pyTorch-models-Accuracies-LogNorm.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##  pretrained models in PyTorch  \n",
    "\n",
    "### Nearly exact copy of WeightedAlphas notebook,  but uses LOG Norm model\n",
    "\n",
    "using this notebook for paper figures\n",
    "\n",
    "BUT\n",
    "\n",
    "### The pytorch-WeighWatchere notebook also works and could be used\n",
    "\n",
    "Its just more painful to get all the figures formatted correctly under a deadline so I include both"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:11.918002Z",
     "start_time": "2018-10-22T21:29:06.826556Z"
    }
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision.models as models\n",
    "import torchvision.transforms as transforms\n",
    "import torch.nn as nn\n",
    "\n",
    "import pandas as pd\n",
    "\n",
    "import numpy as np\n",
    "import powerlaw\n",
    "\n",
    "from sklearn.decomposition import TruncatedSVD\n",
    " \n",
    "\n",
    "from tqdm import tqdm_notebook as tqdm\n",
    "\n",
    "import matplotlib\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Models in Torch\n",
    "\n",
    "Accuracies file is taken from the Cadene website\n",
    "\n",
    "https://github.com/Cadene/pretrained-models.pytorch\n",
    "\n",
    "Although I discovered later that this package does not work consistantly so I just use the prebuilt pytorch models here\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:21.548282Z",
     "start_time": "2018-10-22T21:29:21.506963Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "62"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracies = pd.read_csv('accuracies', delimiter=\"\\t\")\n",
    "len(accuracies)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:22.194786Z",
     "start_time": "2018-10-22T21:29:22.173139Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['Tensorflow', 'ourporting', 'Caffe', 'Torch7', 'Keras', 'Pytorch'],\n",
       "      dtype=object)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracies['platform'].unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:22.635859Z",
     "start_time": "2018-10-22T21:29:22.584119Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(24,                   model    platform    acc1    acc5\n",
       " 1       PNASNet-5-Large  ourporting  82.736  95.992\n",
       " 3        NASNet-A-Large  ourporting  82.566  96.086\n",
       " 5              SENet154  ourporting  81.304  95.498\n",
       " 7               PolyNet  ourporting  81.002  95.624\n",
       " 10  SE-ResNeXt101_32x4d  ourporting  80.236  95.028\n",
       " 12    InceptionResNetV2  ourporting  80.170  95.234\n",
       " 13          InceptionV4  ourporting  80.062  94.926\n",
       " 14    DualPathNet107_5k  ourporting  79.746  94.684\n",
       " 16       DualPathNet131  ourporting  79.432  94.574\n",
       " 17     DualPathNet92_5k  ourporting  79.400  94.620\n",
       " 18        DualPathNet98  ourporting  79.224  94.488\n",
       " 19   SE-ResNeXt50_32x4d  ourporting  79.076  94.434\n",
       " 22     ResNeXt101_64x4d  ourporting  78.956  94.252\n",
       " 23             Xception  ourporting  78.888  94.292\n",
       " 26         SE-ResNet152  ourporting  78.658  94.374\n",
       " 28         SE-ResNet101  ourporting  78.396  94.258\n",
       " 30     ResNeXt101_32x4d  ourporting  78.188  93.886\n",
       " 33          SE-ResNet50  ourporting  77.636  93.752\n",
       " 36          FBResNet152  ourporting  77.386  93.594\n",
       " 39    DualPathNet68b_5k  ourporting  77.034  93.590\n",
       " 41       CaffeResnet101  ourporting  76.200  92.766\n",
       " 44        DualPathNet68  ourporting  75.868  92.774\n",
       " 48      NASNet-A-Mobile  ourporting  74.080  91.740\n",
       " 50          BNInception  ourporting  73.522  91.560)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "our_acc = accuracies[accuracies['platform']=='ourporting']\n",
    "len(our_acc), our_acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:23.152825Z",
     "start_time": "2018-10-22T21:29:23.109944Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(21,             model platform    acc1    acc5\n",
       " 27      ResNet152  Pytorch  78.428  94.110\n",
       " 34    DenseNet161  Pytorch  77.560  93.798\n",
       " 35      ResNet101  Pytorch  77.438  93.672\n",
       " 37    InceptionV3  Pytorch  77.294  93.454\n",
       " 38    DenseNet201  Pytorch  77.152  93.548\n",
       " 42    DenseNet169  Pytorch  76.026  92.992\n",
       " 43       ResNet50  Pytorch  76.002  92.980\n",
       " 45    DenseNet121  Pytorch  74.646  92.136\n",
       " 46       VGG19_BN  Pytorch  74.266  92.066\n",
       " 49       ResNet34  Pytorch  73.554  91.456\n",
       " 51       VGG16_BN  Pytorch  73.518  91.608\n",
       " 52          VGG19  Pytorch  72.080  90.822\n",
       " 53          VGG16  Pytorch  71.636  90.354\n",
       " 54       VGG13_BN  Pytorch  71.508  90.494\n",
       " 55       VGG11_BN  Pytorch  70.452  89.818\n",
       " 56       ResNet18  Pytorch  70.142  89.274\n",
       " 57          VGG13  Pytorch  69.662  89.264\n",
       " 58          VGG11  Pytorch  68.970  88.746\n",
       " 59  SqueezeNet1_1  Pytorch  58.250  80.800\n",
       " 60  SqueezeNet1_0  Pytorch  58.108  80.428\n",
       " 61        Alexnet  Pytorch  56.432  79.194)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "py_acc = accuracies[accuracies['platform']=='Pytorch']\n",
    "len(py_acc), py_acc"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- merge dataframes\n",
    "- get model names\n",
    "- load pretrained\n",
    "- compute lognorm\n",
    "- store in data frame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:24.126421Z",
     "start_time": "2018-10-22T21:29:24.087431Z"
    }
   },
   "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>model</th>\n",
       "      <th>platform</th>\n",
       "      <th>acc1</th>\n",
       "      <th>acc5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>ResNet152</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>78.428</td>\n",
       "      <td>94.110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>DenseNet161</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.560</td>\n",
       "      <td>93.798</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>ResNet101</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.438</td>\n",
       "      <td>93.672</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>InceptionV3</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.294</td>\n",
