfPYALGHr/93KfzQY2ljS2/j1r4h7Se3ttYyvXemsk/2K0anuO9I+nvlW1InB0k7pP5emaTebVuut2bCg/guKf3yIxdeoS0j+1AxvKD8rlF9eVTSMl270b6n6mwOu9DLgtIi4fqlIekbYN8McW1nkpaWThhxvKP9dKQErnoK1eXxZpdMEf8tPa/h5q//dyn/0nqYvwmMYZOZHW/Iz0/jZtQUl6aRdisSbcgrJe+SVwEvArSReRRjztAzzTpO4twPPAFyWtATwN3B8RvyMdDL8bOD4Px76f9Ot/K2B+D2Pq1NdIXW2nS3o96Z/y60jdW3fn+TVfzXH8VNKWpAPy25CGNc9n6BYGkvYmDS1/TtI6pG6+jwLvrw0wqHMasIBlBz4Mtg37AGdKegMwizRSb2tae+9XBR7OLbY/AH8nHW87hDQ4YVquN9T+7+U++zZpgMoxkrYgdSMuIr1/G/GvH0XfJr2/X5e0PXAt6fSFycAOeZm3dSEea1T2MEI/Rs6DoYcDTwfm1j1fAfg8qQtuMfAA6R/eJnkdUxuW3490ns2SPP+sunmvJJ2HspB0cP580jGIuTQfZr5MfEVjosNh5rl8IvA94CHSP9SHSOcFTWiyjvVJ5wQtIP3zuzSXzWeIYdqk404Xkc7bOZ10jOZGYOMmdb9Fajkt8/pDrH8yKUk8mR/TgA0a3/tm7wNplNtJwM2kQQqL83I/AV7Ryv4vuM8G3f80fD7ryscBXyQl30X583ULy54esCLwyTzv6fy4h9QK27ns7+ZofSi/+WZWMbnFOB/4YUR8rMN1nUr6tb99RPiAvo0IPgZlVgGDnEdTO85ydYfr/g6pu8rJyUYUt6DMKkDSdaTuq9+TfjjuQLr46wxg22jzAqyS1iN1qy0mXSao5vqI2LWTmM16zQnKrAIkfZo0Wm6AdB25h0jHlo6LfM0+s+WNE5SZmVWSj0GZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVklOUGZmVkl/R89uIhVg67d5QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('mnistwo9', 'natural')\n",
      "0.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('mnistwo9', 'TRADES(2)')\n",
      "0.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('mnistwo9', 'TRADES(4)')\n",
      "0.12\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('mnistwo9', 'TRADES(8)')\n",
      "0.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('mnistwo9', 'AT(2)')\n",
      "0.08666666666666667\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n",
      "Rare\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "all_settings = []\n",
    "\n",
    "method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(2)\", 'advce', 2.0)]\n",
    "\n",
    "#for i in [0, 4, 9]:\n",
    "#    for method_name, loss, eps in method_names:\n",
    "#        arch = \"WRN_40_10\"\n",
    "#        all_settings.append(\n",
    "#            dict(\n",
    "#                dname = f\"cifar10wo{i}\",\n",
    "#                dataset = f\"cifar10wo{i}\",\n",
    "#                model_path = f\"../models/out_of_sample/pgd-64-cifar10wo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0-ep0070.pt\",\n",
    "#                result_path = f\"../results/oos_repr/cwl2-64-cifar10wo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "#                ood_result_paths = [\n",
    "#                    f\"../results/ood_robustness_correct/cwl2-64-cifar10wo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "#                ],\n",
    "#                method_name = method_name,\n",
    "#                arch = arch,\n",
    "#            )\n",
    "#        )\n",
    "#        \n",
    "#        \n",
    "#        all_settings.append(\n",
    "#            dict(\n",
    "#                dname = f\"cifar100coarsewo{i}\",\n",
    "#                dataset = f\"cifar100coarsewo{i}\",\n",
    "#                model_path = f\"../models/out_of_sample/pgd-64-cifar100coarsewo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0-ep0070.pt\",\n",
    "#                result_path = f\"../results/oos_repr/cwl2-64-cifar100coarsewo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "#                ood_result_paths = [\n",
    "#                    f\"../results/ood_robustness_correct/cwl2-64-cifar100coarsewo{i}-70-{eps}-0.01-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "#                ],\n",
    "#                method_name = method_name,\n",
    "#                arch = arch,\n",
    "#            )\n",
    "#        )\n",
    "        \n",
    "method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(2)\", 'advce', 2.0)]\n",
    "for i in [0, 1, 2]:\n",
    "    arch = \"ResNet50Norm01\"\n",
    "    for method_name, loss, eps in method_names:\n",
    "        lr = 0.01\n",
    "        all_settings.append(\n",
    "            dict(\n",
    "                dname = f\"aug10-imgnet100wo{i}\",\n",
    "                dataset = f\"aug10-imgnet100wo{i}\",\n",
    "                model_path = f\"../models/out_of_sample/pgd-128-aug10-imgnet100wo{i}-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0-ep0070.pt\",\n",
    "                result_path = f\"../results/oos_repr/cwl2-128-aug10-imgnet100wo{i}-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "                ood_result_paths = [\n",
