Adversary_MMTLF-pgd-audio.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_DEVICE_ORDER']='PCI_BUS_ID'\n",
    "os.environ['CUDA_VISIBLE_DEVICES']='0'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import numpy as np\n",
    "from torch.utils import data\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import math\n",
    "import parser\n",
    "\n",
    "import wave\n",
    "from IPython.display import Audio\n",
    "import io\n",
    "\n",
    "import torchaudio\n",
    "import librosa\n",
    "import soundfile as sf\n",
    "from torch.optim import *\n",
    "from torch.optim.lr_scheduler import *\n",
    "from torch.autograd import Variable\n",
    "from Net_mModal import Net, NewNet, TransformerEncoder, Transformer, MMTEncoder, LateFusion\n",
    "# from util_in_origin import *\n",
    "from util_in_multi_h5_unnorm import *\n",
    "from AudioResNet import resnet50\n",
    "\n",
    "import gc\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util_out import *\n",
    "from util_f1 import *\n",
    "from scipy.io import loadmat, savemat\n",
    "import argparse"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.cuda.is_available()\n",
    "torch.cuda.device_count()\n",
    "# torch.cuda.set_device(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VICTIM MODEL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "_StoreAction(option_strings=['--scheduler'], dest='scheduler', nargs=None, const=None, default='reduce', type=<class 'str'>, choices=['reduce', 'warmup-decay'], help=None, metavar=None)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Parse input arguments\n",
    "def mybool(s):\n",
    "    return s.lower() in ['t', 'true', 'y', 'yes', '1']\n",
    "parser = argparse.ArgumentParser()\n",
    "parser.add_argument('--model_type', type = str, default = None, required=True)\n",
    "parser.add_argument('--TALNet', action = 'store_true')              # specify this to evaluate the pre-trained TALNet model\n",
    "parser.add_argument('--embedding_size', type = int, default = 1024) # this is the embedding size after a pooling layer\n",
    "                                                                    # after a non-pooling layer, the embeddings size will be twice this much\n",
    "parser.add_argument('--n_conv_layers', type = int, default = 10)\n",
    "parser.add_argument('--n_trans_layers', type = int, default = 1)\n",
    "parser.add_argument('--kernel_size', type = str, default = '3x3')     # 'n' or 'nxm'\n",
    "parser.add_argument('--n_pool_layers', type = int, default = 5)     # the pooling layers will be inserted uniformly into the conv layers\n",
    "                                                                    # the should be at least 2 and at most 6 pooling layers\n",
    "                                                                    # the first two pooling layers will have stride (2,2); later ones will have stride (1,2)\n",
    "parser.add_argument('--batch_norm', type = mybool, default = True)\n",
    "parser.add_argument('--dropout', type = float, default = 0.0)\n",
    "parser.add_argument('--pooling', type = str, default = 'lin', choices = ['max', 'ave', 'lin', 'exp', 'att'])\n",
    "parser.add_argument('--batch_size', type = int, default = 250)\n",
    "parser.add_argument('--ckpt_size', type = int, default = 1000)      # how many batches per checkpoint\n",
    "parser.add_argument('--optimizer', type = str, default = 'adam', choices = ['adam', 'sgd'])\n",
    "parser.add_argument('--init_lr', type = float, default = 1e-3)\n",
    "parser.add_argument('--weight_decay', type = float, default = 0)\n",
    "parser.add_argument('--beta1', type = float, default = 0.9)\n",
    "parser.add_argument('--beta2', type = float, default = 0.999)\n",
    "parser.add_argument('--lr_patience', type = int, default = 3)\n",
    "parser.add_argument('--lr_factor', type = float, default = 0.8)\n",
    "parser.add_argument('--random_seed', type = int, default = 15213)\n",
    "parser.add_argument('--ckpt', type = int)\n",
    "parser.add_argument('--additional_outname', type = str, default = '')\n",
    "parser.add_argument('--addpos', type = mybool, default = False)\n",
    "parser.add_argument('--transformer_dropout', type = float, default = 0.5)\n",
    "parser.add_argument('--gradient_accumulation', type = int, default = 1)\n",
    "parser.add_argument('--scheduler', type = str, default = 'reduce', choices = ['reduce', 'warmup-decay'])\n",
    "\n",
    "###MODEL INPUT PARAMS###\n",
    "# args = parser.parse_args(args=['--model_type=TAL-trans','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=2', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=18'\n",
    "# ])\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# python train_multimodal_late_fusion.py --model_type MMTLF --embedding_size 1024 --batch_size 100 --ckpt_size 2500 --init_lr 4e-4 --pooling att --max_ckpt 30 --n_trans_layers 2 --gradient_accumulation 3 --transformer_dropout 0.75 --additional_outname dadada --scheduler warmup-decay \n",
    "# MMTLF-embed1024-10C5P-kernel3x3-bn-drop0.0-att-batch100-ckpt2500-adam-lr4e-04-pat3-fac0.8-seed15213-Trans2-weight-decay0.00000000-betas0.900-0.999late_fusion_all_additional_linear"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = parser.parse_args(args=['--model_type=MMTLF','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=1', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=19', '--additional_outname=late_fusion_all_additional_linear'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = parser.parse_args(args=['--model_type=TAL-new','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=2', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=18', '--additional_outname='])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#Ablation Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = parser.parse_args(args=['--model_type=TAL-trans','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=2', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=19', '--additional_outname=sixteenhead_py3'\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = parser.parse_args(args=['--model_type=TAL-trans','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=2', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=18', '--additional_outname=nores_py3'\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = parser.parse_args(args=['--model_type=TAL-trans','--init_lr=4e-4','--embedding_size=1024','--pooling=att','--n_trans_layers=2', '--batch_size=100', '--ckpt_size=2500', '--n_conv_layers=10', '--n_pool_layers=5', '--ckpt=18','--additional_outname='\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "if 'x' not in args.kernel_size:\n",
    "    args.kernel_size = args.kernel_size + 'x' + args.kernel_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Locate model file and prepare directories for prediction and evaluation\n",
    "expid = '%s-embed%d-%dC%dP-kernel%s-%s-drop%.1f-%s-batch%d-ckpt%d-%s-lr%.0e-pat%d-fac%.1f-seed%d-Trans%d-weight-decay%.8f-betas%.3f-%.3f' % (\n",
    "    args.model_type,\n",
    "    args.embedding_size,\n",
    "    args.n_conv_layers,\n",
    "    args.n_pool_layers,\n",
    "    args.kernel_size,\n",
    "    'bn' if args.batch_norm else 'nobn',\n",
    "    args.dropout,\n",
    "    args.pooling,\n",
    "    args.batch_size,\n",
    "    args.ckpt_size,\n",
    "    args.optimizer,\n",
    "    args.init_lr,\n",
    "    args.lr_patience,\n",
    "    args.lr_factor,\n",
    "    args.random_seed,\n",
    "    args.n_trans_layers,\n",
    "    args.weight_decay,\n",
    "    args.beta1, \n",
    "    args.beta2\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "expid += args.additional_outname\n",
    "WORKSPACE = os.path.join('/jet/home/billyli/data_folder/TAL-kaixin_best_model/', expid)\n",
    "PRED_PATH = os.path.join(WORKSPACE, 'pred')\n",
    "if not os.path.exists(PRED_PATH): os.makedirs(PRED_PATH)\n",
    "EVAL_PATH = os.path.join(WORKSPACE, 'eval')\n",
    "if not os.path.exists(EVAL_PATH): os.makedirs(EVAL_PATH)\n",
    "if args.TALNet:\n",
    "    WORKSPACE = '../../workspace/audioset/embed1024-10C5P-kernel3x3-bn-drop0.0-lin-batch250-ckpt1000-adam-lr1e-03-pat3-fac0.8-seed15213'\n",
    "    PRED_PATH = os.path.join(WORKSPACE, 'pred')\n",
    "    EVAL_PATH = os.path.join(WORKSPACE, 'eval')\n",
    "    MODEL_FILE = os.path.join(WORKSPACE, 'model', 'checkpoint15.pt')\n",
    "    PRED_FILE = os.path.join(PRED_PATH, 'checkpoint15.mat')\n",
    "    EVAL_FILE = os.path.join(EVAL_PATH, 'checkpoint15.txt')\n",
    "elif args.model_type == 'MMTLF':\n",
    "#     WORKSPACE = '../../workspace/audioset/MMTLF-embed1024-10C5P-kernel3x3-bn-drop0.0-att-batch100-ckpt2500-adam-lr4e-04-pat3-fac0.8-seed15213-Trans2-weight-decay0.00000000-betas0.900-0.999late_fusion_all_additional_linear/'\n",
    "    WORKSPACE = '../../workspace/audioset/MMTLF-embed1024-10C5P-kernel3x3-bn-drop0.0-att-batch100-ckpt2500-adam-lr4e-04-pat3-fac0.8-seed15213-Trans2-weight-decay0.00000000-betas0.900-0.999late_fusion_all_additional_linear/'\n",
    "    MODEL_FILE = os.path.join(WORKSPACE, 'model', 'checkpoint5.pt')\n",
    "else:\n",
    "    MODEL_FILE = os.path.join(WORKSPACE, 'model', 'checkpoint%d.pt' % args.ckpt)\n",
    "    PRED_FILE = os.path.join(PRED_PATH, 'checkpoint%d.mat' % args.ckpt)\n",
    "    EVAL_FILE = os.path.join(EVAL_PATH, 'checkpoint%d.txt' % args.ckpt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Namespace(TALNet=False, additional_outname='sixteenhead_py3', addpos=False, batch_norm=True, batch_size=100, beta1=0.9, beta2=0.999, ckpt=19, ckpt_size=2500, dropout=0.0, embedding_size=1024, gradient_accumulation=1, init_lr=0.0004, kernel_size='3x3', lr_factor=0.8, lr_patience=3, model_type='TAL-trans', n_conv_layers=10, n_pool_layers=5, n_trans_layers=2, optimizer='adam', pooling='att', random_seed=15213, scheduler='reduce', transformer_dropout=0.5, weight_decay=0)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "args"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/jet/home/billyli/data_folder/TAL-kaixin_best_model/TAL-trans-embed1024-10C5P-kernel3x3-bn-drop0.0-att-batch100-ckpt2500-adam-lr4e-04-pat3-fac0.8-seed15213-Trans2-weight-decay0.00000000-betas0.900-0.999sixteenhead_py3/model/checkpoint19.pt'"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MODEL_FILE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<All keys matched successfully>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Load model\n",
    "args.kernel_size = tuple(int(x) for x in args.kernel_size.split('x'))\n",
    "#model = Net(args).cuda()\n",
    "if args.model_type == 'TAL':\n",
    "    print('debug')\n",
    "    model = Net(args).cuda()\n",
    "elif args.model_type == 'TAL-trans':\n",
    "    model = TransformerEncoder(args).cuda()\n",
    "elif args.model_type == 'TAL-new':\n",
    "    model = NewNet(args).cuda()\n",
    "elif args.model_type == 'Trans':\n",
    "    model = Transformer(args).cuda()\n",
    "elif args.model_type == 'resnet':\n",
    "    model = resnet50().cuda()\n",
    "elif args.model_type == 'MMT':\n",
    "    model = MMTEncoder(0, args).cuda()\n",
    "elif args.model_type == 'MMTLF':\n",
    "    model = LateFusion(args).cuda()\n",
    "else:\n",
    "    print ('model type not recognized')\n",
    "    exit(0)\n",
    "model.load_state_dict(torch.load(MODEL_FILE)['model'])\n",
    "#model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TransformerEncoder(\n",
       "  (conv1): ConvBlock(\n",
