05_label_smoothing_visualization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import torch.nn as nn\n",
    "import torch\n",
    "import torch.nn.functional as F\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm import tqdm\n",
    "import torchvision.transforms as transforms\n",
    "import torchvision\n",
    "\n",
    "from models import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n"
     ]
    }
   ],
   "source": [
    "batch_size=100\n",
    "transform_test = torchvision.transforms.Compose([torchvision.transforms.ToTensor(), \n",
    "                                     torchvision.transforms.Normalize((0.4914, 0.4822, 0.4465), (0.247, 0.243, 0.261))])\n",
    "trainset = torchvision.datasets.CIFAR10(root='./data/CIFAR10', train=True, download=True, transform=transform_test)\n",
    "testset = torchvision.datasets.CIFAR10(root='./data/CIFAR10', train=False, download=True, transform=transform_test)\n",
    "train_loader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=False, num_workers=16)\n",
    "test_loader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False, num_workers=16)\n",
    "dataloaders = {}\n",
    "\n",
    "dataloaders['train'] = train_loader\n",
    "dataloaders['test'] = test_loader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Represent(nn.Module):\n",
    "    def __init__(self, base_model):\n",
    "        super(Represent, self).__init__()\n",
    "        self.conv1 = base_model.conv1\n",
    "        self.bn1 = base_model.bn1\n",
    "        self.layer1 = base_model.layer1\n",
    "        self.layer2 = base_model.layer2\n",
    "        self.layer3 = base_model.layer3\n",
    "        self.fc = base_model.fc\n",
    "        self.avgpool = base_model.avgpool\n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        out = self.conv1(x)\n",
    "        out = self.bn1(out)\n",
    "        out = self.layer1(out)\n",
    "        out = self.layer2(out)\n",
    "        out = self.layer3(out)\n",
    "        out = F.relu(out)\n",
    "        out = self.avgpool(out)\n",
    "        out = out.view(out.size(0), -1)\n",
    "        \n",
    "        return out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_statistics(dataloaders, device, representer, model, mode = 'train'):\n",
    "    label_list = np.empty((0,))\n",
    "    logit_sum_list = np.empty((0,))\n",
    "\n",
    "\n",
    "    for idx, data in enumerate(dataloaders[mode]):\n",
    "        image = data[0].to(device)\n",
    "        label = data[1].long().to(device)\n",
    "\n",
    "        output = representer(image)\n",
    "        pred = model(image)\n",
    "        \n",
    "        logit_sum = torch.sum(pred, dim=1)\n",
    "        pred = F.softmax(pred, dim=1)\n",
    "        score, prediction = torch.max(pred, dim=1)\n",
    "        \n",
    "        label_list = np.concatenate((label_list, label.cpu()))\n",
    "        logit_sum_list = np.concatenate((logit_sum_list, logit_sum.detach().cpu()))\n",
    "\n",
    "        if idx ==0:\n",
    "            out_list  = output.detach().cpu()\n",
    "            pred_list = (prediction == label).detach().cpu()\n",
    "            score_list = score.detach().cpu()\n",
    "        else:\n",
    "            out_list  = np.concatenate((out_list, output.detach().cpu()), axis=0)\n",
    "            pred_list = np.concatenate((pred_list, (prediction == label).detach().cpu()), axis=0)\n",
    "            score_list = np.concatenate((score_list, score.detach().cpu()), axis=0)\n",
    "            \n",
    "    return label_list, out_list, pred_list, score_list, logit_sum_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_singular_value_vector(label_list, out_list):\n",
    "    \n",
    "    singular_dict = {}\n",
    "    v_ortho_dict = {}\n",
    "    \n",
    "    for index in np.unique(label_list):\n",
    "        u, s, v = np.linalg.svd(out_list[label_list==index])\n",
    "        singular_dict[index] = s[0] / s[1]\n",
    "        v_ortho_dict[index] = torch.from_numpy(v[:2])\n",
    "\n",
    "    return singular_dict, v_ortho_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_plane(v_ortho_dict, model_represents, label, device):\n",
    "    # class 0,1,2\n",
    "    model_represents = torch.from_numpy(model_represents).to(device)\n",
    "    out = None\n",
    "    sing_lbl = torch.zeros(model_represents.shape[0]) > 0.\n",
    "    \n",
    "    for i, data in enumerate(model_represents):\n",
    "        if label[i] == 0 or label[i] == 3 or label[i] == 6:\n",
    "            data = data.to(device)\n",
    "            tmp = torch.zeros(3)\n",
    "            for j in range(3):\n",
    "                tmp[j] = torch.dot(v_ortho_dict[j*3].to(device), data)\n",
    "            if out == None:\n",
    "                out = tmp.unsqueeze(0)\n",
    "            else:\n",
    "                out = torch.cat((out, tmp.unsqueeze(0)), dim=0)\n",
    "            sing_lbl[i] = True\n",
    "    return out.detach().cpu().numpy(), sing_lbl.detach().cpu().numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_orthonormal_basis(out):\n",
