02_mnist_basics_clean.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision.all import *\n",
    "from fastbook import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from IPython.display import clear_output, DisplayHandle\n",
    "def update_patch(self, obj):\n",
    "    clear_output(wait=True)\n",
    "    self.display(obj)\n",
    "DisplayHandle.update = update_patch"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training a Neural Net\n",
    "This example trains a simple neural net on the mnist sample data set. We predict if an image is a 3 or 7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "path = untar_data(URLs.MNIST_SAMPLE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import Dataset\n",
    "from torchvision.transforms import ToTensor\n",
    "\n",
    "class MNISTDataset(Dataset):\n",
    "    def __init__(self, path, transform=None, target_transform=None):\n",
    "        self.path = path\n",
    "        self.transform = transform\n",
    "        self.target_transform = target_transform\n",
    "        self.image_paths = list(flatten([(i.ls()) for i in self.path.ls()]))\n",
    "    \n",
    "    def __len__(self):\n",
    "        return sum([len((self.path/str(i)).ls()) for i in self.path.ls()])\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        img_path = self.image_paths[idx]\n",
    "        img = Image.open(img_path)\n",
    "        label = img_path.parent.name\n",
    "        if self.transform:\n",
    "            img = self.transform(img)\n",
    "        if self.target_transform:\n",
    "            label = self.target_transform(label)\n",
    "        return img, torch.tensor(label, dtype=torch.float32).unsqueeze(0)\n",
    "    \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_ds = MNISTDataset(path/\"train\", transform=ToTensor(), target_transform=lambda x: x == \"3\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_sample(ds, rows, cols, seed=42):\n",
    "    rand = random.Random(seed)\n",
    "    figure = plt.figure(figsize=(8, 8))\n",
    "    for i in range(1, cols * rows +1):\n",
    "        sample_idx = rand.randrange(len(ds))\n",
    "        img, label = ds[sample_idx]\n",
    "        figure.add_subplot(rows, cols, i)\n",
    "        plt.title(label)\n",
    "        plt.axis(\"off\")\n",
    "        plt.imshow(img.squeeze(), cmap=\"gray\")\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x800 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_sample(train_ds, 3, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "valid_ds = MNISTDataset(path/\"valid\", transform=ToTensor(), target_transform=lambda x: x == \"3\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x800 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_sample(valid_ds, 3, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "\n",