       "    (conv): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv2): ConvBlock(\n",
       "    (conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv3): ConvBlock(\n",
       "    (conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv4): ConvBlock(\n",
       "    (conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv5): ConvBlock(\n",
       "    (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv6): ConvBlock(\n",
       "    (conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv7): ConvBlock(\n",
       "    (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv8): ConvBlock(\n",
       "    (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv9): ConvBlock(\n",
       "    (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (conv10): ConvBlock(\n",
       "    (conv): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       "  (enc_layer): EncoderLayer(\n",
       "    (slf_attn): MultiHeadAttention(\n",
       "      (w_qs): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (w_ks): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (w_vs): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (attention): ScaledDotProductAttention(\n",
       "        (dropout): Dropout(p=0.5, inplace=False)\n",
       "        (softmax): Softmax(dim=2)\n",
       "      )\n",
       "      (layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "      (fc): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (dropout): Dropout(p=0.5, inplace=False)\n",
       "    )\n",
       "    (pos_ffn): PositionwiseFeedForward(\n",
       "      (w_1): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (w_2): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "      (dropout): Dropout(p=0.5, inplace=False)\n",
       "      (layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "    )\n",
       "  )\n",
       "  (fc_prob): Linear(in_features=1024, out_features=527, bias=True)\n",
       "  (fc_att): Linear(in_features=1024, out_features=527, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# data pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/ocean/projects/iri120008p/billyli/DayLongAudio/audioset_py3/AudioSetDoneRight\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "print(os.getcwd())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "import h5py\n",
    "hf = h5py.File('../../../data/googleAudioSet/data.h5', 'r')\n",
    "feat_as = hf['eval']['data']['feat_a']\n",
    "feat_vs = hf['eval']['data']['feat_v']\n",
    "labels = hf['eval']['data']['labels']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "ename": "TypeError",
     "evalue": "predict() got multiple values for argument 'verbose'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-24-d30f3f950580>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mgas_eval_global_prob\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mgas_eval_global_feat\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_as\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeat_vs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0;31m# feat_as = hf['valid']['data']['feat_a']\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;31m# feat_vs = hf['valid']['data']['feat_v']\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;31m# labels = hf['valid']['data']['labels']\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;31m# gas_valid_global_prob, gas_valid_global_feat = model.predict(feat_as, feat_vs, verbose=False)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTypeError\u001b[0m: predict() got multiple values for argument 'verbose'"
     ]
    }
   ],
   "source": [
    "gas_eval_global_prob, gas_eval_global_feat = model.predict(feat_as, feat_vs, verbose=False)\n",
    "# feat_as = hf['valid']['data']['feat_a']\n",
    "# feat_vs = hf['valid']['data']['feat_v']\n",
    "# labels = hf['valid']['data']['labels']\n",
    "# gas_valid_global_prob, gas_valid_global_feat = model.predict(feat_as, feat_vs, verbose=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "gas_eval_global_prob = model.predict(feat_as, verbose=False)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# gas_eval_global_prob = model.predict(feat_as[:] + perturbation_a,feat_vs[:] + perturbation_v, verbose=False)\n",
    "gas_eval_global_prob = model.predict(feat_as[:] + perturbation_a_only, verbose=False)\n",
    "\n",
    "# gas_eval_global_prob = model.predict(feat_as[:],feat_vs[:], verbose=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hf_feat = h5py.File('latefusion_bottleneck_feat.h5', 'w')\n",
    "# hf_feat.create_dataset('feat_eval', data=gas_eval_global_feat)\n",
    "# hf_feat.create_dataset('feat_valid', data=gas_valid_global_feat)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hf_feat.close()\n",
    "# del gas_eval_global_feat\n",
    "# del gas_valid_global_feat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(20098, 527)"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gas_eval_global_prob[0].shape#, gas_valid_global_prob.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "gas_eval_global_prob = gas_eval_global_prob[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Remember to normalize eval if we are to use normalized model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# %matplotlib inline\n",
    "# import pickle\n",
    "# import matplotlib.pyplot as plt\n",
    "# temp = pickle.load(open('adversarial_temp/delta_global_eps_norm_2_eps_0_1_no_delta_force.pkl', 'rb'))\n",
    "# # temp[0:200, :] =0\n",
    "# plt.imshow(temp.T,interpolation='nearest', aspect='auto')\n",
    "# plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'temp' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-34-f3ba4d7e47ec>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'temp' is not defined"
     ]
    }
   ],
   "source": [
    "np.clip(temp,0,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'perturbation' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-35-2062d3cef409>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtemp\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfeat_vs\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0mperturbation\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mgas_eval_global_prob\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_as\u001b[0m\u001b[0;34m,\u001b[0m  \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m23.27987289428711\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'perturbation' is not defined"
     ]
    }
   ],
   "source": [
    "temp = feat_vs+perturbation\n",
    "gas_eval_global_prob = model.predict(feat_as,  np.clip(temp, 0,23.27987289428711), verbose = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'perturbation' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-36-c0c840c26475>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtemp\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfeat_as\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0mperturbation\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mgas_eval_global_prob\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m-\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m  \u001b[0mfeat_vs\u001b[0m \u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'perturbation' is not defined"
     ]
    }
   ],
   "source": [
    "temp = feat_as+perturbation\n",
    "gas_eval_global_prob = model.predict(np.clip(temp, -2, 2),  feat_vs , verbose = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'perturbation_a' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-37-020d24dce02e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtemp1\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfeat_as\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mperturbation_a\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mtemp2\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfeat_vs\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mperturbation_v\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0mgas_eval_global_prob\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m-\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m23.27987289428711\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'perturbation_a' is not defined"
     ]
    }
   ],
   "source": [
    "temp1 = feat_as + perturbation_a\n",
    "temp2 = feat_vs + perturbation_v\n",
    "\n",
    "gas_eval_global_prob = model.predict(np.clip(temp1, -2, 2), np.clip(temp2, 0,23.27987289428711), verbose = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "ename": "TypeError",
     "evalue": "predict() got multiple values for argument 'verbose'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-38-05270688fa5d>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mgas_eval_global_prob\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_as\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mfeat_vs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mgas_eval_global_prob\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTypeError\u001b[0m: predict() got multiple values for argument 'verbose'"
     ]
    }
   ],
   "source": [
    "gas_eval_global_prob = model.predict(feat_as[:],feat_vs[:] , verbose=False)[0]\n",
    "print(gas_eval_global_prob.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "import csv\n",
    "class_name = []\n",
    "with open('../meta/class_labels_indices.csv', newline='') as csvfile:\n",
    "    spamreader = csv.reader(csvfile, delimiter=',', quotechar='|')\n",
    "    for row_id, row in enumerate(spamreader):\n",
    "        if row_id >= 1:\n",
    "            class_name.append(row[2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Performance on Google Audio Set:\n",
      "   CLASS ||    AP |   AUC |    d' | NAME\n",
      "---------++-------+-------+-------+---\n"
     ]
    }
   ],
   "source": [
    "print('Performance on Google Audio Set:')\n",
    "print(\"   CLASS ||    AP |   AUC |    d' | NAME\")\n",
    "FORMAT  = ' %00007s || %5.3f | %5.3f |%6.03f |%s '\n",
    "SEP     = ''.join('+' if c == '|' else '-' for c in FORMAT)\n",
    "print(SEP)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.18523182243314418 0.9033692283524556 1.839879083984829\n",
      " Average || 0.185 | 0.903 | 1.840 |Name \n",
      "---------++-------+-------+-------+---\n",
      "       0 || 0.675 | 0.895 | 1.773 |0 \n",
      "       1 || 0.062 | 0.805 | 1.216 |1 \n",
      "       2 || 0.099 | 0.920 | 1.982 |2 \n",
      "       3 || 0.245 | 0.946 | 2.273 |3 \n",
      "       4 || 0.117 | 0.937 | 2.165 |4 \n",
      "       5 || 0.031 | 0.920 | 1.988 |5 \n",
      "       6 || 0.287 | 0.991 | 3.320 |6 \n",
      "       7 || 0.324 | 0.977 | 2.818 |7 \n",
      "       8 || 0.049 | 0.902 | 1.832 |8 \n",
      "       9 || 0.055 | 0.886 | 1.708 |9 \n",
      "      10 || 0.047 | 0.924 | 2.024 |10 \n",
      "      11 || 0.041 | 0.916 | 1.954 |11 \n",
      "      12 || 0.674 | 0.996 | 3.744 |12 \n",
      "      13 || 0.236 | 0.967 | 2.594 |13 \n",
      "      14 || 0.114 | 0.960 | 2.476 |14 \n",
      "      15 || 0.540 | 0.960 | 2.471 |15 \n",
      "      16 || 0.229 | 0.951 | 2.338 |16 \n",
      "      17 || 0.337 | 0.979 | 2.876 |17 \n",
      "      18 || 0.121 | 0.961 | 2.486 |18 \n",
      "      19 || 0.109 | 0.951 | 2.341 |19 \n",
      "      20 || 0.240 | 0.977 | 2.819 |20 \n",
      "      21 || 0.135 | 0.946 | 2.278 |21 \n",
      "      22 || 0.257 | 0.962 | 2.517 |22 \n",
      "      23 || 0.248 | 0.973 | 2.735 |23 \n",