    "                    f\"../results/ood_robustness_correct/cwl2-128-aug10-imgnet100wo{i}-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "                    f\"../results/ood_robustness_correct/fbpgdl2-128-aug10-imgnet100wo{i}-70-{eps}-{lr}-{loss}-vtor2-{arch}-0.0-2-adam-0-0.0.pkl\",\n",
    "                ],\n",
    "                method_name = method_name,\n",
    "                arch = arch,\n",
    "            )\n",
    "        )\n",
    "        \n",
    "method_names = [(\"natural\", 'ce', 1.0), (\"TRADES(2)\", 'trades6ce', 2.0), (\"TRADES(4)\", 'trades6ce', 4.0), (\"TRADES(8)\", 'trades6ce', 8.0), (\"AT(2)\", 'advce', 2.0)]\n",
    "for i in [0, 1, 4, 9]:\n",
    "    for method_name, loss, eps in method_names:\n",
    "        all_settings.append(\n",
    "            dict(\n",
    "                dname = f\"mnistwo{i}\",\n",
    "                dataset = f\"mnistwo{i}\",\n",
    "                model_path = f\"../models/out_of_sample/pgd-128-mnistwo{i}-70-{eps}-0.01-{loss}-vtor2-CNN002-0.9-2-sgd-0-0.0-ep0070.pt\",\n",
    "                result_path = f\"../results/oos_repr/cwl2-128-mnistwo{i}-70-{eps}-0.01-{loss}-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ood_result_paths = [\n",
    "                    f\"../results/ood_robustness_correct/cwl2-128-mnistwo{i}-70-{eps}-0.01-{loss}-vtor2-CNN002-0.9-2-sgd-0-0.0.pkl\",\n",
    "                ],\n",
    "                method_name = method_name,\n",
    "                arch = \"CNN002\",\n",
    "            )\n",
    "        )\n",
    "        \n",
    "# fraction of time the label goes from cat to dog\n",
    "\n",
    "for ent in all_settings:\n",
    "    #dataset, dname, method_name, model_path, result_path, ood_robust_paths, arch = ent['dataset'], ent['dname'], ent['method_name'], ent['model_path'], ent['result_path'], ent['ood_robust_paths'], ent['arch']\n",
    "    #adv_dists, ood_dists, in_rob_r_portion = get_rob_results(dataset, dname, method_name, model_path, result_path, ood_robust_paths, arch)\n",
    "    key = (ent['dname'], ent['method_name'])\n",
    "    print(key)\n",
    "    if key not in results:\n",
    "        ent['with_legend'] = False\n",
    "        adv_dists, ood_dists, in_rob_r_portion, ncg_acc, ood_nn_idx, within_list, ncg_correctness = get_rob_results(**ent)\n",
    "        results[key] = [np.mean(adv_dists), np.mean(ood_dists), in_rob_r_portion, ncg_acc, ood_nn_idx, within_list, ncg_correctness]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "5104cf26",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = {}\n",
    "for k, v in results.items():\n",
    "    ood_nn_idx, within_list, ncg_correctness = v[4], v[5], v[6]\n",
    "    data[k] = [v[0], v[1], v[2], v[3], np.logical_and(within_list, ncg_correctness).mean()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "2762e83e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>avg adv dist.</th>\n",
       "      <th>avg OOD dist.</th>\n",
       "      <th>XD</th>\n",
       "      <th>ncg acc</th>\n",
       "      <th>in radius portion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">aug10-imgnet100wo0</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.766684</td>\n",
       "      <td>155.069113</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.044444</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>3.139885</td>\n",
       "      <td>155.624168</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.040741</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>3.535575</td>\n",
       "      <td>160.046210</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.055556</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>2.674800</td>\n",
       "      <td>161.096930</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.066667</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>0.292400</td>\n",
       "      <td>157.418586</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.033333</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">aug10-imgnet100wo1</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.720583</td>\n",
       "      <td>153.034865</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049630</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>2.849707</td>\n",
       "      <td>155.914205</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.051852</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>2.919561</td>\n",
       "      <td>156.972648</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.060741</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>2.324356</td>\n",