    "    a = (out[0]-out[1]) / np.linalg.norm(out[0]-out[1])\n",
    "    b= (out[1] - out[2]) - np.dot(out[1] - out[2], a) * a\n",
    "    b = b / np.linalg.norm(b)\n",
    "    \n",
    "    return a, b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def projection(out, a, b):\n",
    "    x = []\n",
    "    y = []\n",
    "    for idx, data in enumerate(out):\n",
    "        x.append(np.dot(data, a))\n",
    "        y.append(np.dot(data, b))\n",
    "\n",
    "    x = np.array(x)\n",
    "    y = np.array(y)\n",
    "    \n",
    "    return x, y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_intra_cluster_divergence(cluster1):\n",
    "    cl1_mean = np.mean(cluster1, axis=0)\n",
    "    intra_cluster_divergence = 0\n",
    "\n",
    "    for i in range(cluster1.shape[0]):\n",
    "        intra_cluster_divergence += np.linalg.norm(cluster1[i] - cl1_mean) ** 2\n",
    "                \n",
    "    return intra_cluster_divergence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def verify_cluster(x, y, label_list, sing_lbl):\n",
    "    cluster1 = np.concatenate((np.expand_dims(np.array(x[label_list[sing_lbl]==0]), 0), np.expand_dims(np.array(y[label_list[sing_lbl]==0]), 0)),axis=0).transpose()\n",
    "    cluster2 = np.concatenate((np.expand_dims(np.array(x[label_list[sing_lbl]==3]), 0), np.expand_dims(np.array(y[label_list[sing_lbl]==3]), 0)),axis=0).transpose()\n",
    "    cluster3 = np.concatenate((np.expand_dims(np.array(x[label_list[sing_lbl]==6]), 0), np.expand_dims(np.array(y[label_list[sing_lbl]==6]), 0)),axis=0).transpose()\n",
    "    \n",
    "    cl1_mean = np.mean(cluster1, axis=0)\n",
    "    cl2_mean = np.mean(cluster2, axis=0)\n",
    "    cl3_mean = np.mean(cluster3, axis=0)\n",
    "\n",
    "    all_mean = (cl1_mean + cl2_mean + cl3_mean) / 3\n",
    "    \n",
    "    inter_cluster_divergence = (np.linalg.norm(cl1_mean - all_mean)**2 + np.linalg.norm(cl2_mean - all_mean)**2 + np.linalg.norm(cl3_mean - all_mean)**2)/3\n",
    "    intra_cluster_divergence = compute_intra_cluster_divergence(cluster1) + compute_intra_cluster_divergence(cluster2) + compute_intra_cluster_divergence(cluster3)\n",
    "    \n",
    "    return inter_cluster_divergence, intra_cluster_divergence\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SNR"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResNet(\n",
       "  (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (relu): ReLU(inplace=True)\n",
       "  (layer1): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (layer2): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (layer3): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (avgpool): AvgPool2d(kernel_size=8, stride=8, padding=0)\n",
       "  (fc): Linear(in_features=256, out_features=10, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "device = 'cuda:0'\n",
    "\n",
    "# teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= True)\n",
    "teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= False)\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-t-bn/sgd/step/kaiming_both_none_0.1_0.1_0.1_0.1_seed1_False.pt'\n",
    "filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_sin_srip_0.01_0.01_0.01_0.01_seed1_True.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_norm_0.1_0.1_0.1_0.1_seed1_True.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_ortho_0.01_0.01_0.01_0.01_seed4_True.pt' # ortho\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_ort_0.1_0.1_0.1_0.1_seed1_True.pt' #norm\n",
    "\n",
    "checkpoint = torch.load(filename, map_location=device)['best_wts']\n",
    "teacher.load_state_dict(checkpoint, strict=False)\n",
    "teacher.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "teacher = teacher.to(device)\n",
    "representer = Represent(teacher)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='train')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "## snr\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "plt.savefig('snr_visual_train.pdf')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='test')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "# a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(184.60297218622478, 99945.31136108417)\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": [
    "## VANILLA\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "plt.savefig('snr_visual_test.pdf')\n",
    "print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "33.31510378702806"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "99945.31136108417 / 3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "8303.36761336366"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "184.60297218622478 / 33.31510378702806/2 * 2997"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Vanilla"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResNet(\n",
       "  (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (relu): ReLU(inplace=True)\n",
       "  (layer1): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (layer2): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (layer3): Sequential(\n",