       "      <td>93.454</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>38</th>\n",
       "      <td>DenseNet201</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.152</td>\n",
       "      <td>93.548</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          model platform    acc1    acc5\n",
       "27    ResNet152  Pytorch  78.428  94.110\n",
       "34  DenseNet161  Pytorch  77.560  93.798\n",
       "35    ResNet101  Pytorch  77.438  93.672\n",
       "37  InceptionV3  Pytorch  77.294  93.454\n",
       "38  DenseNet201  Pytorch  77.152  93.548"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df =  pd.concat([py_acc, our_acc])\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Minimum dimension for W"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:25.271404Z",
     "start_time": "2018-10-22T21:29:25.262224Z"
    }
   },
   "outputs": [],
   "source": [
    "MIN_DIM = 50"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### All Linear Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_conv2D_Wmats(Wtensor):\n",
    "    \"\"\"Extract W slices from a 4 index conv2D tensor of shape: (N,M,i,j) or (M,N,i,j).  Return ij (N x M) matrices\"\"\"\n",
    "    Wmats = []\n",
    "    s = Wtensor.shape\n",
    "    N, M, imax, jmax = s[0],s[1],s[2],s[3]\n",
    "    #print(\"tensor shape\", N,M,imax,jmax)\n",
    "    for i in range(imax):\n",
    "        for j in range(jmax):\n",
    "            W = Wtensor[:,:,i,j]\n",
    "            if N < M:\n",
    "                W = W.T\n",
    "            Wmats.append(W)\n",
    "    return Wmats"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:26.344742Z",
     "start_time": "2018-10-22T21:29:26.314026Z"
    }
   },
   "outputs": [],
   "source": [
    "def count_layers(model):\n",
    "    \"\"\"Loops over a pytorch model and counts all the weight matrices, linear, conv2d, and conv2d-slices\"\"\"\n",
    "    num_linear, num_conv2d, num_conv2d_slices = 0, 0, 0\n",
    "    \n",
    "    for im, m in enumerate(model.modules()):\n",
    "        #print(im,m)\n",
    "        if isinstance(m, nn.Linear):    \n",
    "            #print(\"linear\")\n",
    "            W = np.array(m.weight.data.clone().cpu())\n",
    "            if W is not None:\n",
    "                #print(im,m, W.shape)\n",
    "                num_linear += 1\n",
    "        elif isinstance(m, nn.Conv2d):\n",
    "            #print(\"conv2d\")\n",
    "            Wtensor = np.array(m.weight.data.clone().cpu())\n",
    "            #print(Wtensor.shape)\n",
    "            Wmats = get_conv2D_Wmats(Wtensor)\n",
    "            #print(im,m)\n",
    "            num_conv2d += 1\n",
    "            for W in Wmats:\n",
    "                #print(W.shape)\n",
    "                if (np.min(W.shape) >= MIN_DIM):\n",
    "                    num_conv2d_slices += 1\n",
    "                \n",
    "    return num_linear, num_conv2d, num_conv2d_slices\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:29:27.441606Z",
     "start_time": "2018-10-22T21:29:27.411156Z"
    }
   },
   "outputs": [],
   "source": [
    "def iter_pytorch_layers(model, min_dim=MIN_DIM):\n",
    "    \"\"\"Iterate over a pytorch layers and  weight matrix.  \n",
    "    Supports linear, conv2d, and conv2d-slices\"\"\"\n",
    "    for im, m in enumerate(model.modules()):\n",
    "        if isinstance(m, nn.Linear):     \n",
    "            W = np.array(m.weight.data.clone().cpu())\n",
    "            if W is not None:\n",
    "                yield W\n",
    "        elif isinstance(m, nn.Conv2d):\n",
    "            Wtensor = np.array(m.weight.data.clone().cpu())\n",
    "            Wmats = get_conv2D_Wmats(Wtensor)\n",
    "            for W in Wmats:\n",
    "                if (np.min(W.shape) >= min_dim):\n",
    "                    yield W"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:30:00.199636Z",
     "start_time": "2018-10-22T21:30:00.179436Z"
    }
   },
   "outputs": [],
   "source": [
    "def calc_model_lognorms(model):\n",
    "    try:\n",
    "        log_norms = []\n",
    "\n",
    "        for W in tqdm(iter_pytorch_layers(model)):            \n",
    "            log_norm = np.log10(np.linalg.norm(W))\n",
    "            log_norms.append(log_norm)\n",
    "\n",
    "    except Exception as e: \n",
    "\n",
    "        print(\"skipping \", model_name)\n",
    "        print(e)\n",
    "\n",
    "        pass\n",
    "    \n",
    "    return log_norms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:32:36.320118Z",
     "start_time": "2018-10-22T21:30:02.517846Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "vgg11\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e5fc852078c7440ca23f25abd08224db",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg11_bn\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "394980eb969449d5b3f10dd92020ea49",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg13\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8ce2074cfc774a468b1068e4bee094fd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg13_bn\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b350554182fa48a79b177b03e33fcc7d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg16\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d6c81002b32d4e79b6fdba3cff4d567e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg16_bn\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e3067a8b486f4621b4871936b185be07",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg19\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d7e412860b6e4f54836561172d1dbcce",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "vgg19_bn\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8c0b916e0d00407882f42aee46ab5bb5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=1, bar_style='info', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "w_alphas_4model = {}\n",
    "counts_4model = {}\n",
    "#models = [m.lower().replace('-','') for m in df['model'].values]\n",
    "model_names = []\n",
    "\n",
    "def update_model(model_name, model):\n",
    "    print(model_name)\n",
    "    w_alphas_4model[model_name] = calc_model_lognorms(model)\n",
    "    counts_4model[model_name] = count_layers(model)\n",
    "    model_names.append(model_name)\n",
    "    \n",
    "    \n",
    "try:\n",
    "    model_name = 'vgg11'\n",
    "    model = models.vgg11(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "   \n",
    "    model_name = 'vgg11_bn'\n",