       "      <td>157.905458</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.072593</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>0.262179</td>\n",
       "      <td>152.469264</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.047407</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">aug10-imgnet100wo2</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.219428</td>\n",
       "      <td>154.279823</td>\n",
       "      <td>0.004474</td>\n",
       "      <td>0.026667</td>\n",
       "      <td>0.004474</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>2.488140</td>\n",
       "      <td>155.756040</td>\n",
       "      <td>0.002141</td>\n",
       "      <td>0.030370</td>\n",
       "      <td>0.002141</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>2.673374</td>\n",
       "      <td>156.826617</td>\n",
       "      <td>0.004149</td>\n",
       "      <td>0.043704</td>\n",
       "      <td>0.004149</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>2.273520</td>\n",
       "      <td>156.272611</td>\n",
       "      <td>0.004246</td>\n",
       "      <td>0.047407</td>\n",
       "      <td>0.004246</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>0.149080</td>\n",
       "      <td>154.157022</td>\n",
       "      <td>0.002045</td>\n",
       "      <td>0.027407</td>\n",
       "      <td>0.002045</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">calcedrepr-mnistwo4</th>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>16.436266</td>\n",
       "      <td>54.225028</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.814918</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>20.389073</td>\n",
       "      <td>53.803331</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.859613</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">calcedrepr-mnistwo9</th>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>16.430955</td>\n",
       "      <td>50.695964</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.741880</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>20.465122</td>\n",
       "      <td>51.481997</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.798362</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">mnistwo0</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.332245</td>\n",
       "      <td>7.001075</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.456758</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>2.165902</td>\n",
       "      <td>6.976460</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.456758</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>2.072912</td>\n",
       "      <td>6.936148</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.476604</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>0.682265</td>\n",
       "      <td>6.971984</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.420976</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>1.257136</td>\n",
       "      <td>6.951552</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.362161</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">mnistwo1</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>1.833255</td>\n",
       "      <td>4.299455</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.513774</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>1.566082</td>\n",
       "      <td>4.307595</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.354831</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>1.439671</td>\n",
       "      <td>4.307122</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.541069</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>0.563512</td>\n",
       "      <td>4.287753</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.284753</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>0.890287</td>\n",
       "      <td>4.327680</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.228006</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">mnistwo4</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.339243</td>\n",
       "      <td>5.639562</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.744138</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>2.323594</td>\n",
       "      <td>5.886802</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.810229</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>1.841858</td>\n",
       "      <td>5.654624</td>\n",
       "      <td>0.001242</td>\n",
       "      <td>0.758353</td>\n",