       "    (0): BasicBlock(\n",
       "      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (5): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (6): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (7): BasicBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (avgpool): AvgPool2d(kernel_size=8, stride=8, padding=0)\n",
       "  (fc): Linear(in_features=256, out_features=10, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "device = 'cuda:0'\n",
    "\n",
    "# teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= True)\n",
    "teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= False)\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-t-bn/sgd/step/kaiming_both_none_0.1_0.1_0.1_0.1_seed1_False.pt'\n",
    "filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_none_0.01_0.01_0.01_0.01_seed3.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_norm_0.1_0.1_0.1_0.1_seed1_True.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_ortho_0.01_0.01_0.01_0.01_seed4_True.pt' # ortho\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_ort_0.1_0.1_0.1_0.1_seed1_True.pt' #norm\n",
    "\n",
    "checkpoint = torch.load(filename, map_location=device)['best_wts']\n",
    "teacher.load_state_dict(checkpoint, strict=False)\n",
    "teacher.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "teacher = teacher.to(device)\n",
    "representer = Represent(teacher)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='train')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "## VANILLA\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "plt.savefig('vanilla_visual_train.pdf')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='test')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "# a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(223.52828111200142, 177637.01646143553)\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": [
    "## VANILLA\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "plt.savefig('vanilla_visual_test.pdf')\n",
    "print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "65.36522383931252"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "196095.67151793756 / 1000 / 3 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5124.393516493725"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "223.52828111200142 / 65.36522383931252 /2 * 2997"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# norm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = 'cuda:0'\n",
    "\n",
    "# teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= True)\n",
    "teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= False)\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-t-bn/sgd/step/kaiming_both_none_0.1_0.1_0.1_0.1_seed1_False.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_none_0.01_0.01_0.01_0.01_seed3.pt'\n",
    "filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_norm_0.1_0.1_0.1_0.1_seed1_True.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_srip_0.1_0.1_0.1_0.1_seed1.pt'\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_ortho_0.01_0.01_0.01_0.01_seed4_True.pt' # ortho\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_ort_0.1_0.1_0.1_0.1_seed1_True.pt' #norm\n",
    "\n",
    "checkpoint = torch.load(filename, map_location=device)['best_wts']\n",
    "teacher.load_state_dict(checkpoint, strict=False)\n",
    "teacher.eval()\n",
    "teacher = teacher.to(device)\n",
    "representer = Represent(teacher)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='train')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "## orn\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "plt.savefig('norm_visual_train.pdf')\n",
    "# print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='test')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "# a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(159.5977688696812, 59464.16261313985)\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": [
    "## orn\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "\n",
    "plt.savefig('norm_visual_test.pdf')\n",
    "print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "19.821387537713285"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "59464.16261313985 / 3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "12065.616304417805"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "159.5977688696812 / 19.821387537713285 /2 * 2997"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SRIP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = 'cuda:0'\n",
    "\n",
    "# teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= True)\n",
    "teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= False)\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-t-bn/sgd/step/kaiming_both_none_0.1_0.1_0.1_0.1_seed1_False.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_none_0.01_0.01_0.01_0.01_seed3.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_norm_0.1_0.1_0.1_0.1_seed1_True.pt'\n",
