tutorial.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Reptrix Tutorial \n",
    "\n",
    "In this tutorial, we show how to use reptrix to assess representations of a model pretrained using self-supervised learning. \n",
    "\n",
    "Here we use the [STL-10](https://cs.stanford.edu/~acoates/stl10/) dataset, which contains 10 classes of images. \n",
    "\n",
    "We will use the [Barlow Twins](https://arxiv.org/abs/2103.03230), and [DINO](https://arxiv.org/abs/2104.14294) models as examples.\n",
    "\n",
    "To assess the quality of the learned representations, we will use various metrics, including:\n",
    "\n",
    "- [**Alpha**](https://proceedings.neurips.cc/paper_files/paper/2022/hash/70596d70542c51c8d9b4e423f4bf2736-Abstract-Conference.html): This metric computes the eigenvalues of the covariance matrix of the representations and fits a power-law distribution to them. The exponent of the power-law distribution is called the alpha exponent, which measures the heavy-tailedness of the distribution. A lower alpha exponent indicates that the representations are more discriminative.\n",
    "\n",
    "- [**RankMe**](https://proceedings.mlr.press/v202/garrido23a): This metric computes the rank of the covariance matrix of the representations. A higher rank indicates representations of higher capacity.\n",
    "\n",
    "- [**LiDAR**](https://openreview.net/forum?id=f3g5XpL9Kb): This metric computes the rank of the Linear Discriminant Analysis (LDA) matrix constructed using representations of augmented versions of images. A higher rank indicates representations of higher discriminability.\n",
    "\n",
    "\n",
    "We will compute these metrics using the Reptrix library, which provides a convenient interface for representation analysis. Let's dive into the code and explore the evaluation process in detail.\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import Libraries, get device and define feature extraction function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision\n",
    "from tqdm import tqdm\n",
    "from reptrix import alpha, rankme, lidar\n",
    "import reptrix\n",
    "import time"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get the device"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Function to get the features of from the pretrained encoder model using a dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_features(encoder_function, dataloader, transform=None, num_augmentations=10):\n",
    "    # Loop over the dataset and collect the representations\n",
    "    all_features = []\n",
    "\n",
    "    # Loop over the dataset and collect the representations\n",
    "    for i, data in enumerate(tqdm(dataloader, 0)):\n",
    "        inputs, _ = data\n",
    "        # apply multiple augmentations for each image\n",
    "        if transform:\n",
    "            inputs = torch.cat([transform(inputs) for _ in range(num_augmentations)], dim=0)\n",
    "        with torch.no_grad():\n",
    "            features = encoder_function(inputs.to(device))\n",
    "        if transform:\n",
    "            # put the augmentations in an additonal dimension\n",
    "            features = features.reshape(num_augmentations, -1, features.shape[1]).transpose(1,0)\n",
    "        all_features.append(features)\n",
    "            \n",
    "            \n",
    "    # Concatenate all the features\n",
    "    all_features = torch.cat(all_features, dim=0)\n",
    "    return all_features"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Get the STL-10 dataset "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "transform_to_tensor = torchvision.transforms.ToTensor()\n",
    "\n",
    "STL_MEAN = (0.4467, 0.4398, 0.4066)\n",
    "STL_STD = (0.2242, 0.2215, 0.2239)\n",
    "transform_base = torchvision.transforms.Normalize(STL_MEAN, STL_STD)\n",
    "\n",
    "# Define additional SSL transformations for the Lidar metric evaluation\n",
    "transform_ssl = torchvision.transforms.Compose([\n",
    "    torchvision.transforms.RandomHorizontalFlip(p=0.5),\n",
    "    torchvision.transforms.ColorJitter(brightness=0.4, contrast=0.4, saturation=0.4, hue=0.1),\n",
    "    torchvision.transforms.RandomGrayscale(p=0.2),\n",
    "    torchvision.transforms.RandomResizedCrop(\n",
    "        96, scale=(0.8, 1.0), \n",
    "        ratio=(0.75, (4/3)), \n",
    "        interpolation=torchvision.transforms.InterpolationMode.BICUBIC),\n",
    "    torchvision.transforms.Normalize(STL_MEAN, STL_STD)\n",
    "])\n",
    "    \n",
    "dataset_folder = '/network/datasets/stl10.var/stl10_torchvision'\n",
    "# Get the STL10 test dataset to measure the quality of the representations learned by the model\n",
    "# testset = torchvision.datasets.STL10(root='./data', split='test', download=False, transform=transform)\n",
    "testset = torchvision.datasets.STL10(root=dataset_folder, split='test', download=False, transform=transform_to_tensor)\n",