    "train_dl = DataLoader(train_ds, batch_size=64, shuffle=True)\n",
    "valid_dl = DataLoader(valid_ds, batch_size=64)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tqdm.notebook import tqdm\n",
    "from pandas import DataFrame\n",
    "from IPython.display import display\n",
    "from ipywidgets import Box\n",
    "\n",
    "class JLearner:\n",
    "    def __init__(self, model, loss_func, opt_func, train_dl, valid_dl, metrics=None):\n",
    "        #self.device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "        self.device = torch.device(\"cpu\")\n",
    "        self.model = model.to(self.device)\n",
    "        self.loss_func = loss_func\n",
    "        self.opt_func = opt_func\n",
    "        self.train_dl = train_dl\n",
    "        self.valid_dl = valid_dl\n",
    "        self.metrics = metrics if type(metrics) in (list, tuple) else [metrics]\n",
    "        self.epoch_results = []\n",
    "        self.o = True\n",
    "\n",
    "    def fit(self, epochs):\n",
    "        self.validate_epoch()\n",
    "        for epoch in tqdm(range(epochs), desc=\"Epochs\"):\n",
    "            self.train_epoch()\n",
    "            with torch.no_grad():\n",
    "                self.validate_epoch()\n",
    "\n",
    "    def train_epoch(self):\n",
    "\n",
    "        for xs, ys in self.train_dl:\n",
    "            xs, ys = xs.to(self.device), ys.to(self.device)\n",
    "            self.train_batch(xs, ys)\n",
    "    \n",
    "    def train_batch(self, xs, ys):\n",
    "        preds = self.model(xs)\n",
    "        loss = self.loss_func(preds, ys)\n",
    "        loss.backward()\n",
    "        self.opt_func.step()\n",
    "        self.opt_func.zero_grad()\n",
    "\n",
    "    def validate_epoch(self):\n",
    "        batch_results = []\n",
    "        for xs, ys in self.valid_dl:\n",
    "            xs, ys = xs.to(self.device), ys.to(self.device)\n",
    "            loss, metrics = self.validate_batch(xs, ys)\n",
    "            batch_results.append([loss] + [metric.item() for _, metric in metrics])\n",
    "        self.epoch_results.append(np.mean(batch_results, axis=0))\n",
    "        df = DataFrame(self.epoch_results, columns=[\"loss\"] + [metric.__name__ for metric in self.metrics])\n",
    "        self.o = display(df, display_id=\"validation_results\", clear=True)\n",
    "\n",
    "\n",
    "    def validate_batch(self, xs, ys):\n",
    "        preds = self.model(xs)\n",
    "        loss = self.loss_func(preds, ys)\n",
    "        metrics = [(metric.__name__, metric(preds, ys)) for metric in self.metrics]\n",
    "        return loss.item(), metrics\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mnist_loss(predictions, targets):\n",