    "    model = models.vgg11_bn(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "    \n",
    "    model_name = 'vgg13'\n",
    "    model = models.vgg13(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "   \n",
    "    model_name = 'vgg13_bn'\n",
    "    model = models.vgg13_bn(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "    \n",
    "    model_name = 'vgg16'\n",
    "    model = models.vgg16(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "   \n",
    "    model_name = 'vgg16_bn'\n",
    "    model = models.vgg16_bn(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "    \n",
    "    model_name = 'vgg19'\n",
    "    model = models.vgg19(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "   \n",
    "    model_name = 'vgg19_bn'\n",
    "    model = models.vgg19_bn(pretrained=True)\n",
    "    update_model(model_name, model)\n",
    "    \n",
    "\n",
    "    \n",
    "except Exception as e: \n",
    "\n",
    "    print(\"skipping \", model_name)\n",
    "    print(e)\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:32:46.115759Z",
     "start_time": "2018-10-22T21:32:46.076314Z"
    }
   },
   "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>model</th>\n",
       "      <th>platform</th>\n",
       "      <th>acc1</th>\n",
       "      <th>acc5</th>\n",
       "      <th>modelname</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>ResNet152</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>78.428</td>\n",
       "      <td>94.110</td>\n",
       "      <td>resnet152</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>DenseNet161</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.560</td>\n",
       "      <td>93.798</td>\n",
       "      <td>densenet161</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>ResNet101</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.438</td>\n",
       "      <td>93.672</td>\n",
       "      <td>resnet101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>InceptionV3</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.294</td>\n",
       "      <td>93.454</td>\n",
       "      <td>inceptionv3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>38</th>\n",
       "      <td>DenseNet201</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.152</td>\n",
       "      <td>93.548</td>\n",
       "      <td>densenet201</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          model platform    acc1    acc5    modelname\n",
       "27    ResNet152  Pytorch  78.428  94.110    resnet152\n",
       "34  DenseNet161  Pytorch  77.560  93.798  densenet161\n",
       "35    ResNet101  Pytorch  77.438  93.672    resnet101\n",
       "37  InceptionV3  Pytorch  77.294  93.454  inceptionv3\n",
       "38  DenseNet201  Pytorch  77.152  93.548  densenet201"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:32:48.706396Z",
     "start_time": "2018-10-22T21:32:48.624300Z"
    }
   },
   "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>model</th>\n",
       "      <th>platform</th>\n",
       "      <th>acc1</th>\n",
       "      <th>acc5</th>\n",
       "      <th>modelname</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>ResNet152</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>78.428</td>\n",
       "      <td>94.110</td>\n",
       "      <td>resnet152</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>DenseNet161</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.560</td>\n",
       "      <td>93.798</td>\n",
       "      <td>densenet161</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>ResNet101</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.438</td>\n",
       "      <td>93.672</td>\n",
       "      <td>resnet101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>InceptionV3</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.294</td>\n",
       "      <td>93.454</td>\n",
       "      <td>inceptionv3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>38</th>\n",
       "      <td>DenseNet201</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.152</td>\n",
       "      <td>93.548</td>\n",
       "      <td>densenet201</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>42</th>\n",
       "      <td>DenseNet169</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>76.026</td>\n",
       "      <td>92.992</td>\n",
       "      <td>densenet169</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>43</th>\n",
       "      <td>ResNet50</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>76.002</td>\n",
       "      <td>92.980</td>\n",
       "      <td>resnet50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>45</th>\n",
       "      <td>DenseNet121</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>74.646</td>\n",
       "      <td>92.136</td>\n",
       "      <td>densenet121</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>46</th>\n",
       "      <td>VGG19_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>74.266</td>\n",
       "      <td>92.066</td>\n",
       "      <td>vgg19_bn</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49</th>\n",
       "      <td>ResNet34</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>73.554</td>\n",
       "      <td>91.456</td>\n",
       "      <td>resnet34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>51</th>\n",
       "      <td>VGG16_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>73.518</td>\n",
       "      <td>91.608</td>\n",
       "      <td>vgg16_bn</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>52</th>\n",
       "      <td>VGG19</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>72.080</td>\n",
       "      <td>90.822</td>\n",
       "      <td>vgg19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>VGG16</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>71.636</td>\n",
       "      <td>90.354</td>\n",
       "      <td>vgg16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54</th>\n",
       "      <td>VGG13_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>71.508</td>\n",
       "      <td>90.494</td>\n",
       "      <td>vgg13_bn</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>55</th>\n",
       "      <td>VGG11_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>70.452</td>\n",
       "      <td>89.818</td>\n",
       "      <td>vgg11_bn</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>56</th>\n",
       "      <td>ResNet18</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>70.142</td>\n",
       "      <td>89.274</td>\n",
       "      <td>resnet18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>57</th>\n",