       "      <td>0.001242</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>0.657575</td>\n",
       "      <td>5.829020</td>\n",
       "      <td>0.002410</td>\n",
       "      <td>0.749121</td>\n",
       "      <td>0.002410</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>1.330655</td>\n",
       "      <td>5.730208</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.760111</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">mnistwo9</th>\n",
       "      <th>AT(2)</th>\n",
       "      <td>2.328356</td>\n",
       "      <td>5.142083</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.714286</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(2)</th>\n",
       "      <td>2.278334</td>\n",
       "      <td>5.246197</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.683099</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(4)</th>\n",
       "      <td>2.181535</td>\n",
       "      <td>5.131419</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.703076</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRADES(8)</th>\n",
       "      <td>0.656668</td>\n",
       "      <td>5.196853</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.648318</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>natural</th>\n",
       "      <td>1.449574</td>\n",
       "      <td>5.081838</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.578471</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                               avg adv dist.  avg OOD dist.        XD  \\\n",
       "aug10-imgnet100wo0  AT(2)           2.766684     155.069113  0.000000   \n",
       "                    TRADES(2)       3.139885     155.624168  0.000000   \n",
       "                    TRADES(4)       3.535575     160.046210  0.000000   \n",
       "                    TRADES(8)       2.674800     161.096930  0.000000   \n",
       "                    natural         0.292400     157.418586  0.000000   \n",
       "aug10-imgnet100wo1  AT(2)           2.720583     153.034865  0.000000   \n",
       "                    TRADES(2)       2.849707     155.914205  0.000000   \n",
       "                    TRADES(4)       2.919561     156.972648  0.000000   \n",
       "                    TRADES(8)       2.324356     157.905458  0.000000   \n",
       "                    natural         0.262179     152.469264  0.000000   \n",
       "aug10-imgnet100wo2  AT(2)           2.219428     154.279823  0.004474   \n",
       "                    TRADES(2)       2.488140     155.756040  0.002141   \n",
       "                    TRADES(4)       2.673374     156.826617  0.004149   \n",
       "                    TRADES(8)       2.273520     156.272611  0.004246   \n",
       "                    natural         0.149080     154.157022  0.002045   \n",
       "calcedrepr-mnistwo4 TRADES(4)      16.436266      54.225028  0.000000   \n",
       "                    TRADES(8)      20.389073      53.803331  0.000000   \n",
       "calcedrepr-mnistwo9 TRADES(4)      16.430955      50.695964  0.000000   \n",
       "                    TRADES(8)      20.465122      51.481997  0.000000   \n",
       "mnistwo0            AT(2)           2.332245       7.001075  0.000000   \n",
       "                    TRADES(2)       2.165902       6.976460  0.000000   \n",
       "                    TRADES(4)       2.072912       6.936148  0.000000   \n",
       "                    TRADES(8)       0.682265       6.971984  0.000000   \n",
       "                    natural         1.257136       6.951552  0.000000   \n",
       "mnistwo1            AT(2)           1.833255       4.299455  0.000000   \n",
       "                    TRADES(2)       1.566082       4.307595  0.000000   \n",
       "                    TRADES(4)       1.439671       4.307122  0.000000   \n",
       "                    TRADES(8)       0.563512       4.287753  0.000000   \n",
       "                    natural         0.890287       4.327680  0.000000   \n",
       "mnistwo4            AT(2)           2.339243       5.639562  0.000000   \n",
       "                    TRADES(2)       2.323594       5.886802  0.000000   \n",
       "                    TRADES(4)       1.841858       5.654624  0.001242   \n",
       "                    TRADES(8)       0.657575       5.829020  0.002410   \n",
       "                    natural         1.330655       5.730208  0.000000   \n",
       "mnistwo9            AT(2)           2.328356       5.142083  0.000000   \n",