    "    return torch.where(targets==1, 1-predictions, predictions).mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def batch_accuracy(xb, yb):\n",
    "    correct = (xb>0.5) == yb\n",
    "    return correct.float().mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = nn.Sequential(nn.Flatten(), nn.Linear(28*28, 30), nn.ReLU(), nn.Linear(30, 1), nn.Sigmoid())\n",
    "loss_func = mnist_loss\n",
    "opt_func = SGD(model.parameters(), lr=0.1)\n",
    "learn = JLearner(model, loss_func, opt_func, train_dl, valid_dl, metrics=batch_accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>loss</th>\n",
       "      <th>batch_accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.509987</td>\n",
       "      <td>0.396864</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.039550</td>\n",
       "      <td>0.971680</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.031704</td>\n",
       "      <td>0.975586</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.028291</td>\n",
       "      <td>0.977051</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.026121</td>\n",
       "      <td>0.979492</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0.024519</td>\n",
       "      <td>0.979004</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       loss  batch_accuracy\n",
       "0  0.509987        0.396864\n",
       "1  0.039550        0.971680\n",
       "2  0.031704        0.975586\n",
       "3  0.028291        0.977051\n",
       "4  0.026121        0.979492\n",
       "5  0.024519        0.979004"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit(5)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Adversarial Example\n",
    "This example tries to convert a 7 to a 3 using the same techniques above"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([0.])"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 100x100 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "img, label = valid_ds[1325]\n",
    "show_image(img.squeeze(), cmap=\"gray\")\n",
    "display(label)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_adv_img(img, noise_img):\n",
    "    adv_img = img + (2*sigmoid(noise_img)-1)\n",
    "    adv_img = torch.clamp(adv_img, 0, 1)\n",
    "    #adv_img = img + sigmoid(noise_img)\n",
    "    return adv_img\n",
    "    #return img "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def adv_model(img, noise_img):\n",
    "    return model(calc_adv_img(img, noise_img))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 100x100 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "noise_img = torch.zeros(img.shape, requires_grad=True)\n",
    "show_image(calc_adv_img(img, noise_img).squeeze(), cmap=\"gray\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def adv_loss(prediction, noise_img, original_img):\n",