       "      <td>VGG13</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>69.662</td>\n",
       "      <td>89.264</td>\n",
       "      <td>vgg13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>58</th>\n",
       "      <td>VGG11</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>68.970</td>\n",
       "      <td>88.746</td>\n",
       "      <td>vgg11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>59</th>\n",
       "      <td>SqueezeNet1_1</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>58.250</td>\n",
       "      <td>80.800</td>\n",
       "      <td>squeezenet1_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>60</th>\n",
       "      <td>SqueezeNet1_0</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>58.108</td>\n",
       "      <td>80.428</td>\n",
       "      <td>squeezenet1_0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>61</th>\n",
       "      <td>Alexnet</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>56.432</td>\n",
       "      <td>79.194</td>\n",
       "      <td>alexnet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>PNASNet-5-Large</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>82.736</td>\n",
       "      <td>95.992</td>\n",
       "      <td>pnasnet5large</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>NASNet-A-Large</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>82.566</td>\n",
       "      <td>96.086</td>\n",
       "      <td>nasnetalarge</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>SENet154</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>81.304</td>\n",
       "      <td>95.498</td>\n",
       "      <td>senet154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>PolyNet</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>81.002</td>\n",
       "      <td>95.624</td>\n",
       "      <td>polynet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>SE-ResNeXt101_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.236</td>\n",
       "      <td>95.028</td>\n",
       "      <td>seresnext101_32x4d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>InceptionResNetV2</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.170</td>\n",
       "      <td>95.234</td>\n",
       "      <td>inceptionresnetv2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>InceptionV4</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.062</td>\n",
       "      <td>94.926</td>\n",
       "      <td>inceptionv4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>DualPathNet107_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.746</td>\n",
       "      <td>94.684</td>\n",
       "      <td>dualpathnet107_5k</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>DualPathNet131</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.432</td>\n",
       "      <td>94.574</td>\n",
       "      <td>dualpathnet131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>DualPathNet92_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.400</td>\n",
       "      <td>94.620</td>\n",
       "      <td>dualpathnet92_5k</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>DualPathNet98</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.224</td>\n",
       "      <td>94.488</td>\n",
       "      <td>dualpathnet98</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>SE-ResNeXt50_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.076</td>\n",
       "      <td>94.434</td>\n",
       "      <td>seresnext50_32x4d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>ResNeXt101_64x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.956</td>\n",
       "      <td>94.252</td>\n",
       "      <td>resnext101_64x4d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>Xception</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.888</td>\n",
       "      <td>94.292</td>\n",
       "      <td>xception</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>SE-ResNet152</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.658</td>\n",
       "      <td>94.374</td>\n",
       "      <td>seresnet152</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>SE-ResNet101</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.396</td>\n",
       "      <td>94.258</td>\n",
       "      <td>seresnet101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>ResNeXt101_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.188</td>\n",
       "      <td>93.886</td>\n",
       "      <td>resnext101_32x4d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>33</th>\n",
       "      <td>SE-ResNet50</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.636</td>\n",
       "      <td>93.752</td>\n",
       "      <td>seresnet50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>36</th>\n",
       "      <td>FBResNet152</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.386</td>\n",
       "      <td>93.594</td>\n",
       "      <td>fbresnet152</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>39</th>\n",
       "      <td>DualPathNet68b_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.034</td>\n",
       "      <td>93.590</td>\n",
       "      <td>dualpathnet68b_5k</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>CaffeResnet101</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>76.200</td>\n",
       "      <td>92.766</td>\n",
       "      <td>cafferesnet101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>44</th>\n",
       "      <td>DualPathNet68</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>75.868</td>\n",
       "      <td>92.774</td>\n",
       "      <td>dualpathnet68</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>NASNet-A-Mobile</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>74.080</td>\n",
       "      <td>91.740</td>\n",
       "      <td>nasnetamobile</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>BNInception</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>73.522</td>\n",
       "      <td>91.560</td>\n",
       "      <td>bninception</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  model    platform    acc1    acc5           modelname\n",
       "27            ResNet152     Pytorch  78.428  94.110           resnet152\n",
       "34          DenseNet161     Pytorch  77.560  93.798         densenet161\n",
       "35            ResNet101     Pytorch  77.438  93.672           resnet101\n",
       "37          InceptionV3     Pytorch  77.294  93.454         inceptionv3\n",
       "38          DenseNet201     Pytorch  77.152  93.548         densenet201\n",