      "      24 || 0.118 | 0.955 | 2.395 |24 \n",
      "      25 || 0.043 | 0.809 | 1.237 |25 \n",
      "      26 || 0.116 | 0.929 | 2.072 |26 \n",
      "      27 || 0.165 | 0.927 | 2.055 |27 \n",
      "      28 || 0.529 | 0.986 | 3.106 |28 \n",
      "      29 || 0.112 | 0.963 | 2.532 |29 \n",
      "      30 || 0.322 | 0.970 | 2.670 |30 \n",
      "      31 || 0.296 | 0.982 | 2.972 |31 \n",
      "      32 || 0.062 | 0.960 | 2.483 |32 \n",
      "      33 || 0.136 | 0.965 | 2.571 |33 \n",
      "      34 || 0.345 | 0.965 | 2.567 |34 \n",
      "      35 || 0.220 | 0.943 | 2.230 |35 \n",
      "      36 || 0.361 | 0.975 | 2.776 |36 \n",
      "      37 || 0.157 | 0.919 | 1.977 |37 \n",
      "      38 || 0.065 | 0.849 | 1.461 |38 \n",
      "      39 || 0.151 | 0.958 | 2.440 |39 \n",
      "      40 || 0.645 | 0.991 | 3.370 |40 \n",
      "      41 || 0.069 | 0.860 | 1.530 |41 \n",
      "      42 || 0.038 | 0.850 | 1.465 |42 \n",
      "      43 || 0.339 | 0.964 | 2.544 |43 \n",
      "      44 || 0.062 | 0.844 | 1.429 |44 \n",
      "      45 || 0.043 | 0.794 | 1.160 |45 \n",
      "      46 || 0.060 | 0.891 | 1.739 |46 \n",
      "      47 || 0.355 | 0.965 | 2.554 |47 \n",
      "      48 || 0.149 | 0.956 | 2.408 |48 \n",
      "      49 || 0.225 | 0.940 | 2.205 |49 \n",
      "      50 || 0.020 | 0.826 | 1.329 |50 \n",
      "      51 || 0.024 | 0.842 | 1.419 |51 \n",
      "      52 || 0.020 | 0.788 | 1.129 |52 \n",
      "      53 || 0.103 | 0.889 | 1.731 |53 \n",
      "      54 || 0.316 | 0.930 | 2.089 |54 \n",
      "      55 || 0.087 | 0.824 | 1.318 |55 \n",
      "      56 || 0.270 | 0.824 | 1.316 |56 \n",
      "      57 || 0.124 | 0.805 | 1.217 |57 \n",
      "      58 || 0.511 | 0.978 | 2.835 |58 \n",
      "      59 || 0.413 | 0.965 | 2.555 |59 \n",
      "      60 || 0.461 | 0.943 | 2.239 |60 \n",
      "      61 || 0.269 | 0.917 | 1.957 |61 \n",
      "      62 || 0.105 | 0.854 | 1.490 |62 \n",
      "      63 || 0.237 | 0.926 | 2.049 |63 \n",
      "      64 || 0.535 | 0.991 | 3.333 |64 \n",
      "      65 || 0.324 | 0.972 | 2.710 |65 \n",
      "      66 || 0.291 | 0.984 | 3.021 |66 \n",
      "      67 || 0.172 | 0.943 | 2.241 |67 \n",
      "      68 || 0.427 | 0.990 | 3.269 |68 \n",
      "      69 || 0.414 | 0.983 | 3.012 |69 \n",
      "      70 || 0.123 | 0.894 | 1.768 |70 \n",
      "      71 || 0.126 | 0.963 | 2.521 |71 \n",
      "      72 || 0.294 | 0.865 | 1.559 |72 \n",
      "      73 || 0.278 | 0.893 | 1.760 |73 \n",
      "      74 || 0.352 | 0.918 | 1.966 |74 \n",
      "      75 || 0.286 | 0.992 | 3.396 |75 \n",
      "      76 || 0.100 | 0.932 | 2.111 |76 \n",
      "      77 || 0.642 | 0.973 | 2.734 |77 \n",
      "      78 || 0.276 | 0.981 | 2.931 |78 \n",
      "      79 || 0.163 | 0.965 | 2.562 |79 \n",
      "      80 || 0.066 | 0.946 | 2.277 |80 \n",
      "      81 || 0.300 | 0.906 | 1.858 |81 \n",
      "      82 || 0.214 | 0.779 | 1.086 |82 \n",
      "      83 || 0.293 | 0.959 | 2.463 |83 \n",
      "      84 || 0.036 | 0.817 | 1.278 |84 \n",
      "      85 || 0.268 | 0.963 | 2.529 |85 \n",
      "      86 || 0.169 | 0.855 | 1.495 |86 \n",
      "      87 || 0.180 | 0.915 | 1.937 |87 \n",
      "      88 || 0.087 | 0.913 | 1.918 |88 \n",
      "      89 || 0.357 | 0.966 | 2.584 |89 \n",
      "      90 || 0.338 | 0.852 | 1.477 |90 \n",
      "      91 || 0.326 | 0.874 | 1.623 |91 \n",
      "      92 || 0.162 | 0.884 | 1.689 |92 \n",
      "      93 || 0.145 | 0.942 | 2.223 |93 \n",
      "      94 || 0.063 | 0.870 | 1.591 |94 \n",
      "      95 || 0.283 | 0.932 | 2.105 |95 \n",
      "      96 || 0.302 | 0.932 | 2.110 |96 \n",
      "      97 || 0.250 | 0.921 | 1.999 |97 \n",
      "      98 || 0.488 | 0.946 | 2.277 |98 \n",
      "      99 || 0.556 | 0.971 | 2.685 |99 \n",
      "     100 || 0.424 | 0.962 | 2.511 |100 \n",
      "     101 || 0.776 | 0.992 | 3.423 |101 \n",
      "     102 || 0.396 | 0.981 | 2.942 |102 \n",
      "     103 || 0.405 | 0.978 | 2.840 |103 \n",
      "     104 || 0.086 | 0.875 | 1.629 |104 \n",
      "     105 || 0.175 | 0.905 | 1.852 |105 \n",
      "     106 || 0.293 | 0.944 | 2.247 |106 \n",
      "     107 || 0.318 | 0.949 | 2.312 |107 \n",
      "     108 || 0.127 | 0.887 | 1.711 |108 \n",
      "     109 || 0.297 | 0.940 | 2.200 |109 \n",
      "     110 || 0.367 | 0.951 | 2.342 |110 \n",
      "     111 || 0.216 | 0.905 | 1.857 |111 \n",
      "     112 || 0.198 | 0.908 | 1.880 |112 \n",
      "     113 || 0.112 | 0.928 | 2.071 |113 \n",
      "     114 || 0.101 | 0.893 | 1.758 |114 \n",
      "     115 || 0.404 | 0.948 | 2.300 |115 \n",
      "     116 || 0.426 | 0.931 | 2.100 |116 \n",
      "     117 || 0.222 | 0.919 | 1.982 |117 \n",
      "     118 || 0.339 | 0.913 | 1.920 |118 \n",
      "     119 || 0.496 | 0.947 | 2.291 |119 \n",
      "     120 || 0.546 | 0.929 | 2.072 |120 \n",
      "     121 || 0.128 | 0.880 | 1.664 |121 \n",
      "     122 || 0.241 | 0.967 | 2.595 |122 \n",
      "     123 || 0.025 | 0.857 | 1.512 |123 \n",
      "     124 || 0.012 | 0.766 | 1.025 |124 \n",
      "     125 || 0.021 | 0.873 | 1.611 |125 \n",
      "     126 || 0.313 | 0.899 | 1.801 |126 \n",
      "     127 || 0.265 | 0.938 | 2.175 |127 \n",
      "     128 || 0.335 | 0.924 | 2.023 |128 \n",
      "     129 || 0.352 | 0.917 | 1.964 |129 \n",
      "     130 || 0.007 | 0.672 | 0.629 |130 \n",
      "     131 || 0.170 | 0.914 | 1.930 |131 \n",
      "     132 || 0.550 | 0.956 | 2.416 |132 \n",
      "     133 || 0.634 | 0.970 | 2.662 |133 \n",
      "     134 || 0.040 | 0.895 | 1.772 |134 \n",
      "     135 || 0.011 | 0.754 | 0.972 |135 \n",
      "     136 || 0.238 | 0.899 | 1.801 |136 \n",
      "     137 || 0.714 | 0.889 | 1.727 |137 \n",
      "     138 || 0.230 | 0.914 | 1.930 |138 \n",
      "     139 || 0.347 | 0.974 | 2.758 |139 \n",
      "     140 || 0.465 | 0.974 | 2.747 |140 \n",
      "     141 || 0.399 | 0.982 | 2.958 |141 \n",
      "     142 || 0.369 | 0.974 | 2.754 |142 \n",
      "     143 || 0.273 | 0.987 | 3.130 |143 \n",
      "     144 || 0.231 | 0.974 | 2.759 |144 \n",
      "     145 || 0.223 | 0.985 | 3.072 |145 \n",
      "     146 || 0.186 | 0.976 | 2.793 |146 \n",
      "     147 || 0.534 | 0.991 | 3.339 |147 \n",
      "     148 || 0.396 | 0.980 | 2.893 |148 \n",
      "     149 || 0.299 | 0.971 | 2.686 |149 \n",
      "     150 || 0.251 | 0.959 | 2.455 |150 \n",
      "     151 || 0.391 | 0.973 | 2.727 |151 \n",
      "     152 || 0.167 | 0.949 | 2.317 |152 \n",
      "     153 || 0.176 | 0.967 | 2.591 |153 \n",
      "     154 || 0.179 | 0.957 | 2.423 |154 \n",
      "     155 || 0.341 | 0.958 | 2.450 |155 \n",
      "     156 || 0.116 | 0.950 | 2.332 |156 \n",
      "     157 || 0.277 | 0.974 | 2.738 |157 \n",
      "     158 || 0.033 | 0.907 | 1.868 |158 \n",
      "     159 || 0.063 | 0.932 | 2.107 |159 \n",
      "     160 || 0.220 | 0.953 | 2.367 |160 \n",
      "     161 || 0.378 | 0.951 | 2.335 |161 \n",
      "     162 || 0.492 | 0.975 | 2.771 |162 \n",
      "     163 || 0.162 | 0.972 | 2.694 |163 \n",
      "     164 || 0.380 | 0.955 | 2.398 |164 \n",
      "     165 || 0.492 | 0.981 | 2.944 |165 \n",
      "     166 || 0.447 | 0.983 | 2.997 |166 \n",
      "     167 || 0.385 | 0.974 | 2.758 |167 \n",
      "     168 || 0.474 | 0.974 | 2.749 |168 \n",
      "     169 || 0.488 | 0.982 | 2.954 |169 \n",
      "     170 || 0.403 | 0.978 | 2.852 |170 \n",
      "     171 || 0.415 | 0.982 | 2.959 |171 \n",
      "     172 || 0.464 | 0.989 | 3.253 |172 \n",
      "     173 || 0.259 | 0.966 | 2.582 |173 \n",
      "     174 || 0.247 | 0.966 | 2.586 |174 \n",
      "     175 || 0.093 | 0.916 | 1.946 |175 \n",
      "     176 || 0.166 | 0.935 | 2.145 |176 \n",
      "     177 || 0.306 | 0.974 | 2.739 |177 \n",
      "     178 || 0.072 | 0.910 | 1.900 |178 \n",
      "     179 || 0.249 | 0.987 | 3.157 |179 \n",
      "     180 || 0.319 | 0.986 | 3.090 |180 \n",
      "     181 || 0.295 | 0.988 | 3.173 |181 \n",
      "     182 || 0.345 | 0.987 | 3.160 |182 \n",
      "     183 || 0.372 | 0.957 | 2.422 |183 \n",
      "     184 || 0.215 | 0.964 | 2.553 |184 \n",
      "     185 || 0.495 | 0.973 | 2.721 |185 \n",
      "     186 || 0.546 | 0.992 | 3.386 |186 \n",
      "     187 || 0.346 | 0.980 | 2.907 |187 \n",
      "     188 || 0.371 | 0.986 | 3.101 |188 \n",
      "     189 || 0.193 | 0.947 | 2.292 |189 \n",
      "     190 || 0.359 | 0.986 | 3.114 |190 \n",
      "     191 || 0.196 | 0.959 | 2.455 |191 \n",
      "     192 || 0.205 | 0.960 | 2.482 |192 \n",
      "     193 || 0.263 | 0.956 | 2.410 |193 \n",
      "     194 || 0.213 | 0.962 | 2.515 |194 \n",
      "     195 || 0.457 | 0.961 | 2.499 |195 \n",
      "     196 || 0.327 | 0.975 | 2.776 |196 \n",
      "     197 || 0.412 | 0.985 | 3.084 |197 \n",
      "     198 || 0.296 | 0.971 | 2.684 |198 \n",
      "     199 || 0.400 | 0.978 | 2.850 |199 \n",
      "     200 || 0.127 | 0.953 | 2.362 |200 \n",
      "     201 || 0.434 | 0.975 | 2.771 |201 \n",
      "     202 || 0.467 | 0.967 | 2.596 |202 \n",
      "     203 || 0.507 | 0.948 | 2.296 |203 \n",
      "     204 || 0.288 | 0.893 | 1.757 |204 \n",
      "     205 || 0.384 | 0.958 | 2.448 |205 \n",
      "     206 || 0.361 | 0.921 | 2.000 |206 \n",
      "     207 || 0.503 | 0.993 | 3.483 |207 \n",
      "     208 || 0.503 | 0.988 | 3.180 |208 \n",
      "     209 || 0.624 | 0.985 | 3.066 |209 \n",
      "     210 || 0.803 | 0.990 | 3.269 |210 \n",
      "     211 || 0.586 | 0.987 | 3.146 |211 \n",
      "     212 || 0.459 | 0.955 | 2.400 |212 \n",
      "     213 || 0.220 | 0.936 | 2.155 |213 \n",
      "     214 || 0.395 | 0.991 | 3.324 |214 \n",
      "     215 || 0.121 | 0.928 | 2.067 |215 \n",
      "     216 || 0.041 | 0.910 | 1.895 |216 \n",
      "     217 || 0.152 | 0.973 | 2.733 |217 \n",
      "     218 || 0.391 | 0.950 | 2.321 |218 \n",
      "     219 || 0.054 | 0.924 | 2.025 |219 \n",
      "     220 || 0.055 | 0.942 | 2.222 |220 \n",
      "     221 || 0.061 | 0.941 | 2.207 |221 \n",
      "     222 || 0.124 | 0.965 | 2.554 |222 \n",
      "     223 || 0.079 | 0.950 | 2.322 |223 \n",
      "     224 || 0.065 | 0.924 | 2.025 |224 \n",
      "     225 || 0.047 | 0.918 | 1.971 |225 \n",
      "     226 || 0.070 | 0.955 | 2.404 |226 \n",
      "     227 || 0.083 | 0.933 | 2.123 |227 \n",
      "     228 || 0.165 | 0.966 | 2.586 |228 \n",
      "     229 || 0.090 | 0.959 | 2.465 |229 \n",
      "     230 || 0.061 | 0.939 | 2.190 |230 \n",
      "     231 || 0.312 | 0.979 | 2.884 |231 \n",
      "     232 || 0.054 | 0.915 | 1.941 |232 \n",
      "     233 || 0.109 | 0.939 | 2.188 |233 \n",
      "     234 || 0.110 | 0.947 | 2.283 |234 \n",
      "     235 || 0.041 | 0.926 | 2.047 |235 \n",
      "     236 || 0.051 | 0.905 | 1.850 |236 \n",
      "     237 || 0.071 | 0.927 | 2.054 |237 \n",
      "     238 || 0.327 | 0.943 | 2.233 |238 \n",
      "     239 || 0.037 | 0.906 | 1.858 |239 \n",
      "     240 || 0.112 | 0.967 | 2.605 |240 \n",
      "     241 || 0.224 | 0.976 | 2.802 |241 \n",
      "     242 || 0.249 | 0.971 | 2.681 |242 \n",
      "     243 || 0.185 | 0.964 | 2.538 |243 \n",
      "     244 || 0.111 | 0.962 | 2.513 |244 \n",
      "     245 || 0.141 | 0.950 | 2.329 |245 \n",
      "     246 || 0.089 | 0.944 | 2.248 |246 \n",
      "     247 || 0.043 | 0.935 | 2.146 |247 \n",
      "     248 || 0.176 | 0.965 | 2.565 |248 \n",
      "     249 || 0.421 | 0.978 | 2.849 |249 \n",
      "     250 || 0.118 | 0.950 | 2.322 |250 \n",
      "     251 || 0.066 | 0.943 | 2.235 |251 \n",