       "                    TRADES(2)       2.278334       5.246197  0.000000   \n",
       "                    TRADES(4)       2.181535       5.131419  0.000000   \n",
       "                    TRADES(8)       0.656668       5.196853  0.000000   \n",
       "                    natural         1.449574       5.081838  0.000000   \n",
       "\n",
       "                                ncg acc  in radius portion  \n",
       "aug10-imgnet100wo0  AT(2)      0.044444           0.000000  \n",
       "                    TRADES(2)  0.040741           0.000000  \n",
       "                    TRADES(4)  0.055556           0.000000  \n",
       "                    TRADES(8)  0.066667           0.000000  \n",
       "                    natural    0.033333           0.000000  \n",
       "aug10-imgnet100wo1  AT(2)      0.049630           0.000000  \n",
       "                    TRADES(2)  0.051852           0.000000  \n",
       "                    TRADES(4)  0.060741           0.000000  \n",
       "                    TRADES(8)  0.072593           0.000000  \n",
       "                    natural    0.047407           0.000000  \n",
       "aug10-imgnet100wo2  AT(2)      0.026667           0.004474  \n",
       "                    TRADES(2)  0.030370           0.002141  \n",
       "                    TRADES(4)  0.043704           0.004149  \n",
       "                    TRADES(8)  0.047407           0.004246  \n",
       "                    natural    0.027407           0.002045  \n",
       "calcedrepr-mnistwo4 TRADES(4)  0.814918           0.000000  \n",
       "                    TRADES(8)  0.859613           0.000000  \n",
       "calcedrepr-mnistwo9 TRADES(4)  0.741880           0.000000  \n",
       "                    TRADES(8)  0.798362           0.000000  \n",
       "mnistwo0            AT(2)      0.456758           0.000000  \n",
       "                    TRADES(2)  0.456758           0.000000  \n",
       "                    TRADES(4)  0.476604           0.000000  \n",
       "                    TRADES(8)  0.420976           0.000000  \n",
       "                    natural    0.362161           0.000000  \n",
       "mnistwo1            AT(2)      0.513774           0.000000  \n",
       "                    TRADES(2)  0.354831           0.000000  \n",
       "                    TRADES(4)  0.541069           0.000000  \n",
       "                    TRADES(8)  0.284753           0.000000  \n",
       "                    natural    0.228006           0.000000  \n",
       "mnistwo4            AT(2)      0.744138           0.000000  \n",
       "                    TRADES(2)  0.810229           0.000000  \n",
       "                    TRADES(4)  0.758353           0.001242  \n",
       "                    TRADES(8)  0.749121           0.002410  \n",
       "                    natural    0.760111           0.000000  \n",
       "mnistwo9            AT(2)      0.714286           0.000000  \n",
       "                    TRADES(2)  0.683099           0.000000  \n",
       "                    TRADES(4)  0.703076           0.000000  \n",
       "                    TRADES(8)  0.648318           0.000000  \n",
       "                    natural    0.578471           0.000000  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame.from_dict(data, orient=\"index\",\n",
    "                            columns=[\"avg adv dist.\", \"avg OOD dist.\", \"XD\", \"ncg acc\", \"in radius portion\"])\n",
    "df = df.sort_index()\n",
    "df.index = pd.MultiIndex.from_tuples(df.index)\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "9122e8f4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\begin{tabular}{llrrrr}\n",
      "\\toprule\n",
      "         &         &  empirical robust radius &  OOD dist. &  percentage covered &  NCG acc. \\\\\n",
      "\\midrule\n",
      "\\multirow{5}{*}{I-0} & AT(2) &           2.77 &         155.07 &               0.00 &     0.04 \\\\\n",
      "         & TRADES(2) &           3.14 &         155.62 &               0.00 &     0.04 \\\\\n",
      "         & TRADES(4) &           3.54 &         160.05 &               0.00 &     0.06 \\\\\n",
      "         & TRADES(8) &           2.67 &         161.10 &               0.00 &     0.07 \\\\\n",
      "         & natural &           0.29 &         157.42 &               0.00 &     0.03 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{I-1} & AT(2) &           2.72 &         153.03 &               0.00 &     0.05 \\\\\n",