       "42          DenseNet169     Pytorch  76.026  92.992         densenet169\n",
       "43             ResNet50     Pytorch  76.002  92.980            resnet50\n",
       "45          DenseNet121     Pytorch  74.646  92.136         densenet121\n",
       "46             VGG19_BN     Pytorch  74.266  92.066            vgg19_bn\n",
       "49             ResNet34     Pytorch  73.554  91.456            resnet34\n",
       "51             VGG16_BN     Pytorch  73.518  91.608            vgg16_bn\n",
       "52                VGG19     Pytorch  72.080  90.822               vgg19\n",
       "53                VGG16     Pytorch  71.636  90.354               vgg16\n",
       "54             VGG13_BN     Pytorch  71.508  90.494            vgg13_bn\n",
       "55             VGG11_BN     Pytorch  70.452  89.818            vgg11_bn\n",
       "56             ResNet18     Pytorch  70.142  89.274            resnet18\n",
       "57                VGG13     Pytorch  69.662  89.264               vgg13\n",
       "58                VGG11     Pytorch  68.970  88.746               vgg11\n",
       "59        SqueezeNet1_1     Pytorch  58.250  80.800       squeezenet1_1\n",
       "60        SqueezeNet1_0     Pytorch  58.108  80.428       squeezenet1_0\n",
       "61              Alexnet     Pytorch  56.432  79.194             alexnet\n",
       "1       PNASNet-5-Large  ourporting  82.736  95.992       pnasnet5large\n",
       "3        NASNet-A-Large  ourporting  82.566  96.086        nasnetalarge\n",
       "5              SENet154  ourporting  81.304  95.498            senet154\n",
       "7               PolyNet  ourporting  81.002  95.624             polynet\n",
       "10  SE-ResNeXt101_32x4d  ourporting  80.236  95.028  seresnext101_32x4d\n",
       "12    InceptionResNetV2  ourporting  80.170  95.234   inceptionresnetv2\n",
       "13          InceptionV4  ourporting  80.062  94.926         inceptionv4\n",
       "14    DualPathNet107_5k  ourporting  79.746  94.684   dualpathnet107_5k\n",
       "16       DualPathNet131  ourporting  79.432  94.574      dualpathnet131\n",
       "17     DualPathNet92_5k  ourporting  79.400  94.620    dualpathnet92_5k\n",
       "18        DualPathNet98  ourporting  79.224  94.488       dualpathnet98\n",
       "19   SE-ResNeXt50_32x4d  ourporting  79.076  94.434   seresnext50_32x4d\n",
       "22     ResNeXt101_64x4d  ourporting  78.956  94.252    resnext101_64x4d\n",
       "23             Xception  ourporting  78.888  94.292            xception\n",
       "26         SE-ResNet152  ourporting  78.658  94.374         seresnet152\n",
       "28         SE-ResNet101  ourporting  78.396  94.258         seresnet101\n",
       "30     ResNeXt101_32x4d  ourporting  78.188  93.886    resnext101_32x4d\n",
       "33          SE-ResNet50  ourporting  77.636  93.752          seresnet50\n",
       "36          FBResNet152  ourporting  77.386  93.594         fbresnet152\n",
       "39    DualPathNet68b_5k  ourporting  77.034  93.590   dualpathnet68b_5k\n",
       "41       CaffeResnet101  ourporting  76.200  92.766      cafferesnet101\n",
       "44        DualPathNet68  ourporting  75.868  92.774       dualpathnet68\n",
       "48      NASNet-A-Mobile  ourporting  74.080  91.740       nasnetamobile\n",
       "50          BNInception  ourporting  73.522  91.560         bninception"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['modelname'] = df['model'].apply(lambda x: x.lower().replace('-',''))\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:32:52.793310Z",
     "start_time": "2018-10-22T21:32:52.769740Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "27             resnet152\n",
       "34           densenet161\n",
       "35             resnet101\n",
       "37           inceptionv3\n",
       "38           densenet201\n",
       "42           densenet169\n",
       "43              resnet50\n",
       "45           densenet121\n",
       "46              vgg19_bn\n",
       "49              resnet34\n",
       "51              vgg16_bn\n",
       "52                 vgg19\n",
       "53                 vgg16\n",
       "54              vgg13_bn\n",
       "55              vgg11_bn\n",
       "56              resnet18\n",
       "57                 vgg13\n",
       "58                 vgg11\n",
       "59         squeezenet1_1\n",
       "60         squeezenet1_0\n",
       "61               alexnet\n",
       "1          pnasnet5large\n",
       "3           nasnetalarge\n",
       "5               senet154\n",
       "7                polynet\n",
       "10    seresnext101_32x4d\n",
       "12     inceptionresnetv2\n",
       "13           inceptionv4\n",
       "14     dualpathnet107_5k\n",
       "16        dualpathnet131\n",
       "17      dualpathnet92_5k\n",
       "18         dualpathnet98\n",
       "19     seresnext50_32x4d\n",
       "22      resnext101_64x4d\n",
       "23              xception\n",
       "26           seresnet152\n",
       "28           seresnet101\n",
       "30      resnext101_32x4d\n",
       "33            seresnet50\n",
       "36           fbresnet152\n",
       "39     dualpathnet68b_5k\n",
       "41        cafferesnet101\n",
       "44         dualpathnet68\n",
       "48         nasnetamobile\n",
       "50           bninception\n",
       "Name: modelname, dtype: object"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['modelname']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:32:54.226821Z",
     "start_time": "2018-10-22T21:32:54.201505Z"
    }
   },
   "outputs": [],
   "source": [
    "avg_w_alphas = {}\n",
    "len_w_alphas = {}\n",
    "\n",
    "for model_name in df['modelname'].values:\n",
    "    avg_w_alphas[model_name] = 0.0\n",
    "    len_w_alphas[model_name] = 0.0\n",
    "    \n",
    "    if(model_name in w_alphas_4model):\n",
    "        avg_w_alphas[model_name] = np.average(w_alphas_4model[model_name])\n",
    "        len_w_alphas[model_name] = len(w_alphas_4model[model_name])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:33:28.433606Z",
     "start_time": "2018-10-22T21:33:28.410000Z"
    }
   },
   "outputs": [],
   "source": [
    "df['avg_w_alphas'] = df['modelname'].apply(lambda x: avg_w_alphas[x] )\n",
    "df['num_layers'] = df['modelname'].apply(lambda x: int(len_w_alphas[x]) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:33:30.176440Z",
     "start_time": "2018-10-22T21:33:30.058643Z"
    }
   },
   "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>model</th>\n",
       "      <th>platform</th>\n",