      "     252 || 0.145 | 0.954 | 2.384 |252 \n",
      "     253 || 0.045 | 0.929 | 2.079 |253 \n",
      "     254 || 0.070 | 0.947 | 2.280 |254 \n",
      "     255 || 0.353 | 0.981 | 2.942 |255 \n",
      "     256 || 0.229 | 0.980 | 2.908 |256 \n",
      "     257 || 0.256 | 0.981 | 2.933 |257 \n",
      "     258 || 0.075 | 0.948 | 2.304 |258 \n",
      "     259 || 0.200 | 0.967 | 2.595 |259 \n",
      "     260 || 0.034 | 0.873 | 1.614 |260 \n",
      "     261 || 0.401 | 0.982 | 2.971 |261 \n",
      "     262 || 0.164 | 0.942 | 2.228 |262 \n",
      "     263 || 0.112 | 0.943 | 2.235 |263 \n",
      "     264 || 0.043 | 0.928 | 2.062 |264 \n",
      "     265 || 0.097 | 0.962 | 2.508 |265 \n",
      "     266 || 0.071 | 0.932 | 2.112 |266 \n",
      "     267 || 0.052 | 0.933 | 2.116 |267 \n",
      "     268 || 0.054 | 0.905 | 1.852 |268 \n",
      "     269 || 0.052 | 0.950 | 2.330 |269 \n",
      "     270 || 0.073 | 0.918 | 1.964 |270 \n",
      "     271 || 0.313 | 0.977 | 2.816 |271 \n",
      "     272 || 0.114 | 0.952 | 2.348 |272 \n",
      "     273 || 0.040 | 0.937 | 2.162 |273 \n",
      "     274 || 0.146 | 0.968 | 2.618 |274 \n",
      "     275 || 0.041 | 0.936 | 2.149 |275 \n",
      "     276 || 0.047 | 0.947 | 2.280 |276 \n",
      "     277 || 0.062 | 0.911 | 1.909 |277 \n",
      "     278 || 0.043 | 0.928 | 2.071 |278 \n",
      "     279 || 0.028 | 0.921 | 1.998 |279 \n",
      "     280 || 0.030 | 0.929 | 2.080 |280 \n",
      "     281 || 0.097 | 0.976 | 2.793 |281 \n",
      "     282 || 0.083 | 0.901 | 1.820 |282 \n",
      "     283 || 0.146 | 0.900 | 1.811 |283 \n",
      "     284 || 0.050 | 0.849 | 1.460 |284 \n",
      "     285 || 0.159 | 0.927 | 2.057 |285 \n",
      "     286 || 0.061 | 0.908 | 1.875 |286 \n",
      "     287 || 0.065 | 0.914 | 1.932 |287 \n",
      "     288 || 0.067 | 0.755 | 0.977 |288 \n",
      "     289 || 0.025 | 0.786 | 1.120 |289 \n",
      "     290 || 0.022 | 0.765 | 1.022 |290 \n",
      "     291 || 0.023 | 0.835 | 1.378 |291 \n",
      "     292 || 0.009 | 0.783 | 1.108 |292 \n",
      "     293 || 0.015 | 0.894 | 1.763 |293 \n",
      "     294 || 0.044 | 0.886 | 1.707 |294 \n",
      "     295 || 0.046 | 0.903 | 1.835 |295 \n",
      "     296 || 0.016 | 0.813 | 1.259 |296 \n",
      "     297 || 0.011 | 0.793 | 1.153 |297 \n",
      "     298 || 0.007 | 0.745 | 0.931 |298 \n",
      "     299 || 0.027 | 0.859 | 1.524 |299 \n",
      "     300 || 0.183 | 0.826 | 1.327 |300 \n",
      "     301 || 0.127 | 0.872 | 1.606 |301 \n",
      "     302 || 0.087 | 0.942 | 2.224 |302 \n",
      "     303 || 0.067 | 0.891 | 1.743 |303 \n",
      "     304 || 0.030 | 0.859 | 1.524 |304 \n",
      "     305 || 0.047 | 0.882 | 1.677 |305 \n",
      "     306 || 0.027 | 0.855 | 1.499 |306 \n",
      "     307 || 0.140 | 0.863 | 1.546 |307 \n",
      "     308 || 0.186 | 0.960 | 2.483 |308 \n",
      "     309 || 0.211 | 0.956 | 2.417 |309 \n",
      "     310 || 0.420 | 0.972 | 2.694 |310 \n",
      "     311 || 0.008 | 0.729 | 0.864 |311 \n",
      "     312 || 0.132 | 0.848 | 1.455 |312 \n",
      "     313 || 0.081 | 0.756 | 0.981 |313 \n",
      "     314 || 0.039 | 0.875 | 1.629 |314 \n",
      "     315 || 0.034 | 0.816 | 1.271 |315 \n",
      "     316 || 0.066 | 0.864 | 1.555 |316 \n",
      "     317 || 0.023 | 0.859 | 1.524 |317 \n",
      "     318 || 0.382 | 0.983 | 3.001 |318 \n",
      "     319 || 0.214 | 0.918 | 1.965 |319 \n",
      "     320 || 0.536 | 0.967 | 2.608 |320 \n",
      "     321 || 0.015 | 0.833 | 1.366 |321 \n",
      "     322 || 0.745 | 0.982 | 2.962 |322 \n",
      "     323 || 0.594 | 0.988 | 3.174 |323 \n",
      "     324 || 0.621 | 0.973 | 2.721 |324 \n",
      "     325 || 0.447 | 0.978 | 2.848 |325 \n",
      "     326 || 0.042 | 0.855 | 1.497 |326 \n",
      "     327 || 0.010 | 0.770 | 1.043 |327 \n",
      "     328 || 0.044 | 0.722 | 0.831 |328 \n",
      "     329 || 0.053 | 0.727 | 0.854 |329 \n",
      "     330 || 0.214 | 0.918 | 1.965 |330 \n",
      "     331 || 0.189 | 0.954 | 2.386 |331 \n",
      "     332 || 0.047 | 0.741 | 0.913 |332 \n",
      "     333 || 0.051 | 0.800 | 1.188 |333 \n",
      "     334 || 0.015 | 0.734 | 0.883 |334 \n",
      "     335 || 0.046 | 0.888 | 1.720 |335 \n",
      "     336 || 0.059 | 0.872 | 1.606 |336 \n",
      "     337 || 0.147 | 0.925 | 2.032 |337 \n",
      "     338 || 0.017 | 0.864 | 1.552 |338 \n",
      "     339 || 0.024 | 0.900 | 1.815 |339 \n",
      "     340 || 0.029 | 0.843 | 1.421 |340 \n",
      "     341 || 0.029 | 0.755 | 0.977 |341 \n",
      "     342 || 0.023 | 0.814 | 1.263 |342 \n",
      "     343 || 0.060 | 0.873 | 1.610 |343 \n",
      "     344 || 0.011 | 0.787 | 1.127 |344 \n",
      "     345 || 0.009 | 0.664 | 0.598 |345 \n",
      "     346 || 0.043 | 0.934 | 2.131 |346 \n",
      "     347 || 0.192 | 0.910 | 1.898 |347 \n",
      "     348 || 0.044 | 0.926 | 2.041 |348 \n",
      "     349 || 0.022 | 0.851 | 1.474 |349 \n",
      "     350 || 0.029 | 0.851 | 1.472 |350 \n",
      "     351 || 0.021 | 0.889 | 1.730 |351 \n",
      "     352 || 0.045 | 0.908 | 1.881 |352 \n",
      "     353 || 0.147 | 0.910 | 1.900 |353 \n",
      "     354 || 0.018 | 0.849 | 1.461 |354 \n",
      "     355 || 0.302 | 0.912 | 1.910 |355 \n",
      "     356 || 0.169 | 0.887 | 1.712 |356 \n",
      "     357 || 0.016 | 0.825 | 1.324 |357 \n",
      "     358 || 0.021 | 0.818 | 1.281 |358 \n",
      "     359 || 0.218 | 0.905 | 1.851 |359 \n",
      "     360 || 0.078 | 0.867 | 1.572 |360 \n",
      "     361 || 0.196 | 0.879 | 1.655 |361 \n",
      "     362 || 0.029 | 0.824 | 1.316 |362 \n",
      "     363 || 0.040 | 0.881 | 1.670 |363 \n",
      "     364 || 0.047 | 0.911 | 1.906 |364 \n",
      "     365 || 0.044 | 0.924 | 2.021 |365 \n",
      "     366 || 0.287 | 0.845 | 1.436 |366 \n",
      "     367 || 0.046 | 0.841 | 1.414 |367 \n",
      "     368 || 0.056 | 0.816 | 1.273 |368 \n",
      "     369 || 0.073 | 0.879 | 1.658 |369 \n",
      "     370 || 0.019 | 0.809 | 1.235 |370 \n",
      "     371 || 0.026 | 0.865 | 1.560 |371 \n",
      "     372 || 0.112 | 0.769 | 1.042 |372 \n",
      "     373 || 0.029 | 0.796 | 1.169 |373 \n",
      "     374 || 0.320 | 0.952 | 2.360 |374 \n",
      "     375 || 0.026 | 0.657 | 0.572 |375 \n",
      "     376 || 0.041 | 0.846 | 1.442 |376 \n",
      "     377 || 0.024 | 0.816 | 1.275 |377 \n",
      "     378 || 0.049 | 0.813 | 1.255 |378 \n",
      "     379 || 0.037 | 0.833 | 1.368 |379 \n",
      "     380 || 0.043 | 0.878 | 1.645 |380 \n",
      "     381 || 0.106 | 0.871 | 1.598 |381 \n",
      "     382 || 0.222 | 0.938 | 2.175 |382 \n",
      "     383 || 0.129 | 0.945 | 2.254 |383 \n",
      "     384 || 0.298 | 0.909 | 1.883 |384 \n",
      "     385 || 0.079 | 0.906 | 1.861 |385 \n",
      "     386 || 0.385 | 0.916 | 1.951 |386 \n",
      "     387 || 0.078 | 0.847 | 1.447 |387 \n",
      "     388 || 0.126 | 0.933 | 2.119 |388 \n",
      "     389 || 0.128 | 0.958 | 2.449 |389 \n",
      "     390 || 0.466 | 0.984 | 3.045 |390 \n",
      "     391 || 0.212 | 0.946 | 2.279 |391 \n",
      "     392 || 0.242 | 0.955 | 2.397 |392 \n",
      "     393 || 0.577 | 0.989 | 3.219 |393 \n",
      "     394 || 0.478 | 0.976 | 2.799 |394 \n",
      "     395 || 0.328 | 0.929 | 2.077 |395 \n",
      "     396 || 0.725 | 0.979 | 2.879 |396 \n",
      "     397 || 0.733 | 0.995 | 3.684 |397 \n",
      "     398 || 0.070 | 0.916 | 1.949 |398 \n",
      "     399 || 0.639 | 0.992 | 3.415 |399 \n",
      "     400 || 0.562 | 0.990 | 3.297 |400 \n",
      "     401 || 0.025 | 0.822 | 1.306 |401 \n",
      "     402 || 0.260 | 0.958 | 2.448 |402 \n",
      "     403 || 0.310 | 0.975 | 2.762 |403 \n",
      "     404 || 0.038 | 0.785 | 1.115 |404 \n",
      "     405 || 0.020 | 0.806 | 1.221 |405 \n",
      "     406 || 0.124 | 0.887 | 1.715 |406 \n",
      "     407 || 0.064 | 0.794 | 1.158 |407 \n",
      "     408 || 0.116 | 0.832 | 1.359 |408 \n",
      "     409 || 0.032 | 0.830 | 1.351 |409 \n",
      "     410 || 0.004 | 0.516 | 0.057 |410 \n",
      "     411 || 0.123 | 0.925 | 2.040 |411 \n",
      "     412 || 0.062 | 0.858 | 1.516 |412 \n",
      "     413 || 0.028 | 0.827 | 1.332 |413 \n",
      "     414 || 0.175 | 0.870 | 1.591 |414 \n",
      "     415 || 0.022 | 0.807 | 1.227 |415 \n",
      "     416 || 0.049 | 0.886 | 1.708 |416 \n",
      "     417 || 0.067 | 0.816 | 1.271 |417 \n",
      "     418 || 0.058 | 0.860 | 1.528 |418 \n",
      "     419 || 0.062 | 0.937 | 2.164 |419 \n",
      "     420 || 0.151 | 0.954 | 2.389 |420 \n",
      "     421 || 0.080 | 0.887 | 1.716 |421 \n",
      "     422 || 0.061 | 0.920 | 1.989 |422 \n",
      "     423 || 0.134 | 0.872 | 1.609 |423 \n",
      "     424 || 0.060 | 0.899 | 1.805 |424 \n",
      "     425 || 0.060 | 0.902 | 1.827 |425 \n",
      "     426 || 0.060 | 0.920 | 1.990 |426 \n",
      "     427 || 0.127 | 0.908 | 1.876 |427 \n",
      "     428 || 0.114 | 0.958 | 2.442 |428 \n",
      "     429 || 0.037 | 0.915 | 1.936 |429 \n",
      "     430 || 0.044 | 0.911 | 1.907 |430 \n",
      "     431 || 0.332 | 0.981 | 2.922 |431 \n",
      "     432 || 0.141 | 0.942 | 2.224 |432 \n",
      "     433 || 0.158 | 0.965 | 2.557 |433 \n",
      "     434 || 0.010 | 0.670 | 0.622 |434 \n",
      "     435 || 0.048 | 0.933 | 2.124 |435 \n",
      "     436 || 0.033 | 0.934 | 2.131 |436 \n",
      "     437 || 0.040 | 0.758 | 0.991 |437 \n",
      "     438 || 0.049 | 0.903 | 1.837 |438 \n",
      "     439 || 0.014 | 0.595 | 0.338 |439 \n",
      "     440 || 0.026 | 0.800 | 1.189 |440 \n",
      "     441 || 0.066 | 0.883 | 1.685 |441 \n",
      "     442 || 0.070 | 0.878 | 1.650 |442 \n",
      "     443 || 0.051 | 0.886 | 1.702 |443 \n",
      "     444 || 0.053 | 0.771 | 1.048 |444 \n",
      "     445 || 0.035 | 0.908 | 1.875 |445 \n",
      "     446 || 0.027 | 0.818 | 1.285 |446 \n",
      "     447 || 0.029 | 0.741 | 0.915 |447 \n",
      "     448 || 0.049 | 0.889 | 1.726 |448 \n",
      "     449 || 0.294 | 0.945 | 2.257 |449 \n",
      "     450 || 0.033 | 0.825 | 1.320 |450 \n",
      "     451 || 0.007 | 0.624 | 0.449 |451 \n",
      "     452 || 0.025 | 0.785 | 1.115 |452 \n",
      "     453 || 0.023 | 0.749 | 0.949 |453 \n",
      "     454 || 0.009 | 0.740 | 0.908 |454 \n",
      "     455 || 0.046 | 0.902 | 1.826 |455 \n",
      "     456 || 0.008 | 0.734 | 0.885 |456 \n",
      "     457 || 0.289 | 0.945 | 2.257 |457 \n",
      "     458 || 0.128 | 0.898 | 1.796 |458 \n",
      "     459 || 0.178 | 0.881 | 1.670 |459 \n",
      "     460 || 0.083 | 0.858 | 1.515 |460 \n",
      "     461 || 0.095 | 0.901 | 1.824 |461 \n",
      "     462 || 0.345 | 0.962 | 2.510 |462 \n",
      "     463 || 0.384 | 0.955 | 2.404 |463 \n",
      "     464 || 0.268 | 0.953 | 2.371 |464 \n",
      "     465 || 0.185 | 0.941 | 2.215 |465 \n",
      "     466 || 0.006 | 0.703 | 0.754 |466 \n",
      "     467 || 0.201 | 0.940 | 2.198 |467 \n",
      "     468 || 0.021 | 0.834 | 1.375 |468 \n",
      "     469 || 0.009 | 0.756 | 0.980 |469 \n",
      "     470 || 0.084 | 0.915 | 1.938 |470 \n",
      "     471 || 0.015 | 0.754 | 0.971 |471 \n",
      "     472 || 0.178 | 0.920 | 1.991 |472 \n",
      "     473 || 0.017 | 0.753 | 0.969 |473 \n",
      "     474 || 0.179 | 0.890 | 1.734 |474 \n",
      "     475 || 0.008 | 0.739 | 0.907 |475 \n",
      "     476 || 0.118 | 0.881 | 1.671 |476 \n",
      "     477 || 0.038 | 0.924 | 2.029 |477 \n",