    "    noise_power = noise_img.abs().sum()\n",
    "    return -5.0*torch.log(prediction) + 0.2*noise_power\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[0.0013]], grad_fn=<SigmoidBackward0>)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "noise_img = torch.zeros(img.shape, requires_grad=True)\n",
    "img_dev = img.to(learn.device)\n",
    "noise_dev = noise_img.to(learn.device)\n",
    "model(img_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "25.033588409423828 0.019757194444537163 27.062002182006836\n",
      "9.006372451782227 0.8003026247024536 39.462730407714844\n",
      "8.898905754089355 0.8020345568656921 38.97943878173828\n",
      "8.877422332763672 0.7983251810073853 38.756126403808594\n",
      "8.913603782653809 0.7988259792327881 38.95271301269531\n",
      "8.897480964660645 0.7983980178833008 38.85870361328125\n",
      "8.965872764587402 0.806667685508728 39.458274841308594\n",
      "8.89414119720459 0.7941275835037231 38.70792770385742\n",
      "8.91778564453125 0.7978670597076416 38.94359588623047\n",
      "8.936525344848633 0.7982958555221558 39.050724029541016\n"
     ]
    }
   ],
   "source": [
    "num_epoch = 1000\n",
    "sgd = SGD([noise_dev], lr=0.11)\n",
    "for epoch in range(num_epoch):\n",
    "    preds = adv_model(img_dev, noise_dev)\n",
    "    #if preds.item() > 0.6:\n",
    "    #    break\n",
    "    loss = adv_loss(preds, noise_dev, img_dev)\n",
    "    loss.backward()\n",
    "    sgd.step()\n",
    "    sgd.zero_grad()\n",
    "    with torch.no_grad():\n",
    "        preds = adv_model(img_dev, noise_dev)\n",
    "        loss = adv_loss(preds, noise_dev, img_dev)\n",
    "        if epoch % (num_epoch/10) == 0:\n",
    "            print(loss.item(), preds.item(), noise_dev.abs().sum().item())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f6643da9690>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(noise_dev.squeeze().detach().numpy(), cmap=\"gray\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f6643eae150>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAn8AAAFFCAYAAACdR4utAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAdf0lEQVR4nO3de5BWdf0H8M8DCwgkhBqoTYm3MpXUIIv7bQqTMSPyXoKJZU4TNVaO9SuWNQ0dlbLGjDTEsZi8peQFDVmgADUqNc1LKgiJlSiQNMZleX5/NOy47iLf1QNn2e/rNcM/5/lwvt9znrNn33ue85xPpVqtVgMAgCx0KHsCAADsOsIfAEBGhD8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh87VKlUYsSIEW97PSNGjIhKpfL2J9RKK1asiEqlEhMnTtzlYwO7n4kTJ0alUokVK1aUPZU24/rrr49KpRLXX3/921pPbW1tVCqVWLBgQSHz4q0R/tqZZcuWxVlnnRUHHXRQdO3aNXr06BH9+vWLb3zjG/HCCy+UPT2Ane7iiy+OSqUSlUolnnrqqbKnA22O8NdOVKvVuOCCC+LDH/5w3HjjjXHYYYfFV77ylTj77LOjW7ducfnll8f73ve+uOWWW1q97ieeeCJuuOGGtz3HG264IZ544om3vR6A7alWq3Hdddc1fspw7bXXljyj9mHcuHHxxBNPxLhx48qeCgWoKXsCFKOuri4uu+yy6Nu3b9x5551xxBFHNHn91ltvjc9+9rNx6qmnxn333RejRo1KXvdhhx1WyBzf+973FrIegO257777Yvny5XHOOefEHXfcEbNmzYqLL744OnfuXPbUdms9e/aMnj17lj0NCuLKXzuwfPny+N73vhedOnWKOXPmNAt+ERHjx4+P6dOnR0NDQ3zpS1+KrVu3RkTT+zjuuuuuGDZsWPTo0aPJvXnbu+fvxRdfjLPOOit69+4dXbt2jaOPPjpmzZoVCxYsiEqlErW1tU3qW7rn7/W1Dz/8cIwdOzbe+c53Rrdu3WLYsGGxePHiZuOuXr066urqYvDgwbHvvvtG586dY//994/TTjstHn/88bewB4H24mc/+1lEREyaNCnOOOOMeOmll+L2229vsXbevHkxdOjQ6N69e+y1117xqU99qsVPJ5YuXRqVSiU+/elPb3fcQw89NLp06RKvvPJKk+X33ntvHH/88bHPPvtEly5d4uCDD45vfOMbsW7dumbr6Nu3b/Tt2zfWr18fkydPjgMOOCA6derUeC5dv359TJ06NY444ojYc8894x3veEf07ds3TjrppPjjH//YZF3XX399jB8/vsktQIMHD97upzjbzs8bN26M7373u3HooYdG586dG++V3t49f/X19fGFL3whDj/88OjRo0d07do1jjjiiJgyZUq89tpr291flMuVv3Zg5syZsWXLljjppJOiX79+262bNGlS1NXVxdNPPx0LFy6MkSNHNr528803x9y5c+P444+Pc889N5YvX/6mY/7rX/+KQYMGxYoVK2LYsGExaNCg+Mc//hHnnXdefPzjH2/1Nixbtiwuu+yyGDhwYEyaNClWrlwZt956a4wePTr+/Oc/xwc+8IHG2kWLFsW0adNi5MiRMX78+OjevXv87W9/i1tuuSXmzJkTixcvjqOPPrrVcwB2b//85z9jzpw58YEPfCCOPfbY6Nq1a0yfPj1mzJgRJ598cpPaW265JU455ZTo3LlznHLKKbHffvvF73//+xg4cGAcddRRTWoHDhwY73vf++LOO++Ml19+Ofbee+8mry9ZsiSeeeaZGD9+fOy1116Ny+vq6mLKlCmx9957x9ixY6N3797x6KOPxuWXXx533313LFmypNnVtI0bN8aoUaNi7dq1MWbMmMaAV61W47jjjosHHnggBg4cGOecc07U1NTEqlWrYsGCBbF06dLo379/43q+9KUvxeGHHx7Dhg2L/fbbL9asWRN33XVXTJgwIZ588sm45JJLWtyH48ePj2XLlsUnPvGJ+NSnPhV9+vR5031+6aWXxpNPPhmDBg2KsWPHxmuvvRaLFy+Ourq6qK+vj/nz50dNjajR5lTZ7Y0cObIaEdUZM2bssPa0006rRkT1oosuqlar1erMmTOrEVGtVCrVe+65p8X/ExHV4cOHN1n2+c9/vhoR1W9+85tNlj/88MPVzp07VyOiOmXKlCavDR8+vPrGQ66+vr4aEdWIqF5//fVNXrvmmmuqEVE999xzmyz/5z//Wf33v//dbJ5//OMfq926dauOGTOmyfLly5dXI6I6YcKEFrcPaB++//3vVyOiOm3atMZlxxxzTLVSqVSfffbZxmWvvvpqda+99qrW1NRU//CHPzRZx1e/+tXGc9Ly5csbl1988cXViKj+6Ec/ajbuF7/4xWpEVOfMmdO4bP78+dWIqA4ePLi6bt26JvXbzruTJ09usvyAAw6oRkR19OjR1Q0bNjR57ZFHHqlGRPXEE09sNn5DQ0P1lVdeabLsmWeeaVb33//+tzpixIhqTU1NddWqVU1e23Z+7tevX/Wll15q9n+3zXnmzJlNlj/77LPVrVu3Nqu/8MILqxFRnT17dpPlU6ZMqUZEtb6+vtn/YdfxsW878I9//CMiIt7znvfssHZbzerVq5ss/+QnPxnHHXdc0nibNm2K2bNnR8+ePeP//u//mrx21FFHxZlnnpm0ntcbMmRITJgwocmyz3/+81FTUxN/+MMfmizv3bt37Lnnns3W8aEPfShGjRoVCxYsiM2bN7d6DsDuq1qtxrXXXhsdO3aMz33uc43LzzrrrMbXtrnjjjvilVdeidNPPz0GDBjQZD21tbUt3tv2uc99Ljp06BCzZs1qsnzjxo3xq1/9Knr37h2f+MQnGpdfddVVERExY8aMZuubOHFiHH300fHLX/6yxW25/PLLo3v37k2Wbbtlplu3bs3qO3ToEL169Wqy7OCDD25W16VLl/jyl78cW7Zsifnz57c4dl1dXeyzzz4tvtaSgw46qMVHeJ1//vkR8b97MGl7XIttB6rVakRE0jP0ttW8sfYjH/lI8nhPPfVUvPbaazFgwIAWQ9iQIUNa/Q27N56AIyI6deoUffr0ibVr1zZ77a677oprrrkmli1bFmvWrIktW7Y0eX3NmjWx3377tWoOwO5r/vz58eyzz8Zxxx0X+++/f+Py008/Pb7+9a/HzJkzo66uLmpqauJPf/pTREQMHz682Xp69uwZRx99dCxcuLDJ8ve85z0xatSomDdvXvz1r3+Nww8/PCIi5syZE+vWrYuvfe1rTT7eXLp0aXTq1CluuummFue7adOmeOmll5p9jNylS5dmHztHRBx++OFxzDHHxOzZs2PVqlXxyU9+MgYPHhwDBgxo8cssK1eujEsvvTTuv//+WLlyZbP777b36K/W/C6IiPjPf/4TP/zhD+PXv/51PP300/Hqq682/k56s