       "      <th>acc1</th>\n",
       "      <th>acc5</th>\n",
       "      <th>modelname</th>\n",
       "      <th>avg_w_alphas</th>\n",
       "      <th>num_layers</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>61</th>\n",
       "      <td>Alexnet</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>56.432</td>\n",
       "      <td>79.194</td>\n",
       "      <td>alexnet</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>BNInception</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>73.522</td>\n",
       "      <td>91.560</td>\n",
       "      <td>bninception</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>CaffeResnet101</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>76.200</td>\n",
       "      <td>92.766</td>\n",
       "      <td>cafferesnet101</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>45</th>\n",
       "      <td>DenseNet121</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>74.646</td>\n",
       "      <td>92.136</td>\n",
       "      <td>densenet121</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>DenseNet161</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.560</td>\n",
       "      <td>93.798</td>\n",
       "      <td>densenet161</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>42</th>\n",
       "      <td>DenseNet169</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>76.026</td>\n",
       "      <td>92.992</td>\n",
       "      <td>densenet169</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>38</th>\n",
       "      <td>DenseNet201</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.152</td>\n",
       "      <td>93.548</td>\n",
       "      <td>densenet201</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>DualPathNet107_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.746</td>\n",
       "      <td>94.684</td>\n",
       "      <td>dualpathnet107_5k</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>DualPathNet131</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.432</td>\n",
       "      <td>94.574</td>\n",
       "      <td>dualpathnet131</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>44</th>\n",
       "      <td>DualPathNet68</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>75.868</td>\n",
       "      <td>92.774</td>\n",
       "      <td>dualpathnet68</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>39</th>\n",
       "      <td>DualPathNet68b_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.034</td>\n",
       "      <td>93.590</td>\n",
       "      <td>dualpathnet68b_5k</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>DualPathNet92_5k</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.400</td>\n",
       "      <td>94.620</td>\n",
       "      <td>dualpathnet92_5k</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>DualPathNet98</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.224</td>\n",
       "      <td>94.488</td>\n",
       "      <td>dualpathnet98</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>36</th>\n",
       "      <td>FBResNet152</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.386</td>\n",
       "      <td>93.594</td>\n",
       "      <td>fbresnet152</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>InceptionResNetV2</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.170</td>\n",
       "      <td>95.234</td>\n",
       "      <td>inceptionresnetv2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>InceptionV3</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.294</td>\n",
       "      <td>93.454</td>\n",
       "      <td>inceptionv3</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>InceptionV4</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.062</td>\n",
       "      <td>94.926</td>\n",
       "      <td>inceptionv4</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>NASNet-A-Large</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>82.566</td>\n",
       "      <td>96.086</td>\n",
       "      <td>nasnetalarge</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>NASNet-A-Mobile</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>74.080</td>\n",
       "      <td>91.740</td>\n",
       "      <td>nasnetamobile</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>PNASNet-5-Large</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>82.736</td>\n",
       "      <td>95.992</td>\n",
       "      <td>pnasnet5large</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>PolyNet</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>81.002</td>\n",
       "      <td>95.624</td>\n",
       "      <td>polynet</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>ResNeXt101_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.188</td>\n",
       "      <td>93.886</td>\n",
       "      <td>resnext101_32x4d</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>ResNeXt101_64x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.956</td>\n",
       "      <td>94.252</td>\n",
       "      <td>resnext101_64x4d</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>ResNet101</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>77.438</td>\n",
       "      <td>93.672</td>\n",
       "      <td>resnet101</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>ResNet152</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>78.428</td>\n",
       "      <td>94.110</td>\n",
       "      <td>resnet152</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>56</th>\n",
       "      <td>ResNet18</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>70.142</td>\n",
       "      <td>89.274</td>\n",
       "      <td>resnet18</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49</th>\n",
       "      <td>ResNet34</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>73.554</td>\n",
       "      <td>91.456</td>\n",
       "      <td>resnet34</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>43</th>\n",
       "      <td>ResNet50</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>76.002</td>\n",
       "      <td>92.980</td>\n",
       "      <td>resnet50</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>SE-ResNeXt101_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>80.236</td>\n",
       "      <td>95.028</td>\n",
       "      <td>seresnext101_32x4d</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>SE-ResNeXt50_32x4d</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>79.076</td>\n",
       "      <td>94.434</td>\n",