    "filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/kaiming_both_srip_0.1_0.1_0.1_0.1_seed1.pt' # ortho\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_ort_0.1_0.1_0.1_0.1_seed1_True.pt' #norm\n",
    "\n",
    "checkpoint = torch.load(filename, map_location=device)['best_wts']\n",
    "teacher.load_state_dict(checkpoint, strict=False)\n",
    "teacher.eval()\n",
    "teacher = teacher.to(device)\n",
    "representer = Represent(teacher)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='train')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "## ortho\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.rcParams.update({'font.size': 18})\n",
    "\n",
    "plt.savefig('srip_visual_train.pdf')\n",
    "# print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='test')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "# a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(151.23108302328515, 63581.0422315414)\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": [
    "## ortho\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.savefig('srip_visual_test.pdf')\n",
    "print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "21.193680743847136"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "63581.0422315414 / 3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10692.799455148166"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "151.23108302328515 / 21.193680743847136/2 * 2997\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# srip +"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = 'cuda:0'\n",
    "\n",
    "# teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= True)\n",
    "teacher = ResNet(opt = 'none', init = 'kaiming', num_classes = 10, batchnorm = True, width = 4, depth = 50, bottleneck= False)\n",
    "\n",
    "# filename = './checkpoint/cifar10/res-50-4-t-bn/sgd/step/kaiming_both_none_0.1_0.1_0.1_0.1_seed1_False.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_none_0.01_0.01_0.01_0.01_seed3.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_norm_0.1_0.1_0.1_0.1_seed1_True.pt'\n",
    "# filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/ort_both_ortho_0.01_0.01_0.01_0.01_seed4_True.pt' # ortho\n",
    "filename = './checkpoint/cifar10/res-50-4-f-bn/sgd/step/xavier_both_srip_0.1_0.1_0.1_0.1_seed1.pt' #norm\n",
    "\n",
    "checkpoint = torch.load(filename, map_location=device)['best_wts']\n",
    "teacher.load_state_dict(checkpoint, strict=False)\n",
    "teacher.eval()\n",
    "teacher = teacher.to(device)\n",
    "representer = Represent(teacher)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='train')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "## norm\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.savefig('srip_p_visual_train.pdf')\n",
    "# print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list, out_list, pred_list, score_list, logit_sum_list = get_statistics(dataloaders, device, representer, teacher, mode='test')\n",
    "# singular_dict, v_ortho_dict = get_singular_value_vector(label_list, out_list)\n",
    "v_ortho_dict = teacher.fc.weight\n",
    "out, sing_lbl = get_plane(v_ortho_dict,out_list, label_list, device)\n",
    "# a, b = make_orthonormal_basis(out)\n",
    "x, y = projection(out, a, b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(145.01338327096437, 60335.325587749496)\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": [
    "## norm\n",
    "plt.plot(x[label_list[sing_lbl]==0],y[label_list[sing_lbl]==0],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==3],y[label_list[sing_lbl]==3],'.')\n",
    "plt.plot(x[label_list[sing_lbl]==6],y[label_list[sing_lbl]==6],'.')\n",
    "plt.xlim(-40,40)\n",
    "plt.ylim(-40,40)\n",
    "plt.savefig('srip_p_visual_test.pdf')\n",
    "print(verify_cluster(x,y,label_list, sing_lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "20.111775195916497"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "60335.325587749496 / 3000 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10804.742630360213"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "145.01338327096437 / 20.111775195916497/2 * 2997\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 240,
   "metadata": {},
   "outputs": [],
   "source": [
    "vanilla_norm = np.linalg.norm(out_list, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 225,
   "metadata": {},
   "outputs": [],
   "source": [
    "orn_norm = np.linalg.norm(out_list, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 241,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([2.200e+01, 3.350e+02, 2.222e+03, 3.797e+03, 2.482e+03, 8.630e+02,\n",
       "        2.230e+02, 4.500e+01, 9.000e+00, 2.000e+00]),\n",
       " array([ 9.162148, 10.817671, 12.473192, 14.128715, 15.784237, 17.439758,\n",
       "        19.095282, 20.750803, 22.406324, 24.061848, 25.71737 ],\n",
       "       dtype=float32),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 241,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "plt.hist(vanilla_norm)\n",
    "plt.hist(orn_norm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}