      "     478 || 0.028 | 0.761 | 1.004 |478 \n",
      "     479 || 0.180 | 0.898 | 1.797 |479 \n",
      "     480 || 0.071 | 0.861 | 1.536 |480 \n",
      "     481 || 0.292 | 0.964 | 2.547 |481 \n",
      "     482 || 0.080 | 0.840 | 1.408 |482 \n",
      "     483 || 0.081 | 0.916 | 1.953 |483 \n",
      "     484 || 0.079 | 0.905 | 1.852 |484 \n",
      "     485 || 0.033 | 0.804 | 1.209 |485 \n",
      "     486 || 0.004 | 0.567 | 0.240 |486 \n",
      "     487 || 0.012 | 0.824 | 1.317 |487 \n",
      "     488 || 0.017 | 0.837 | 1.391 |488 \n",
      "     489 || 0.053 | 0.883 | 1.680 |489 \n",
      "     490 || 0.037 | 0.780 | 1.090 |490 \n",
      "     491 || 0.306 | 0.939 | 2.183 |491 \n",
      "     492 || 0.013 | 0.710 | 0.782 |492 \n",
      "     493 || 0.168 | 0.894 | 1.762 |493 \n",
      "     494 || 0.735 | 0.995 | 3.606 |494 \n",
      "     495 || 0.101 | 0.964 | 2.544 |495 \n",
      "     496 || 0.057 | 0.858 | 1.512 |496 \n",
      "     497 || 0.061 | 0.874 | 1.623 |497 \n",
      "     498 || 0.629 | 0.984 | 3.043 |498 \n",
      "     499 || 0.225 | 0.923 | 2.014 |499 \n",
      "     500 || 0.125 | 0.915 | 1.939 |500 \n",
      "     501 || 0.425 | 0.975 | 2.781 |501 \n",
      "     502 || 0.015 | 0.877 | 1.643 |502 \n",
      "     503 || 0.123 | 0.905 | 1.851 |503 \n",
      "     504 || 0.022 | 0.805 | 1.214 |504 \n",
      "     505 || 0.056 | 0.731 | 0.871 |505 \n",
      "     506 || 0.077 | 0.746 | 0.935 |506 \n",
      "     507 || 0.020 | 0.710 | 0.783 |507 \n",
      "     508 || 0.027 | 0.876 | 1.634 |508 \n",
      "     509 || 0.032 | 0.697 | 0.728 |509 \n",
      "     510 || 0.023 | 0.745 | 0.931 |510 \n",
      "     511 || 0.217 | 0.891 | 1.745 |511 \n",
      "     512 || 0.037 | 0.831 | 1.352 |512 \n",
      "     513 || 0.012 | 0.787 | 1.126 |513 \n",
      "     514 || 0.183 | 0.903 | 1.836 |514 \n",
      "     515 || 0.104 | 0.907 | 1.871 |515 \n",
      "     516 || 0.026 | 0.770 | 1.043 |516 \n",
      "     517 || 0.371 | 0.970 | 2.668 |517 \n",
      "     518 || 0.065 | 0.899 | 1.801 |518 \n",
      "     519 || 0.011 | 0.796 | 1.172 |519 \n",
      "     520 || 0.015 | 0.826 | 1.327 |520 \n",
      "     521 || 0.021 | 0.903 | 1.841 |521 \n",
      "     522 || 0.168 | 0.963 | 2.531 |522 \n",
      "     523 || 0.039 | 0.908 | 1.878 |523 \n",
      "     524 || 0.049 | 0.841 | 1.414 |524 \n",
      "     525 || 0.140 | 0.752 | 0.961 |525 \n",
      "     526 || 0.028 | 0.773 | 1.059 |526 \n"
     ]
    }
   ],
   "source": [
    "\n",
    "# 0.427 | 0.971 | 2.686\n",
    "# 0.37905724410180536 0.9626179342278633 2.519999363956429\n",
    "gas_eval_y =labels\n",
    "classwise = []\n",
    "classwise_dict = []\n",
    "N_CLASSES = gas_eval_global_prob.shape[-1]\n",
    "for i in range(N_CLASSES):\n",
    "    a, b, c = gas_eval(gas_eval_global_prob[:,i], gas_eval_y[:,i])     # AP, AUC, dprime\n",
    "    classwise_dict.append({'mAP':a, 'AUC':b, 'dprime':c, 'id':i})\n",
    "    classwise.append((a, b, c, i ))\n",
    "map, mauc = numpy.array(classwise).mean(axis = 0)[:2]\n",
    "print (map, mauc, dprime(mauc))\n",
    "# classwise.sort(key=lambda x: x[0])\n",
    "\n",
    "print(FORMAT % ('Average', map, mauc, dprime(mauc), 'Name'))\n",
    "print(SEP)\n",
    "# for c in classwise:\n",
    "#     print(c, class_name[c[3]])\n",
    "for i in range(N_CLASSES):\n",
    "    print(FORMAT % ((str(i),) + classwise[i]))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Adversarial Routine"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "pgd_dataset = torch.utils.data.TensorDataset(torch.from_numpy(feat_as[:]), torch.from_numpy(labels[:].astype('float32')))\n",
    "pgd_dataloader = torch.utils.data.DataLoader(pgd_dataset, batch_size=100, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pgd(model, dataloader, delta_global, num_steps, step_size, step_norm, eps, eps_norm, x_min, x_max, y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    model.cuda().eval()\n",
    "    targeted = y_target is not None\n",
    "    criterion = nn.BCELoss().cuda()\n",
    "    loss_global = 0\n",
    "    for epoch in range(num_steps):\n",
    "        torch.cuda.empty_cache()\n",
    "        loss_global = 0\n",
    "        dataiter = iter(dataloader)\n",
    "        for step, (feat_a, y) in enumerate(dataiter):\n",
    "            feat_a = feat_a.cuda()\n",
    "            \n",
    "            x_adv_1 = feat_a + delta_global #.to(x.device)\n",
    "            x_adv_1 = x_adv_1.clamp(x_min, x_max)\n",
    "            # only video\n",
    "           \n",
    "            N_1, T_1, V_1 = x_adv_1.shape\n",
    "#             print(N_1,T_1,V_1)\n",
    "\n",
    "            #######   data loaded ###################\n",
    "            _x_adv_1 = x_adv_1.clone().detach().requires_grad_(True).cuda()\n",
    "\n",
    "            #######   get gradient #################\n",
    "#             with torch.no_grad():\n",
    "#             model.train()# double check\n",
    "            out = model(_x_adv_1)\n",
    "            \n",
    "            temp_out = out[0].clamp(0,1).cuda()\n",
    "#             print(temp_out.shape, y.shape, temp_out.max(), temp_out.min())\n",
    "            loss = criterion(temp_out, y.cuda())\n",
    "            loss_global += loss.item()\n",
    "            loss.backward()\n",
    "#             break\n",
    "            ########   updated _x_adv ###############\n",
    "            with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "                if step_norm == 'inf':\n",
    "                    gradients = _x_adv_1.grad.sign() * step_size\n",
    "                else:\n",
    "                    # The grad is normalized here.\n",
    "                    # gradients = _x_adv.grad \n",
    "                    # Only normalize w.r.t. to feats at a timestep\n",
    "                    gradients = _x_adv_1.grad/_x_adv_1.grad.norm(step_norm, dim=-1).view(-1, T_1, 1)\n",
    "                    gradients *= step_size\n",
    "                if targeted:\n",
    "                    # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                    # w.r.t. the image data\n",
    "                    x_adv_1 -= gradients\n",
    "                else:\n",
    "                    # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                    # the model parameters\n",
    "                    x_adv_1 += gradients\n",
    "            # Project back into l_norm ball and correct range\n",
    "            if eps_norm == 'inf':\n",
    "                # Workaround as PyTorch doesn't have elementwise clip\n",
    "                x_adv_1 = torch.max(torch.min(x_adv_1, feat_a + eps), feat_a - eps)\n",
    "                delta = x_adv_1 - feat_a\n",
    "            else:\n",
    "                delta = x_adv_1 - feat_a\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "            # 100 10 2048\n",
    "            mask = delta.view(delta.shape[0], -1).norm(float(eps_norm), dim=1) <= eps #100*(20480) => 100\n",
    "            scaling_factor = delta.view(delta.shape[0], T_1, -1).norm(float(eps_norm), dim=-1) #100 10 2048 => 100 40\n",
    "            scaling_factor[mask] = eps\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta *= eps / scaling_factor.view(-1, T_1, 1)\n",
    "            delta = torch.mean(delta,0)\n",
    "            delta_global += delta * N_1 / 20098\n",
    "            \n",
    "            for name, param in model.named_parameters():\n",
    "                if param.requires_grad:\n",
    "                    try:\n",
    "                        param.grad.zero_()\n",
    "                    except:\n",
    "                        pass\n",
    "        print(f'global loss:{loss_global}')\n",
    "    return delta_global"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "global loss:3.574622900225222\n",
      "global loss:3.5811952967196703\n",
      "global loss:3.5911570321768522\n",
      "global loss:3.6069952035322785\n",
      "global loss:3.6304941214621067\n",
      "global loss:3.6661494290456176\n",
      "global loss:3.7179373959079385\n",
      "global loss:3.789903696626425\n",
      "global loss:3.856285208836198\n",
      "global loss:3.9273139648139477\n",
      "global loss:4.014958452433348\n",
      "global loss:4.134525720030069\n",
      "global loss:4.291621016338468\n",
      "global loss:4.500108953565359\n",
      "global loss:4.769953498616815\n",
      "global loss:5.113639527931809\n",
      "global loss:5.5408785585314035\n",
      "global loss:6.0441518146544695\n",
      "global loss:6.612309375777841\n",
      "global loss:7.221106166020036\n"
     ]
    }
   ],
   "source": [
    "torch.cuda.empty_cache()\n",
    "perturb = pgd(model, pgd_dataloader, torch.zeros((400,64)).cuda(), num_steps=20, step_size=0.1, step_norm='inf', eps= 0.1, eps_norm='inf', x_min=-2, x_max=2, y_target=None )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "perturbation = perturb.detach().cpu()\n",
    "perturbation_a_only = perturbation.numpy()"
   ]
  },
  {
   "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": []
  },
  {
   "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": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "import numpy as np\n",
    "temp = pickle.load(open('video_stats.pkl','rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "23.27987289428711"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.max(pickle.load(open('min_max_values.pkl','rb'))['max_frame_v'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "pgd_dataset = torch.utils.data.TensorDataset(torch.from_numpy(feat_as[:]), torch.from_numpy(feat_vs[:]), torch.from_numpy(labels[:].astype('float32')))\n",
    "pgd_dataloader = torch.utils.data.DataLoader(pgd_dataset, batch_size=100, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pgd(model, dataloader, delta_global, num_steps, step_size, step_norm, eps, eps_norm, x_min, x_max, y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    model.cuda().eval()\n",
    "    targeted = y_target is not None\n",
    "    criterion = nn.BCELoss().cuda()\n",
    "    loss_global = 0\n",
    "    for epoch in range(num_steps):\n",
    "        torch.cuda.empty_cache()\n",
    "        loss_global = 0\n",
    "        dataiter = iter(dataloader)\n",
    "        for step, (feat_a, feat_v, y) in enumerate(dataiter):\n",
    "            feat_a = feat_a.cuda()\n",
    "            feat_v = feat_v.cuda()\n",
    "            \n",
    "            x_adv_1 = feat_a + delta_global #.to(x.device)\n",
    "            x_adv_1 = x_adv_1.clamp(x_min, x_max)\n",
    "            # only video\n",
    "            x_adv_2 = feat_v \n",
    "            \n",
    "            N_1, T_1, V_1 = x_adv_1.shape\n",
    "#             print(N_1,T_1,V_1)\n",
    "            N_2, T_2, V_2 = x_adv_2.shape\n",
    "#             print(N_2,T_2,V_2)\n",
    "            #######   data loaded ###################\n",
    "            _x_adv_1 = x_adv_1.clone().detach().requires_grad_(True).cuda()\n",
    "            _x_adv_2 = x_adv_2\n",
    "            #######   get gradient #################\n",
    "#             with torch.no_grad():\n",
    "            out = model(_x_adv_1, _x_adv_2)\n",
    "#             print(out[0].shape)\n",
    "            loss = criterion(out[0], y.cuda())\n",
    "            loss_global += loss.item()\n",
    "#             print(f'loss:{loss}')\n",
    "            loss.backward()\n",