3Eol/DXDuy3337x5JNPxqpVq3ZY+/e//73x/7zevvvumzze+vXrIyK2ey/Iju4Racn2vkVWU1MTDQ0NTZZdddVVMXny5OjVq1d87GMfi/e+973RrVu3qFQqcfvtt8cjjzwSGzdubPUcgN3XjBkzIiKaPcx97733jhNOOCFuvfXW+M1vfhPjxo3b4Tlse+fDCRMmxLx582LWrFlx6aWXRkQ0Xgl84ycXL7/8cmzZsiWmTp36pvPesGFDk/DXp0+fFv+Q79ixY9x///1RV1cXt9xyS3zzm9+MiIgePXrExIkT45JLLmm8Wvjcc8/FscceG2vXro2hQ4fGxz/+8ejZs2d07NgxVqxYEbNmzdruObI1vws2b94co0aNioceeiiOPPLIOOWUU+Jd73pXdOrUKSIipk6d6lzcRgl/7cCQIUOivr4+5s2bF5MmTdpuXUNDQ9TX10dExODBg5u81prOGz169IiI/91c3ZLtLS/Cli1bYsqUKbHvvvvGn/70p2YhdunSpTttbKBtev03ek899dQ49dRTW6ybMWNGjBs3rvGPze2dq7bdSvNGn/70p+O8886LG2+8MS655JJ4+eWX4957742jjjqq2dW6nj17xtatW5t9+3dH3uxc3KtXr5g+fXpMnz49nnnmmVi4cGH89Kc/jauuuirWrVvXGESvvPLKePnll2PmzJnNwvDs2bObfXSdOv4b3XHHHfHQQw/FhAkTmn0L+MUXX9xh8KU87vlrByZOnBgdO3aM2267Lf76179ut+7nP/95rF69Ot7//ve3+HFHqsMOOyy6du0ajz76aLz66qvNXv/973//lte9I2vWrIl169bFoEGDmgW/DRs2NH6cA+Rj1qxZsWnTpujfv3+cffbZLf7bZ5994r777ovnn38+PvShD0VENPtoN+J/n2w8/PDDLY7TrVu3+MxnPhOrV6+OefPmxS9+8YvYsmVLs6t+EREf/ehHY+3atTvt8VOHHHJInH322bFw4cJ4xzveEb/+9a8bX3vmmWci4n/f3H2jlrb5rdpV41A84a8dOOigg+Jb3/pWbN68OU444YQWA+Dtt98ekydPjo4dO8bVV18dHTq89bd+26MR1q9fH9/73veavPbII48U0g1ke3r37h3dunWLZcuWxYYNGxqXb968OSZPnhxr1qzZaWMDbdO2e4yvvvrquPbaa1v8N2nSpNi6dWtcd911ceKJJ0avXr3il7/8ZSxbtqzJumpraxs/Fm7JtitpN9xwQ9xwww1RU1MTZ5xxRrO6r33taxERcc455zT7gl3E/+6Ve+CBB5K3cfny5S0GybVr18bGjRtjjz32aFzWt2/fiIjGT3q2uffeewvteLK9cZ577rm44IILChuH4vnYt52ora2N//znP3HllVfGUUcdFWPGjIkjjjgiNm/eHEuWLIkHH3wwunbtGrNnz25Vd4/tmTZtWsyfPz8uu+yyePDBB2PQoEHx4osvxk033RTHH3983H777W8rYG5Phw4d4itf+UpMmzYt+vXrFyeeeGJs2rQp6uvr45VXXomRI0c2OxEB7deCBQviqaeein79+sWxxx673bpJkybFpZdeGj//+c9jypQpMWPGjDjllFNi6NChTZ7z99hjj8WwYcNi0aJFLa5n6NChceCBB8bNN9/c+Ad37969m9WNHj06pk2bFhdeeGEceuihcfzxx8eBBx4YGzZsiOeffz4WLlwYQ4YMiblz5yZt5yOPPBLjxo2L/v37x5FHHhn7779/vPTSS3HHHXfE5s2bm4St8847L2bOnBknn3xyjB8/Pt797nfHY489FnPnzo2TTz45fvWrXyWNuSMnnHBCHHLIITF9+vR47LHH4phjjomVK1fGnXfeGWPHjo2VK1cWMg7Fc+WvnejQoUNcccUV8eCDD8bpp58ejz/+eFx11VUxY8aM2LBhQ5x//vnx9NNPx0knnVTIeH369IklS5bEmWeeGY8//nhMnz49/vznP8fVV1/d+FfwzmoFdNFFF8UVV1wRXbt2jZ/+9Kdx2223xYABA+Khhx7SQg4y8/qOHm/m4IMPjhEjRsQLL7wQd999d3zmM5+JuXPnRv/+/eOmm26Ka665Jvbaa69YunRpHHjggdtdT6VSiTPPPLPxcVItfeS7zQUXXBCLFi2KsWPHxuLFi+MHP/hB3HzzzfHCCy/EF77whWafnLyZAQMGxIUXXhhdunSJuXPnxhVXXBH33HNP9O/fP+6+++7GR6tERHzwgx+M+vr6GDRoUNx9993xk5/8JP7973/HbbfdFueee27ymDvSvXv3mD9/fpPfOY8++mh85zvfiRtvvLGwcShepfr672RDAb797W/HJZdcEnPnzo0xY8aUPR0A4HWEP96y1atXN3meVkTEX/7ylxg0aFB06dIl/v73vze5DwUAKJ97/njLBgwYEIccckgceeSRjf1177rrrti6dWv87Gc/E/wAoA1y5Y+3rK6uLubMmRMrVqyI9evXR8+ePWPgwIHx9a9//W09SgYA2HmEPwCAjPi2LwBARoQ/AICMCH8AABlJ/rZva5o9A7wV7f0W5LZ+Hk3typP6Pu2M7d26dWvh68xJ6nvcnvZz0ducelwXfT5LHTdlO1z5AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMhIpZr4COq2/mR6YPenw0f7UlYnhIg8O1mUwX5ue1J+nlz5AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMhITdkTAMhFmR0vUqTOL7WuzK4OqWO3lw4VRR9bqftld+jKU3RnnaL3YRnHlit/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICMCH8AABnR4QNgFymrM0DR3R/aU1eHsvZhWV0ncux2UdbxWvS4Rb53rvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZESHD4A2pujOAGV1OCizq0Oq1DmmKrrTRlkdSFIV/R6nrq8127E7dKRJUeR2uPIHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBEdPgDamLK6NRQ97s7orFDWvim6c8cFF1yQVDdt2rSkutra2qS6VatWJdU9//zzSXW//e1vk+pS6fby9utSuPIHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMiI8AcAkJFKNfER1EU+WRqgJTujI0RbUvST/IveX6kdDsrswlDWvklV9O/KU089NanugAMOSKrbsGFDUt2Pf/zjpLqyjund4VxRdMeQ1J+7lH3jyh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICPCHwBARmrKnsCuVltbW9rYU6ZMKW3snEydOjWprsxjAd5MaveCsromfPe7302qS+1w0JpuDaljF915omip3RqK7hKRap999kmqSz2PFn1M7wxl/TyV0a3ElT8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZqVQT+4qU2XKlSGW0UaFtGjlyZFLdggULdu5EaNTefz7by3m0oaEhqS61NVlqq7Odsc6y1le0stqnDR06NKlu8eLFhY5b5rliZxzXRUrZN678AQBkRPgDAMiI8AcAkBHhDwAgI8IfAEBGhD8AgIwIfwAAGRH+AAAyIvwBAGSkpuwJ7GqpXR1GjBiRVNea7g86RbQsdV/X19eXMq73jV2t6C4MRXd/SD2Pjh49OqmuNT/bixYtSqorq8NHWYYMGZJU97vf/S6pLnV7U8+jS5YsKXTc1ij6PW4PnYhc+QMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDISHYdPlK7NejqAJSlrA4CqeOmdtlIrStT0d1PUtdXdNeJjh07JtWlKrrLTOp2FL2fWzN2qrLe4yK58gcAkBHhDwAgI8IfAEBGhD8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQkew6fND2jBgxouwpACVK7YQQkd4NoTXrLGPcojtejB49OqkuVeq4nTp1KnR9RXdcac06U6WOXfQxU+R2uPIHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBEdPgB2kaK7HJSl6C4WqXWtkboPi+4o0adPn6S6L3/5y0l1e+65Z1LdunXrkuqKVvR+Lnrc1iij00bEzjn+d8SVPwCAjAh/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICM6PABsIsU3RmgrI4hZXQkaK2yuqRMnz49qa5Xr15JdamdO9avX59UV/R719DQUOj6iu4e0xq7w3FdFFf+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADKiwwdAG5Pa5SBVWd0uUrVme1O7MKR2P0mtS9WzZ8+kus6dOyfVPffcc0l1mzdvTqor2u5wrBY9x1RtuWOIK38AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBHhDwAgI8IfAEBGhD8AgIwIfwAAGdHhA6CNSe0MUFbngqK1phNCakeOojtFfPazn02q+8tf/lLouBdeeGFS3Xe+852kuk2bNiXV7bHHHkl1RXexSH3fWtOZpeiuMGV1zCmyG037OHMAAJBE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZKRSTXxUdZFPlobXq62tTaqbMmVKoeM6ptuesp6cv6uUdcy19c4F7Ulq15Wiu7ikvncNDQ1JdanHTNHHdHs6Vot+74qsc+UPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICM1ZU8Ahg8fXvYUoE1J7QyQKrWbxO6gre+bstZ31llnJdW19c5GqV0sWnMcFP2epO7DssZN4cofAEBGhD8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEYq1cTHabf1p4Kz+0p9onvRHNNtT1nHwq5S1jGXOm7R80vtcNCabg2px0hqXeo2F31sFt2pZMuWLYWuL3W/DB06NKluyZIlSXW7wzGTquhjK3WbGxoadryupDUBANAuCH8AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBHhDwAgI8IfAEBGhD8AgIzUlD0BKNrUqVPLngK0qKxuEqlSuyukbkdqR4LUcVsjt31ddJeI6667LqnuscceS6obM2ZMUt29996bVNea962s7ixtuWORK38AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBHhDwAgI8IfAEBGhD8AgIwIfwAAGdHhA2AXactP/G+N1O04//zzk+quuOKKtzOdFhXdGSN1falSxy26c0eqhoaGpLqBAwcm1e2xxx5vZzrNlPmz1B5+jl35AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICMCH8AABkR/gAAMiL8AQBkRPgDAMiIDh8Au0hZXSeKXl9q3bx585Lqzj333KS6iIju3bsn1aV2DSl6Hxat6M4dqf71r38l1a1duzap7oEHHkiqS93e1uyXrVu3FrrO1PUVrchj0JU/AICMCH8AABkR/gAAMiL8AQBkRPgDAMiI8AcAkBHhDwAgI8IfAEBGhD8AgIwIfwAAGdHejZ1mxIgRZU8B2pSy2qyljlt0C7OHH3640Lr2JLWV2JAhQ3byTN6e1LZtRWtNi7XUn5Oy2raV0VbOlT8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjOjwwU5TVoeP4cOHlzIuFKXoThupUjsNFC21A0NrFN3VpOguDKl1qefRorvHDB48OKmuaDvj2C/rPU5V9PyS1lXYmgAAaPOEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICPCHwBARirVxEdL74wnsENEed0MRo4cmVS3YMGCnTsRGpV1LOwquZ1Hd0bHhLK6MBSt6O0ouqNJ6rGaeh5duHBhUl17Pwe8HanvScox48ofAEBGhD8AgIwIfwAAGRH+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEZqyp4AlGXEiBFJdTp8UJTUrg5Fd2soa9yd0WWjrXfuKEvR711q3fDhw5Pq2tN5tKyfpyK7n7jyBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDIiPAHAJARHT4oXeqT31M7ckydOjWprra2NqkOilJWd4qix03tcJA6bqVSSR47tTZ17KK3JVXR61u0aFFS3bBhw5LqLrrooqS6urq6pLoypR4zZXau2dVc+QMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADIi/AEAZET4AwDISKWa+Ejr1jyBHeCtSH3C/u4q9Tya2nUiVdH7NXV9O6N7RlndRYreh2W9x0VvR+r+K3o/t2Y72ssxkyplXFf+AAAyIvwBAGRE+AMAyIjwBwCQEeEPACAjwh8AQEaEPwCAjAh/AAAZEf4AADKS3OEDAIDdnyt/AAAZEf4AADIi/AEAZET4AwDIiPAHAJAR4Q8AICPCHwBARoQ/AICMCH8AABn5f4OB7nqcaUoBAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 800x800 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8, 8))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.title(\"Original\")\n",
    "plt.axis(\"off\")\n",
    "plt.imshow(img.squeeze(), cmap=\"gray\")\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.title(\"Adversarial\")\n",
    "plt.axis(\"off\")\n",
    "plt.imshow(calc_adv_img(img, noise_dev).squeeze().detach().numpy(), cmap=\"gray\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor(1., grad_fn=<MaxBackward1>),\n",
       " tensor(0., grad_fn=<MinBackward1>),\n",
       " tensor(1.),\n",
       " tensor(0.))"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "calc_adv_img(img, noise_img).max(), calc_adv_img(img, noise_img).min(), img.max(), img.min()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor([[0.0013]], grad_fn=<SigmoidBackward0>),\n",
       " tensor([[0.7984]], grad_fn=<SigmoidBackward0>))"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model(img_dev), adv_model(img_dev, noise_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor(0.0976), tensor(0.1149, grad_fn=<DivBackward0>))"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.sum()/(28*28), calc_adv_img(img, noise_dev).sum()/(28*28)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "ldm",
   "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": "3.11.3"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}