       "      <td>seresnext50_32x4d</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>SE-ResNet101</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.396</td>\n",
       "      <td>94.258</td>\n",
       "      <td>seresnet101</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>SE-ResNet152</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.658</td>\n",
       "      <td>94.374</td>\n",
       "      <td>seresnet152</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>33</th>\n",
       "      <td>SE-ResNet50</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>77.636</td>\n",
       "      <td>93.752</td>\n",
       "      <td>seresnet50</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>SENet154</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>81.304</td>\n",
       "      <td>95.498</td>\n",
       "      <td>senet154</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>60</th>\n",
       "      <td>SqueezeNet1_0</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>58.108</td>\n",
       "      <td>80.428</td>\n",
       "      <td>squeezenet1_0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>59</th>\n",
       "      <td>SqueezeNet1_1</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>58.250</td>\n",
       "      <td>80.800</td>\n",
       "      <td>squeezenet1_1</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>58</th>\n",
       "      <td>VGG11</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>68.970</td>\n",
       "      <td>88.746</td>\n",
       "      <td>vgg11</td>\n",
       "      <td>0.947488</td>\n",
       "      <td>66</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>55</th>\n",
       "      <td>VGG11_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>70.452</td>\n",
       "      <td>89.818</td>\n",
       "      <td>vgg11_bn</td>\n",
       "      <td>0.890248</td>\n",
       "      <td>66</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>57</th>\n",
       "      <td>VGG13</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>69.662</td>\n",
       "      <td>89.264</td>\n",
       "      <td>vgg13</td>\n",
       "      <td>0.864038</td>\n",
       "      <td>84</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54</th>\n",
       "      <td>VGG13_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>71.508</td>\n",
       "      <td>90.494</td>\n",
       "      <td>vgg13_bn</td>\n",
       "      <td>0.793754</td>\n",
       "      <td>84</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>VGG16</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>71.636</td>\n",
       "      <td>90.354</td>\n",
       "      <td>vgg16</td>\n",
       "      <td>0.838180</td>\n",
       "      <td>111</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>51</th>\n",
       "      <td>VGG16_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>73.518</td>\n",
       "      <td>91.608</td>\n",
       "      <td>vgg16_bn</td>\n",
       "      <td>0.770092</td>\n",
       "      <td>111</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>52</th>\n",
       "      <td>VGG19</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>72.080</td>\n",
       "      <td>90.822</td>\n",
       "      <td>vgg19</td>\n",
       "      <td>0.818500</td>\n",
       "      <td>138</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>46</th>\n",
       "      <td>VGG19_BN</td>\n",
       "      <td>Pytorch</td>\n",
       "      <td>74.266</td>\n",
       "      <td>92.066</td>\n",
       "      <td>vgg19_bn</td>\n",
       "      <td>0.736490</td>\n",
       "      <td>138</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>Xception</td>\n",
       "      <td>ourporting</td>\n",
       "      <td>78.888</td>\n",
       "      <td>94.292</td>\n",
       "      <td>xception</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  model    platform    acc1    acc5           modelname  \\\n",
       "61              Alexnet     Pytorch  56.432  79.194             alexnet   \n",
       "50          BNInception  ourporting  73.522  91.560         bninception   \n",
       "41       CaffeResnet101  ourporting  76.200  92.766      cafferesnet101   \n",
       "45          DenseNet121     Pytorch  74.646  92.136         densenet121   \n",
       "34          DenseNet161     Pytorch  77.560  93.798         densenet161   \n",
       "42          DenseNet169     Pytorch  76.026  92.992         densenet169   \n",
       "38          DenseNet201     Pytorch  77.152  93.548         densenet201   \n",
       "14    DualPathNet107_5k  ourporting  79.746  94.684   dualpathnet107_5k   \n",
       "16       DualPathNet131  ourporting  79.432  94.574      dualpathnet131   \n",
       "44        DualPathNet68  ourporting  75.868  92.774       dualpathnet68   \n",
       "39    DualPathNet68b_5k  ourporting  77.034  93.590   dualpathnet68b_5k   \n",
       "17     DualPathNet92_5k  ourporting  79.400  94.620    dualpathnet92_5k   \n",
       "18        DualPathNet98  ourporting  79.224  94.488       dualpathnet98   \n",
       "36          FBResNet152  ourporting  77.386  93.594         fbresnet152   \n",
       "12    InceptionResNetV2  ourporting  80.170  95.234   inceptionresnetv2   \n",
       "37          InceptionV3     Pytorch  77.294  93.454         inceptionv3   \n",
       "13          InceptionV4  ourporting  80.062  94.926         inceptionv4   \n",
       "3        NASNet-A-Large  ourporting  82.566  96.086        nasnetalarge   \n",
       "48      NASNet-A-Mobile  ourporting  74.080  91.740       nasnetamobile   \n",
       "1       PNASNet-5-Large  ourporting  82.736  95.992       pnasnet5large   \n",
       "7               PolyNet  ourporting  81.002  95.624             polynet   \n",
       "30     ResNeXt101_32x4d  ourporting  78.188  93.886    resnext101_32x4d   \n",
       "22     ResNeXt101_64x4d  ourporting  78.956  94.252    resnext101_64x4d   \n",
       "35            ResNet101     Pytorch  77.438  93.672           resnet101   \n",
       "27            ResNet152     Pytorch  78.428  94.110           resnet152   \n",
       "56             ResNet18     Pytorch  70.142  89.274            resnet18   \n",
       "49             ResNet34     Pytorch  73.554  91.456            resnet34   \n",
       "43             ResNet50     Pytorch  76.002  92.980            resnet50   \n",
       "10  SE-ResNeXt101_32x4d  ourporting  80.236  95.028  seresnext101_32x4d   \n",
       "19   SE-ResNeXt50_32x4d  ourporting  79.076  94.434   seresnext50_32x4d   \n",
       "28         SE-ResNet101  ourporting  78.396  94.258         seresnet101   \n",