    "            ########   updated _x_adv ###############\n",
    "            with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "                if step_norm == 'inf':\n",
    "                    gradients = _x_adv_1.grad.sign() * step_size\n",
    "                else:\n",
    "                    # The grad is normalized here.\n",
    "                    # gradients = _x_adv.grad \n",
    "                    # Only normalize w.r.t. to feats at a timestep\n",
    "                    gradients = _x_adv_1.grad/_x_adv_1.grad.norm(step_norm, dim=-1).view(-1, T_1, 1)\n",
    "                    gradients *= step_size\n",
    "                if targeted:\n",
    "                    # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                    # w.r.t. the image data\n",
    "                    x_adv_1 -= gradients\n",
    "                else:\n",
    "                    # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                    # the model parameters\n",
    "                    x_adv_1 += gradients\n",
    "            # Project back into l_norm ball and correct range\n",
    "            if eps_norm == 'inf':\n",
    "                # Workaround as PyTorch doesn't have elementwise clip\n",
    "                x_adv_1 = torch.max(torch.min(x_adv_1, feat_a + eps), feat_a - eps)\n",
    "                delta = x_adv_1 - feat_a\n",
    "            else:\n",
    "                delta = x_adv_1 - feat_a\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "            # 100 10 2048\n",
    "            mask = delta.view(delta.shape[0], -1).norm(float(eps_norm), dim=1) <= eps #100*(20480) => 100\n",
    "            scaling_factor = delta.view(delta.shape[0], T_1, -1).norm(float(eps_norm), dim=-1) #100 10 2048 => 100 40\n",
    "            scaling_factor[mask] = eps\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta *= eps / scaling_factor.view(-1, T_1, 1)\n",
    "            delta = torch.mean(delta,0)\n",
    "            delta_global += delta * N_1 / 20098\n",
    "            \n",
    "            for name, param in model.named_parameters():\n",
    "                if param.requires_grad:\n",
    "                    try:\n",
    "                        param.grad.zero_()\n",
    "                    except:\n",
    "                        pass\n",
    "        print(f'global loss:{loss_global}')\n",
    "    return delta_global"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<generator object Module.named_parameters at 0x7f1e3d762c80>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.named_parameters()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "ename": "TypeError",
     "evalue": "forward() takes 2 positional arguments but 3 were given",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-44-a05befb925fd>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mempty_cache\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mperturb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mpgd\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpgd_dataloader\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mzeros\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m400\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m64\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnum_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m20\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstep_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0.1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstep_norm\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'inf'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0meps\u001b[0m\u001b[0;34m=\u001b[0m \u001b[0;36m0.1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0meps_norm\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'inf'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx_min\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx_max\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_target\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mNone\u001b[0m \u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m<ipython-input-42-57af21057140>\u001b[0m in \u001b[0;36mpgd\u001b[0;34m(model, dataloader, delta_global, num_steps, step_size, step_norm, eps, eps_norm, x_min, x_max, y_target)\u001b[0m\n\u001b[1;32m     27\u001b[0m             \u001b[0;31m#######   get gradient #################\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     28\u001b[0m \u001b[0;31m#             with torch.no_grad():\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 29\u001b[0;31m             \u001b[0mout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0m_x_adv_1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0m_x_adv_2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     30\u001b[0m \u001b[0;31m#             print(out[0].shape)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     31\u001b[0m             \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcriterion\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m_call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1100\u001b[0m         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks\n\u001b[1;32m   1101\u001b[0m                 or _global_forward_hooks or _global_forward_pre_hooks):\n\u001b[0;32m-> 1102\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mforward_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1103\u001b[0m         \u001b[0;31m# Do not call functions when jit is used\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1104\u001b[0m         \u001b[0mfull_backward_hooks\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnon_full_backward_hooks\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTypeError\u001b[0m: forward() takes 2 positional arguments but 3 were given"
     ]
    }
   ],
   "source": [
    "torch.cuda.empty_cache()\n",
    "perturb = pgd(model, pgd_dataloader, torch.zeros((400,64)).cuda(), num_steps=20, step_size=0.1, step_norm='inf', eps= 0.1, eps_norm='inf', x_min=-2, x_max=2, y_target=None )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "perturbation = perturb.detach().cpu()\n",
    "perturbation_a_only = perturbation.numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "pgd_dataset = torch.utils.data.TensorDataset(torch.from_numpy(feat_as[:]), torch.from_numpy(feat_vs[:]), torch.from_numpy(labels[:].astype('float32')))\n",
    "pgd_dataloader = torch.utils.data.DataLoader(pgd_dataset, batch_size=100, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pgd(model, dataloader, delta_global, num_steps, step_size, step_norm, eps, eps_norm, x_min, x_max, y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    model.cuda().eval()\n",
    "    targeted = y_target is not None\n",
    "    criterion = nn.BCELoss().cuda()\n",
    "    loss_global = 0\n",
    "    for epoch in range(num_steps):\n",
    "        torch.cuda.empty_cache()\n",
    "        loss_global = 0\n",
    "        dataiter = iter(dataloader)\n",
    "        for step, (feat_a, feat_v, y) in enumerate(dataiter):\n",
    "            feat_a = feat_a.cuda()\n",
    "            x_adv_1 = feat_a\n",
    "            feat_v = feat_v.cuda()\n",
    "            # only video\n",
    "            x_adv_2 = feat_v + delta_global #.to(x.device)\n",
    "            x_adv_2 = x_adv_2.clamp(x_min, x_max)\n",
    "            \n",
    "            N_1, T_1, V_1 = x_adv_1.shape\n",
    "#             print(N_1,T_1,V_1)\n",
    "            N_2, T_2, V_2 = x_adv_2.shape\n",
    "#             print(N_2,T_2,V_2)\n",
    "            #######   data loaded ###################\n",
    "            _x_adv_1 = feat_a\n",
    "            _x_adv_2 = x_adv_2.clone().detach().requires_grad_(True).cuda()\n",
    "            #######   get gradient #################\n",
    "#             with torch.no_grad():\n",
    "            out = model(_x_adv_1, _x_adv_2)\n",
    "#             print(out[0].shape)\n",
    "            loss = criterion(out[0], y.cuda())\n",
    "            loss_global += loss.item()\n",
    "#             print(f'loss:{loss}')\n",
    "            loss.backward()\n",
    "            ########   updated _x_adv ###############\n",
    "            with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "                if step_norm == 'inf':\n",
    "                    gradients = _x_adv_2.grad.sign() * step_size\n",
    "                else:\n",
    "                    # The grad is normalized here.\n",
    "                    # gradients = _x_adv.grad \n",
    "                    # Only normalize w.r.t. to feats at a timestep\n",
    "                    gradients = _x_adv_2.grad/_x_adv_2.grad.norm(step_norm, dim=-1).view(-1, T_2, 1)\n",
    "                    gradients *= step_size\n",
    "                if targeted:\n",
    "                    # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                    # w.r.t. the image data\n",
    "                    x_adv_2 -= gradients\n",
    "                else:\n",
    "                    # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                    # the model parameters\n",
    "                    x_adv_2 += gradients\n",
    "            # Project back into l_norm ball and correct range\n",
    "            if eps_norm == 'inf':\n",
    "                # Workaround as PyTorch doesn't have elementwise clip\n",
    "                x_adv_2 = torch.max(torch.min(x_adv_2, feat_v + eps), feat_v - eps)\n",
    "                delta = x_adv_2 - feat_v\n",
    "            else:\n",
    "                delta = x_adv_2 - feat_v\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "            # 100 10 2048\n",
    "            mask = delta.view(delta.shape[0], -1).norm(float(eps_norm), dim=1) <= eps #100*(20480) => 100\n",
    "            scaling_factor = delta.view(delta.shape[0], T_2, -1).norm(float(eps_norm), dim=-1) #100 10 2048 => 100 40\n",
    "            scaling_factor[mask] = eps\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta *= eps / scaling_factor.view(-1, T_2, 1)\n",
    "            delta = torch.mean(delta,0)\n",
    "            delta_global += delta * N_2 / 20098\n",
    "            \n",
    "            for name, param in model.named_parameters():\n",
    "                if param.requires_grad:\n",
    "                    try:\n",
    "                        param.grad.zero_()\n",
    "                    except:\n",
    "                        pass\n",
    "        print(f'global loss:{loss_global}')\n",
    "    return delta_global"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "global loss:4.052812464535236\n",
      "global loss:4.084545601159334\n",
      "global loss:4.138830292969942\n",
      "global loss:4.233388312160969\n",
      "global loss:4.399552414193749\n",
      "global loss:4.65200812369585\n",
      "global loss:4.9532044641673565\n",
      "global loss:5.246565429493785\n",
      "global loss:5.495519766584039\n",
      "global loss:5.700055651366711\n",
      "global loss:5.8773121275007725\n",
      "global loss:6.027423532679677\n",
      "global loss:6.15042126365006\n",
      "global loss:6.245867017656565\n",
      "global loss:6.314192967489362\n",
      "global loss:6.362946743145585\n",
      "global loss:6.397345703095198\n",
      "global loss:6.421141775324941\n",
      "global loss:6.437668612226844\n",
      "global loss:6.445729523897171\n"
     ]
    }
   ],
   "source": [
    "torch.cuda.empty_cache()\n",