    "\n",
    "# Define a dataloader to load the test dataset\n",
    "testloader = torch.utils.data.DataLoader(testset, batch_size=256, shuffle=False, num_workers=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the pretrained models, extract the features and compute the metrics"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. ResNet50 encoder trained with Barlow Twins\n",
    "\n",
    "In this section, we will evaluate the representations learned by an encoder trained with the Barlow Twins method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using cache found in /home/mila/a/arnab.mondal/.cache/torch/hub/facebookresearch_barlowtwins_main\n",
      "/home/mila/a/arnab.mondal/.conda/envs/reptrix/lib/python3.10/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.\n",
      "  warnings.warn(\n",
      "/home/mila/a/arnab.mondal/.conda/envs/reptrix/lib/python3.10/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=None`.\n",
      "  warnings.warn(msg)\n"
     ]
    }
   ],
   "source": [
    "# Define a resnet encoder using pytorch where you can load the weights from a pre-trained model\n",
    "# We will use encoder that is trained using Barlow Twins\n",
    "encoder = torch.hub.load('facebookresearch/barlowtwins:main', 'resnet50')\n",
    "# Remove the final fully connected layer so that the model outputs the 2048 feature vector\n",
    "encoder = torch.nn.Sequential(*(list(encoder.children())[:-1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 32/32 [00:03<00:00,  9.16it/s]\n"
     ]
    }
   ],
   "source": [
    "# Move the model to the device\n",
    "encoder = encoder.to(device)\n",
    "\n",
    "# Set the model to evaluation mode\n",
    "encoder.eval()\n",
    "\n",
    "all_representations = get_features(encoder, testloader, transform=transform_base, num_augmentations=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([8000, 1, 2048])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_representations.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "start_time = time.time()\n",
    "metric_alpha = alpha.get_alpha(all_representations)\n",
    "alpha_time = time.time()\n",
    "metric_rankme = rankme.get_rankme(all_representations)\n",
    "rankme_time = time.time()\n",
    "alpha_compute_time = alpha_time - start_time\n",
    "rankme_compute_time = rankme_time - alpha_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the eigenspectrum and the powerlaw fit\n",
    "eigenspectrum = reptrix.utils.get_eigenspectrum(all_representations.cpu())\n",
    "\n",
    "# reptrix.utils.plot_eigenspectrum(eigenspectrum)\n",
    "reptrix.alpha.plot_powerlaw(eigenspectrum, metric_alpha)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/32 [00:00<?, ?it/s]/network/scratch/g/ghosharn/conda_envs/ffcv_ssl_fastssl/lib/python3.9/site-packages/torchvision/transforms/functional.py:1603: UserWarning: The default value of the antialias parameter of all the resizing transforms (Resize(), RandomResizedCrop(), etc.) will change from None to True in v0.17, in order to be consistent across the PIL and Tensor backends. To suppress this warning, directly pass antialias=True (recommended, future default), antialias=None (current default, which means False for Tensors and True for PIL), or antialias=False (only works on Tensors - PIL will still use antialiasing). This also applies if you are using the inference transforms from the models weights: update the call to weights.transforms(antialias=True).\n",
      "  warnings.warn(\n",
      "100%|██████████| 32/32 [01:31<00:00,  2.87s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Values of different metrics: Alpha: 0.9461134930695289, Rankme: 364.6830749511719, LiDAR: 102.7093505859375\n",
      "Compute time for different metrics: Alpha: 2.400s, Rankme: 2.364s, LiDAR: 7.929s\n"
     ]
    }
   ],
   "source": [
    "# Get features of all the augmentations for LiDAR metric evaluation\n",
    "num_augmentations = 10\n",
    "all_representations_ssl = get_features(encoder.forward, testloader, transform=transform_ssl, num_augmentations=num_augmentations)\n",
    "\n",
    "num_samples = all_representations_ssl.shape[0] // num_augmentations\n",
    "\n",
    "# Randomly sample num_samples number of samples from the all_representations_ssl\n",
    "ramdom_sample_index = torch.randperm(all_representations_ssl.shape[0])[:num_samples]\n",
    "subsampled_representations_ssl = all_representations_ssl[ramdom_sample_index]\n",
    "\n",
    "# Get the Lidar metric\n",
    "start_time = time.time()\n",
    "lidar_metric = lidar.get_lidar(subsampled_representations_ssl, num_samples, num_augmentations, del_sigma_augs=0.001)\n",
    "lidar_time = time.time()\n",
    "lidar_compute_time = lidar_time - start_time\n",
    "\n",
    "\n",
    "# For lidar metric evaluation, we need to use the SSL transformations over the test dataset\n",
    "\n",