       "26         SE-ResNet152  ourporting  78.658  94.374         seresnet152   \n",
       "33          SE-ResNet50  ourporting  77.636  93.752          seresnet50   \n",
       "5              SENet154  ourporting  81.304  95.498            senet154   \n",
       "60        SqueezeNet1_0     Pytorch  58.108  80.428       squeezenet1_0   \n",
       "59        SqueezeNet1_1     Pytorch  58.250  80.800       squeezenet1_1   \n",
       "58                VGG11     Pytorch  68.970  88.746               vgg11   \n",
       "55             VGG11_BN     Pytorch  70.452  89.818            vgg11_bn   \n",
       "57                VGG13     Pytorch  69.662  89.264               vgg13   \n",
       "54             VGG13_BN     Pytorch  71.508  90.494            vgg13_bn   \n",
       "53                VGG16     Pytorch  71.636  90.354               vgg16   \n",
       "51             VGG16_BN     Pytorch  73.518  91.608            vgg16_bn   \n",
       "52                VGG19     Pytorch  72.080  90.822               vgg19   \n",
       "46             VGG19_BN     Pytorch  74.266  92.066            vgg19_bn   \n",
       "23             Xception  ourporting  78.888  94.292            xception   \n",
       "\n",
       "    avg_w_alphas  num_layers  \n",
       "61      0.000000           0  \n",
       "50      0.000000           0  \n",
       "41      0.000000           0  \n",
       "45      0.000000           0  \n",
       "34      0.000000           0  \n",
       "42      0.000000           0  \n",
       "38      0.000000           0  \n",
       "14      0.000000           0  \n",
       "16      0.000000           0  \n",
       "44      0.000000           0  \n",
       "39      0.000000           0  \n",
       "17      0.000000           0  \n",
       "18      0.000000           0  \n",
       "36      0.000000           0  \n",
       "12      0.000000           0  \n",
       "37      0.000000           0  \n",
       "13      0.000000           0  \n",
       "3       0.000000           0  \n",
       "48      0.000000           0  \n",
       "1       0.000000           0  \n",
       "7       0.000000           0  \n",
       "30      0.000000           0  \n",
       "22      0.000000           0  \n",
       "35      0.000000           0  \n",
       "27      0.000000           0  \n",
       "56      0.000000           0  \n",
       "49      0.000000           0  \n",
       "43      0.000000           0  \n",
       "10      0.000000           0  \n",
       "19      0.000000           0  \n",
       "28      0.000000           0  \n",
       "26      0.000000           0  \n",
       "33      0.000000           0  \n",
       "5       0.000000           0  \n",
       "60      0.000000           0  \n",
       "59      0.000000           0  \n",
       "58      0.947488          66  \n",
       "55      0.890248          66  \n",
       "57      0.864038          84  \n",
       "54      0.793754          84  \n",
       "53      0.838180         111  \n",
       "51      0.770092         111  \n",
       "52      0.818500         138  \n",
       "46      0.736490         138  \n",
       "23      0.000000           0  "
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.sort_values('model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:33:33.710273Z",
     "start_time": "2018-10-22T21:33:32.844105Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/charleshmartin/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:2: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  \n",
      "/Users/charleshmartin/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  \"\"\"\n",
      "/Users/charleshmartin/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:8: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  \n",
      "/Users/charleshmartin/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:11: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  # This is added back by InteractiveShellApp.init_path()\n"
     ]
    }
   ],
   "source": [
    "df_vgg11 = df[(df['model']=='VGG11')| (df['model']=='VGG11_BN')]\n",
    "df_vgg11['legend']='VGG11'\n",
    "\n",
    "df_vgg13 = df[(df['model']=='VGG13')| (df['model']=='VGG13_BN')]\n",
    "df_vgg13['legend']='VGG13'\n",
    "\n",
    "df_vgg16 = df[(df['model']=='VGG16')| (df['model']=='VGG16_BN')]\n",
    "df_vgg16['legend']='VGG16'\n",
    "\n",
    "df_vgg19 = df[(df['model']=='VGG19')| (df['model']=='VGG19_BN')]\n",
    "df_vgg19['legend']='VGG19'\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.rcParams['figure.figsize'] = [10,10]\n",
    "matplotlib.rcParams.update({'font.size': 24})\n",
    "legend_prop={'size':20}\n",
    "marker_size = 75\n",
    "\n",
    "title = r\"Test Accuracy vs. Average Log Norm $\\langle\\log_{10}\\Vert W\\Vert_{F}\\rangle$\"+\"\\nPretrained\"\n",
    "x_label = r\"top1 Test Accuracy\"\n",
    "y_label = r\"$\\langle\\log_{10}\\Vert W\\Vert_{F}\\rangle$\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib.legend_handler import HandlerLineCollection, HandlerTuple"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-22T21:34:58.599857Z",
     "start_time": "2018-10-22T21:34:57.583864Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "colors=['blue', 'orange', 'green', 'red']\n",
    "\n",
    "plts = []\n",
    "labels = []\n",
    "for im, mf in enumerate([df_vgg11, df_vgg13, df_vgg16, df_vgg19]):\n",
    "#for mf in [df_vgg11, df_vgg13]:\n",
    "    x = mf['acc1'].values\n",
    "    y = mf['avg_w_alphas'].values\n",
    "    label = mf['legend'].values[0]\n",
    "    color = colors[im]\n",
    "    \n",
    "    p1 = plt.scatter(x[1],y[1], c=color, marker=\"o\", s=marker_size)\n",
    "    p2 = plt.scatter(x[0],y[0],  c=color,  marker=\"s\", s=marker_size)\n",
    "    \n",
    "    plts.append((p1,p2))\n",
    "    labels.append(label+\", BN\")\n",
    "\n",
    "plt.legend(plts, labels, scatterpoints=1,\n",
    "           numpoints=1, handler_map={tuple: HandlerTuple(ndivide=None)},prop=legend_prop)\n",
    "    \n",
    "plt.title(title+\"VGG and VGG_BN Models\")\n",
    "plt.xlabel(x_label)\n",
    "plt.ylabel(y_label)\n",
    "plt.savefig(\"img/vgg-lognorms.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.6.5"
  },
  "widgets": {
   "state": {
    "52ecdae62c394a67868cd3ee83e19861": {
     "views": [
      {
       "cell_index": 14
      }
     ]
    },
    "e232317699e24554b5ed09fa6c78c662": {
     "views": [
      {
       "cell_index": 8
      }
     ]
    }
   },
   "version": "1.2.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}