    "perturb = pgd(model, pgd_dataloader, torch.zeros((10,2048)).cuda(), num_steps=20, step_size=0.1, step_norm='inf', eps= 0.1, eps_norm='inf', x_min=-0.51, x_max=1.51, y_target=None )\n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "perturbation = perturb.detach().cpu()\n",
    "perturbation_v_only = perturbation.numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "pgd_dataset = torch.utils.data.TensorDataset(torch.from_numpy(feat_as[:]), torch.from_numpy(feat_vs[:]), torch.from_numpy(labels[:].astype('float32')))\n",
    "pgd_dataloader = torch.utils.data.DataLoader(pgd_dataset, batch_size=100, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pgd(model, dataloader, delta_global_a, delta_global_v, num_steps, step_size, step_norm, eps, eps_norm, x_min_a, x_max_a, x_min_v, x_max_v, y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    model.cuda().eval()\n",
    "    targeted = y_target is not None\n",
    "    criterion = nn.BCELoss().cuda()\n",
    "    loss_global = 0\n",
    "    for epoch in range(num_steps):\n",
    "        torch.cuda.empty_cache()\n",
    "        loss_global = 0\n",
    "        dataiter = iter(dataloader)\n",
    "        for step, (feat_a, feat_v, y) in enumerate(dataiter):\n",
    "            feat_a = feat_a.cuda()\n",
    "            feat_v = feat_v.cuda()\n",
    "            # only video\n",
    "            x_adv_1 = feat_a + delta_global_a #.to(x.device)\n",
    "            x_adv_1 = x_adv_1.clamp(x_min_a, x_max_a)\n",
    "            \n",
    "            x_adv_2 = feat_v + delta_global_v #.to(x.device)\n",
    "            x_adv_2 = x_adv_2.clamp(x_min_v, x_max_v)\n",
    "            \n",
    "            N_1, T_1, V_1 = x_adv_1.shape\n",
    "#             print(N_1,T_1,V_1)\n",
    "            N_2, T_2, V_2 = x_adv_2.shape\n",
    "#             print(N_2,T_2,V_2)\n",
    "            #######   data loaded ###################\n",
    "            _x_adv_1 = x_adv_1.clone().detach().requires_grad_(True).cuda()\n",
    "            _x_adv_2 = x_adv_2.clone().detach().requires_grad_(True).cuda()\n",
    "            #######   get gradient #################\n",
    "#             with torch.no_grad():\n",
    "            out = model(_x_adv_1, _x_adv_2)\n",
    "#             print(out[0].shape)\n",
    "            loss = criterion(out[0], y.cuda())\n",
    "            loss_global += loss.item()\n",
    "#             print(f'loss:{loss}')\n",
    "            loss.backward()\n",
    "            ########   updated _x_adv ###############\n",
    "            with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "                if step_norm == 'inf':\n",
    "                    gradients_a = _x_adv_1.grad.sign() * step_size\n",
    "                    gradients_v = _x_adv_2.grad.sign() * step_size\n",
    "                else:\n",
    "                    # The grad is normalized here.\n",
    "                    # gradients = _x_adv.grad \n",
    "                    # Only normalize w.r.t. to feats at a timestep\n",
    "                    gradients_a = _x_adv_1.grad/_x_adv_1.grad.norm(step_norm, dim=-1).view(-1, T_1, 1)\n",
    "                    gradients_a *= step_size\n",
    "                    gradients_v = _x_adv_2.grad/_x_adv_2.grad.norm(step_norm, dim=-1).view(-1, T_2, 1)\n",
    "                    gradients_v *= step_size\n",
    "                if targeted:\n",
    "                    # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                    # w.r.t. the image data\n",
    "                    x_adv_1 -= gradients_a\n",
    "                    x_adv_2 -= gradients_v\n",
    "                else:\n",
    "                    # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                    # the model parameters\n",
    "                    x_adv_1 += gradients_a\n",
    "                    x_adv_2 += gradients_v\n",
    "            # Project back into l_norm ball and correct range\n",
    "            if eps_norm == 'inf':\n",
    "                # Workaround as PyTorch doesn't have elementwise clip\n",
    "                x_adv_1 = torch.max(torch.min(x_adv_1, feat_a + eps), feat_a - eps)\n",
    "                delta_a = x_adv_1 - feat_a\n",
    "                \n",
    "                x_adv_2 = torch.max(torch.min(x_adv_2, feat_v + eps), feat_v - eps)\n",
    "                delta_v = x_adv_2 - feat_v\n",
    "            else:\n",
    "                delta_a = x_adv_1 - feat_a\n",
    "                delta_v = x_adv_2 - feat_v\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "            # 100 10 2048\n",
    "            mask_a = delta_a.view(delta_a.shape[0], -1).norm(float(eps_norm), dim=1) <= eps #100*(20480) => 100\n",
    "            scaling_factor_a = delta_a.view(delta_a.shape[0], T_1, -1).norm(float(eps_norm), dim=-1) #100 10 2048 => 100 40\n",
    "            scaling_factor_a[mask_a] = eps\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta_a *= eps / scaling_factor_a.view(-1, T_1, 1)\n",
    "            delta_a = torch.mean(delta_a,0)\n",
    "            delta_global_a += delta_a * N_2 / 20098\n",
    "            \n",
    "            \n",
    "            mask_v = delta_v.view(delta_v.shape[0], -1).norm(float(eps_norm), dim=1) <= eps #100*(20480) => 100\n",
    "            scaling_factor_v = delta_v.view(delta_v.shape[0], T_2, -1).norm(float(eps_norm), dim=-1) #100 10 2048 => 100 40\n",
    "            scaling_factor_v[mask_v] = eps\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta_v *= eps / scaling_factor_v.view(-1, T_2, 1)\n",
    "            delta_v = torch.mean(delta_v,0)\n",
    "            delta_global_v += delta_v * N_2 / 20098\n",
    "            \n",
    "            for name, param in model.named_parameters():\n",
    "                if param.requires_grad:\n",
    "                    try:\n",
    "                        param.grad.zero_()\n",
    "                    except:\n",
    "                        pass\n",
    "                    \n",
    "        print(f'global loss:{loss_global}')\n",
    "    return delta_global_a, delta_global_v"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "global loss:4.026436904445291\n",
      "global loss:4.06182736903429\n",
      "global loss:4.122058970853686\n",
      "global loss:4.228114370256662\n",
      "global loss:4.4108707923442125\n",
      "global loss:4.690501457080245\n",
      "global loss:5.034160507842898\n",
      "global loss:5.389105902984738\n",
      "global loss:5.7030198108404875\n",
      "global loss:5.975775713101029\n",
      "global loss:6.233841169625521\n",
      "global loss:6.484253194183111\n",
      "global loss:6.729882122948766\n",
      "global loss:6.97159150056541\n",
      "global loss:7.210782241076231\n",
      "global loss:7.461076285690069\n",
      "global loss:7.722201336175203\n",
      "global loss:7.994646478444338\n",
      "global loss:8.276697803288698\n",
      "global loss:8.563009224832058\n"
     ]
    }
   ],
   "source": [
    "torch.cuda.empty_cache()\n",
    "perturb = pgd(model, pgd_dataloader, torch.zeros((400,64)).cuda(), torch.zeros((10,2048)).cuda(), num_steps=20, step_size=0.1, step_norm='inf', eps= 0.1, eps_norm='inf', x_min_a = -2, x_max_a = 2, x_min_v=-0.51, x_max_v=1.51, y_target=None )\n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "perturbation_a = perturb[0].detach().cpu()\n",
    "perturbation_a = perturbation_a.numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "perturbation_v = perturb[1].detach().cpu()\n",
    "perturbation_v = perturbation_v.numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "res = feat_as[0, :]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp = perturbation_a.T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(64, 40)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "temp[:,range(100,0,-1)].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp[:,300:] = temp[:,range(100,0,-1)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "plt.imshow(res.T,interpolation='nearest', aspect='auto')\n",
    "plt.xlabel('Time frame')\n",
    "plt.ylabel('Mel filter bank')\n",
    "plt.colorbar()\n",
    "plt.savefig('Audio raw')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "plt.imshow(res.T+temp,interpolation='nearest', aspect='auto')\n",
    "plt.xlabel('Time frame')\n",
    "plt.ylabel('Mel filter bank')\n",
    "plt.colorbar()\n",
    "plt.savefig('Audio attack')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.cuda.empty_cache()\n",
    "perturb = pgd(model, pgd_dataloader, perturb, num_steps=1, step_size=0.01, step_norm=2, eps= 0.1, eps_norm=2, x_min=-10, x_max=20, y_target=None )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "ename": "FileNotFoundError",
     "evalue": "[Errno 2] No such file or directory: 'delta_global_eps_norm_2_eps_0_3_delta_force.pkl'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mFileNotFoundError\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-47-01875afdae36>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mmatplotlib\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpyplot\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mplt\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mpickle\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0mtemp\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mpickle\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'delta_global_eps_norm_2_eps_0_3_delta_force.pkl'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'rb'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      5\u001b[0m \u001b[0;31m# temp[:, 20:64] =0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0mplt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0minterpolation\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'nearest'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maspect\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'auto'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mFileNotFoundError\u001b[0m: [Errno 2] No such file or directory: 'delta_global_eps_norm_2_eps_0_3_delta_force.pkl'"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import pickle\n",
    "temp = pickle.load(open('delta_global_eps_norm_2_eps_0_3_delta_force.pkl', 'rb'))\n",
    "# temp[:, 20:64] =0\n",
    "plt.imshow(temp.T,interpolation='nearest', aspect='auto')\n",
    "plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.colorbar.Colorbar at 0x7e7e4a5f0590>"
      ]
     },
     "execution_count": 132,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "perturbation = perturb.detach().cpu()\n",
    "perturbation = perturbation.numpy()\n",
    "plt.imshow(perturbation,interpolation='nearest', aspect='auto')\n",
    "plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.cuda.empty_cache()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "pickle.dump(perturb, open('delta_video_trial.pkl','wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 194,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get middle hidden layer"
   ]
  },
  {
   "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": []
  },
  {
   "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": [
    "def pgd(model, x, y, num_steps, step_size, step_norm, eps, eps_norm, x_min, x_max, y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    model.eval()\n",
    "    x,y = torch.from_numpy(x).cuda(), torch.from_numpy(y).cuda()\n",
    "    x_adv = x.clone().detach().requires_grad_(True)#.to(x.device)\n",
    "    N, T, V = x.shape\n",