    "print(f'Values of different metrics: Alpha: {metric_alpha[0]}, Rankme: {metric_rankme}, LiDAR: {lidar_metric}')\n",
    "print(f'Compute time for different metrics: Alpha: {alpha_compute_time:.3f}s, Rankme: {rankme_compute_time:.3f}s, LiDAR: {lidar_compute_time:.3f}s')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Vit Encoder trained with DINO"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using cache found in /home/mila/g/ghosharn/.cache/torch/hub/facebookresearch_dino_main\n"
     ]
    }
   ],
   "source": [
    "# Define a vit encoder using pytorch where you can load the weights from a pre-trained model\n",
    "# We will use encoder that is trained using Dino\n",
    "encoder = torch.hub.load('facebookresearch/dino:main', 'dino_vits16')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 32/32 [00:02<00:00, 10.68it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of the representations: torch.Size([8000, 1, 384])\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Move the model to the device\n",
    "encoder = encoder.to(device)\n",
    "\n",
    "# Set the model to evaluation mode\n",
    "encoder.eval()\n",
    "    \n",
    "all_representations = get_features(encoder.forward, testloader, transform=transform_base, num_augmentations=1)\n",
    "\n",
    "print('Shape of the representations:', all_representations.shape) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "start_time = time.time()\n",
    "metric_alpha = alpha.get_alpha(all_representations)\n",
    "alpha_time = time.time()\n",
    "metric_rankme = rankme.get_rankme(all_representations)\n",
    "rankme_time = time.time()\n",
    "alpha_compute_time = alpha_time - start_time\n",
    "rankme_compute_time = rankme_time - alpha_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the eigenspectrum and the powerlaw fit\n",
    "eigenspectrum = reptrix.utils.get_eigenspectrum(all_representations.cpu())\n",
    "\n",
    "# reptrix.utils.plot_eigenspectrum(eigenspectrum)\n",
    "reptrix.alpha.plot_powerlaw(eigenspectrum, metric_alpha)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/32 [00:00<?, ?it/s]/network/scratch/g/ghosharn/conda_envs/ffcv_ssl_fastssl/lib/python3.9/site-packages/torchvision/transforms/functional.py:1603: UserWarning: The default value of the antialias parameter of all the resizing transforms (Resize(), RandomResizedCrop(), etc.) will change from None to True in v0.17, in order to be consistent across the PIL and Tensor backends. To suppress this warning, directly pass antialias=True (recommended, future default), antialias=None (current default, which means False for Tensors and True for PIL), or antialias=False (only works on Tensors - PIL will still use antialiasing). This also applies if you are using the inference transforms from the models weights: update the call to weights.transforms(antialias=True).\n",
      "  warnings.warn(\n",
      "100%|██████████| 32/32 [01:31<00:00,  2.87s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of the representations after SSL transformations: torch.Size([8000, 10, 384])\n",
      "Values of different metrics: Alpha: 0.7854639552258638, Rankme: 156.16427612304688, LiDAR: 15.42029094696045\n",
      "Compute time for different metrics: Alpha: 0.137s, Rankme: 0.091s, LiDAR: 0.162s\n"
     ]
    }
   ],
   "source": [
    "# Get features of all the augmentations for LiDAR metric evaluation\n",
    "num_augmentations = 10\n",
    "all_representations_ssl = get_features(encoder.forward, testloader, transform=transform_ssl, num_augmentations=num_augmentations)\n",
    "\n",
    "print(\"Shape of the representations after SSL transformations:\", all_representations_ssl.shape)\n",
    "\n",
    "num_samples = all_representations_ssl.shape[0] // num_augmentations\n",
    "\n",
    "# Randomly sample num_samples number of samples from the all_representations_ssl\n",
    "ramdom_sample_index = torch.randperm(all_representations_ssl.shape[0])[:num_samples]\n",
    "subsampled_representations_ssl = all_representations_ssl[ramdom_sample_index]\n",
    "\n",
    "# Get the Lidar metric\n",
    "start_time = time.time()\n",
    "lidar_metric = lidar.get_lidar(subsampled_representations_ssl, num_samples, num_augmentations, del_sigma_augs=0.001)\n",
    "lidar_time = time.time()\n",
    "lidar_compute_time = lidar_time - start_time\n",
    "\n",
    "# For lidar metric evaluation, we need to use the SSL transformations over the test dataset\n",
    "\n",
    "print(f'Values of different metrics: Alpha: {metric_alpha[0]}, Rankme: {metric_rankme}, LiDAR: {lidar_metric}')\n",
    "print(f'Compute time for different metrics: Alpha: {alpha_compute_time:.3f}s, Rankme: {rankme_compute_time:.3f}s, LiDAR: {lidar_compute_time:.3f}s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "reptrix",
   "language": "python",
   "name": "ffcv_ssl_fastssl"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}