      "         & TRADES(2) &           2.85 &         155.91 &               0.00 &     0.05 \\\\\n",
      "         & TRADES(4) &           2.92 &         156.97 &               0.00 &     0.06 \\\\\n",
      "         & TRADES(8) &           2.32 &         157.91 &               0.00 &     0.07 \\\\\n",
      "         & natural &           0.26 &         152.47 &               0.00 &     0.05 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{I-2} & AT(2) &           2.22 &         154.28 &               0.00 &     0.03 \\\\\n",
      "         & TRADES(2) &           2.49 &         155.76 &               0.00 &     0.03 \\\\\n",
      "         & TRADES(4) &           2.67 &         156.83 &               0.00 &     0.04 \\\\\n",
      "         & TRADES(8) &           2.27 &         156.27 &               0.00 &     0.05 \\\\\n",
      "         & natural &           0.15 &         154.16 &               0.00 &     0.03 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{2}{*}{calcedrepr-mnistwo4} & TRADES(4) &          16.44 &          54.23 &               0.00 &     0.81 \\\\\n",
      "         & TRADES(8) &          20.39 &          53.80 &               0.00 &     0.86 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{2}{*}{calcedrepr-mnistwo9} & TRADES(4) &          16.43 &          50.70 &               0.00 &     0.74 \\\\\n",
      "         & TRADES(8) &          20.47 &          51.48 &               0.00 &     0.80 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{mnistwo0} & AT(2) &           2.33 &           7.00 &               0.00 &     0.46 \\\\\n",
      "         & TRADES(2) &           2.17 &           6.98 &               0.00 &     0.46 \\\\\n",
      "         & TRADES(4) &           2.07 &           6.94 &               0.00 &     0.48 \\\\\n",
      "         & TRADES(8) &           0.68 &           6.97 &               0.00 &     0.42 \\\\\n",
      "         & natural &           1.26 &           6.95 &               0.00 &     0.36 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{mnistwo1} & AT(2) &           1.83 &           4.30 &               0.00 &     0.51 \\\\\n",
      "         & TRADES(2) &           1.57 &           4.31 &               0.00 &     0.35 \\\\\n",
      "         & TRADES(4) &           1.44 &           4.31 &               0.00 &     0.54 \\\\\n",
      "         & TRADES(8) &           0.56 &           4.29 &               0.00 &     0.28 \\\\\n",
      "         & natural &           0.89 &           4.33 &               0.00 &     0.23 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{mnistwo4} & AT(2) &           2.34 &           5.64 &               0.00 &     0.74 \\\\\n",
      "         & TRADES(2) &           2.32 &           5.89 &               0.00 &     0.81 \\\\\n",
      "         & TRADES(4) &           1.84 &           5.65 &               0.00 &     0.76 \\\\\n",
      "         & TRADES(8) &           0.66 &           5.83 &               0.00 &     0.75 \\\\\n",
      "         & natural &           1.33 &           5.73 &               0.00 &     0.76 \\\\\n",
      "\\cline{1-6}\n",
      "\\multirow{5}{*}{mnistwo9} & AT(2) &           2.33 &           5.14 &               0.00 &     0.71 \\\\\n",
      "         & TRADES(2) &           2.28 &           5.25 &               0.00 &     0.68 \\\\\n",
      "         & TRADES(4) &           2.18 &           5.13 &               0.00 &     0.70 \\\\\n",
      "         & TRADES(8) &           0.66 &           5.20 &               0.00 &     0.65 \\\\\n",
      "         & natural &           1.45 &           5.08 &               0.00 &     0.58 \\\\\n",
      "\\bottomrule\n",
      "\\end{tabular}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "text = df[[\"avg adv dist.\", \"avg OOD dist.\", \"in radius portion\", \"ncg acc\",]].to_latex(float_format=\"%.2f\", multirow=True)\n",
    "text = text.replace(\"avg adv dist.\", \"empirical robust radius\")\n",
    "text = text.replace(\"avg OOD dist.\", \"OOD dist.\")\n",
    "text = text.replace(\"in radius portion\", \"percentage covered\")\n",
    "text = text.replace(\"ncg acc\", \"NCG acc.\")\n",
    "text = text.replace(\"aug10-imgnet100wo\", \"I-\")\n",
    "text = text.replace(\"cifar100coarsewo\", \"C100-\")\n",
    "text = text.replace(\"cifar10wo\", \"C10-\")\n",
    "print(text)"
   ]
  }
 ],
 "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}