    "    print('NTV:', N, T, V)\n",
    "    targeted = y_target is not None\n",
    "    num_channels = x.shape[1]\n",
    "    criterion = nn.BCELoss()    \n",
    "    \n",
    "    for i in range(num_steps):\n",
    "        _x_adv = x_adv.clone().detach().requires_grad_(True)\n",
    "#         x_adv = torch.zeros((400,64)).detach().requires_grad_(True)\n",
    "        out = model(_x_adv)[0] # N x T x 1\n",
    "        print(out.shape)\n",
    "        loss = criterion(out, y)\n",
    "        print(f'loss:{loss}')\n",
    "        #prediction = model(_x_adv)\n",
    "        #loss = criterion(prediction, y_target if targeted else y)\n",
    "        loss.backward()\n",
    "\n",
    "        with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "            if step_norm == 'inf':\n",
    "                gradients = _x_adv.grad.sign() * step_size\n",
    "            else:\n",
    "                # The grad is normalized here.\n",
    "                #gradients = _x_adv.grad \n",
    "                # Only normalize w.r.t. to feats at a timestep\n",
    "                gradients = _x_adv.grad/_x_adv.grad.norm(step_norm, dim=-1).view(-1, T, 1)\n",
    "                gradients *= step_size\n",
    "#                 print(f'gradients shape:{gradients.shape}')\n",
    "            if targeted:\n",
    "                # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                # w.r.t. the image data\n",
    "                x_adv -= gradients\n",
    "            else:\n",
    "                # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                # the model parameters\n",
    "#                 print(f'gradient:{gradients[0]}')\n",
    "#                 print(f'gradient:{gradients[1]}')\n",
    "                x_adv += gradients\n",
    "\n",
    "        # Project back into l_norm ball and correct range\n",
    "        if eps_norm == 'inf':\n",
    "            # Workaround as PyTorch doesn't have elementwise clip\n",
    "            x_adv = torch.max(torch.min(x_adv, x + eps), x - eps)\n",
    "        else:\n",
    "            delta = x_adv - x\n",
    "#             print(f'deltashape:{delta.shape}')\n",
    "\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "# #             mask = delta.view(delta.shape[0], -1).norm(eps_norm, dim=1) <= eps\n",
    "#             mask = torch.max(delta.view(delta.shape[0], -1).norm(eps_norm, dim=1), dim=0, keepdim=True) <= eps #1 400 64\n",
    "#             mask, ind = torch.max(delta.view(delta.shape[0], -1).norm(eps_norm, dim=1), dim=0, keepdim=True) #1 400 64\n",
    "#             print(f'mask shape1:{mask.shape}')\n",
    "# #             mask = mask.repeat(mask, N, 1)\n",
    "# #             print(f'mask shape2:{mask.shape}')\n",
    "#             # 100 400 64\n",
    "#             scaling_factor = delta.view(delta.shape[0], T, -1).norm(eps_norm, dim=-1)\n",
    "#             scaling_factor[mask] = eps\n",
    "            \n",
    "#             \n",
    "            \n",
    "            mask = delta.view(delta.shape[0], -1).norm(eps_norm, dim=1) <= eps #100*(25600) => 100\n",
    "#             print(f'mask shape: {mask.shape}')\n",
    "            scaling_factor = delta.view(delta.shape[0], T, -1).norm(eps_norm, dim=-1)\n",
    "            #100 400 64 => 100 400\n",
    "#             print(f'scaling shape: {scaling_factor.shape}')\n",
    "#             print(mask)\n",
    "            scaling_factor[mask] = eps\n",
    "#             print(scaling_factor)\n",
    "            #print(scaling_factor.shape)\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta *= eps / scaling_factor.view(-1, T, 1)\n",
    "            delta = torch.mean(delta,0)\n",
    "            x_adv = x + delta\n",
    "#             print(f'delta:{delta.shape}')\n",
    "            # x_adv = x + torch.audio.MelSpec(delta)\n",
    "        x_adv = x_adv.clamp(x_min, x_max)\n",
    "\n",
    "    return x_adv.detach()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "perturb = pgd(model, gas_eval_x, gas_eval_y, num_steps=1000, step_size=0.01, step_norm=2, eps= 0.1, eps_norm=2, x_min=-10, x_max=20, y_target=None )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(perturbation)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "perturbation = perturb[0].detach().cpu() - x[0]\n",
    "perturbation = perturbation.numpy()\n",
    "perturbation[perturbation< np.max(perturbation)-0.05] = 0\n",
    "plt.imshow(perturbation,interpolation='nearest', aspect='auto')\n",
    "plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(hash_val[0])\n",
    "result = model(perturb)\n",
    "print(torch.argmax(result[0]))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "result_origin = model(torch.from_numpy(x[0:1]).cuda())\n",
    "print(result_origin[0].shape)\n",
    "print(torch.argmax(result_origin[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(np.argmax(y[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "perturbation = (x[0])\n",
    "plt.imshow(perturbation.T,interpolation='nearest', aspect='auto')\n",
    "plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for batch in range(1, args.ckpt_size + 1):\n",
    "        x, y = next(train_gen)\n",
    "        global_prob = model(x)[0]\n",
    "        global_prob.clamp_(min = 1e-7, max = 1 - 1e-7)\n",
    "        loss = criterion(global_prob, y)\n",
    "        if args.gradient_accumulation > 1:\n",
    "            loss = loss / args.gradient_accumulation\n",
    "        #if n_gpu > 1:\n",
    "        #    loss = loss.mean()\n",
    "        train_loss += loss.item()\n",
    "        if numpy.isnan(train_loss) or numpy.isinf(train_loss): break\n",
    "        loss.backward()\n",
    "        global_step += 1\n",
    "        if global_step % args.gradient_accumulation == 0:\n",
    "            optimizer.step()\n",
    "            if args.scheduler == 'warmup-decay':\n",
    "                scheduler.step() \n",
    "            optimizer.zero_grad()\n",
    "        if batch % 500 == 0:\n",
    "            sys.stderr.write('Checkpoint %d, Batch %d / %d, avg train loss = %f\\r' % \\\n",
    "                            (checkpoint, batch, args.ckpt_size, train_loss / batch))\n",
    "            tb_writer.add_scalar('lr', optimizer.param_groups[0]['lr'], global_step)\n",
    "            tb_writer.add_scalar('loss', train_loss / batch, global_step)\n",
    "            \n",
    "        del x, y, global_prob, loss         # This line and next line: to save GPU memory\n",
    "        torch.cuda.empty_cache()            # I don't know if they're useful or not\n",
    "    train_loss /= args.ckpt_size\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(opt, model, adv_jam, trainloader, max_batches = 100):\n",
    "    model.eval()\n",
    "        \n",
    "    batches = 0\n",
    "    total = 0\n",
    "    error = 0\n",
    "    loss_ = 0.\n",
    "    torch.manual_seed(0)\n",
    "\n",
    "    for X_batch, y_batch in trainloader:\n",
    "        # forward pass\n",
    "        X_batch, y_batch = X_batch.cuda(), y_batch.cuda()\n",
    "        # plot(X_batch[0], y_batch[0])\n",
    "        output = model(adv_jam(X))\n",
    "        opt.zero_grad()\n",
    "        neg_loss = -nn.CrossEntropyLoss()(output, y)\n",
    "\n",
    "        neg_loss.backward()\n",
    "        opt.step()\n",
    "\n",
    "\n",
    "\n",
    "        # Evaluate performance\n",
    "        adv_out = model(X_adv)\n",
    "        adv_prediction = sigmoid(adv_out.squeeze().cpu().detach())>0.5\n",
    "        adv_accuracy = (adv_prediction.numpy()==y_batch.cpu().numpy()).mean()\n",
    "        print(\"Accuracy = {0}\".format(adv_accuracy))\n",
    "                # project into feasible region\n",
    "        adv.jam.param1.clamp_(0,1)\n",
    "        adv.jam.param2.clamp_(1, 40)\n",
    "        adv.jam.param3.clamp_(0,224)\n",
    "        total += X.shape[0]\n",
    "        error += (output.max(1)[1] != y).sum().item()\n",
    "        loss_ += nn.CrossEntropyLoss(size_average=False)(output, y).item()\n",
    "            batches += 1\n",
    "            if batches > max_batches:\n",
    "                break\n",
    "\n",
    "        return error / total, loss_ / total"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def projected_gradient_descent(model, x, y, num_steps = 20, step_size = 0.1, step_norm=2, eps = 10, eps_norm=2,\n",
    "                               clamp=(x_min, x_max), y_target=None):\n",
    "    \"\"\"Performs the projected gradient descent attack on a batch of images.\"\"\"\n",
    "    x_adv = x.clone().detach().requires_grad_(True)#.to(x.device)\n",
    "    N, T, V = x.shape\n",
    "    targeted = y_target is not None\n",
    "    num_channels = x.shape[1]\n",
    "    criterion = nn.BCELoss()\n",
    "\n",
    "    for i in range(num_steps):\n",
    "        _x_adv = x_adv.clone().detach().requires_grad_(True)\n",
    "        \n",
    "        out = model(_x_adv) # N x T x 1\n",
    "        out_flatten = out.reshape(-1) \n",
    "        y_flatten = y.reshape(-1)\n",
    "        loss = criterion(out_flatten, y_flatten)\n",
    "        \n",
    "        #prediction = model(_x_adv)\n",
    "        #loss = criterion(prediction, y_target if targeted else y)\n",
    "        loss.backward()\n",
    "\n",
    "        with torch.no_grad():\n",
    "            # Force the gradient step to be a fixed size in a certain norm\n",
    "            if step_norm == 'inf':\n",
    "                gradients = _x_adv.grad.sign() * step_size\n",
    "            else:\n",
    "                # The grad is normalized here.\n",
    "                #gradients = _x_adv.grad \n",
    "                # Only normalize w.r.t. to feats at a timestep\n",
    "                gradients = _x_adv.grad/_x_adv.grad.norm(step_norm, dim=-1).view(-1, T, 1)\n",
    "                gradients *= step_size\n",
    "                \n",
    "            if targeted:\n",
    "                # Targeted: Gradient descent with on the loss of the (incorrect) target label\n",
    "                # w.r.t. the image data\n",
    "                x_adv -= gradients\n",
    "            else:\n",
    "                # Untargeted: Gradient ascent on the loss of the correct label w.r.t.\n",
    "                # the model parameters\n",
    "                x_adv += gradients\n",
    "\n",
    "        # Project back into l_norm ball and correct range\n",
    "        if eps_norm == 'inf':\n",
    "            # Workaround as PyTorch doesn't have elementwise clip\n",
    "            x_adv = torch.max(torch.min(x_adv, x + eps), x - eps)\n",
    "        else:\n",
    "            delta = x_adv - x\n",
    "\n",
    "            # Assume x and x_adv are batched tensors where the first dimension is\n",
    "            # a batch dimension\n",
    "            mask = delta.view(delta.shape[0], -1).norm(eps_norm, dim=1) <= eps\n",
    "\n",
    "            scaling_factor = delta.view(delta.shape[0], T, -1).norm(eps_norm, dim=-1)\n",
    "            scaling_factor[mask] = eps\n",
    "            #print(scaling_factor.shape)\n",
    "            # .view() assumes batched images as a 4D Tensor\n",
    "            delta *= eps / scaling_factor.view(-1, T, 1)\n",
    "\n",
    "            x_adv = x + delta\n",
    "            \n",
    "        x_adv = x_adv.clamp(*clamp)\n",
    "\n",
    "    return x_adv.detach()"
   ]
  },
  {
   "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}