transfer_learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Convolutional Neural Networks\n",
    "\n",
    "## Project: Write an Algorithm for Landmark Classification\n",
    "\n",
    "\n",
    "### Transfer learning\n",
    "\n",
    "In the previous notebook we have trained our own CNN and we got a certain performance. Let's see how hard it is to match that performance with transfer learning.\n",
    "\n",
    "---\n",
    "## <img src=\"static_images/icons/noun-advance-2109145.png\" alt=\">\" style=\"width:50px\"/> Step 0: Setting up\n",
    "\n",
    "The following cells make sure that your environment is setup correctly and check that your GPU is available and ready to go. You have to execute them every time you restart your notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Install requirements\n",
    "!pip install -r requirements.txt | grep -v \"already satisfied\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GPU available\n",
      "Dataset already downloaded. If you need to re-download, please delete the directory landmark_images\n",
      "Reusing cached mean and std\n"
     ]
    }
   ],
   "source": [
    "from src.helpers import setup_env\n",
    "\n",
    "# If running locally, this will download dataset (make sure you have at \n",
    "# least 2 Gb of space on your hard drive)\n",
    "setup_env()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "## <img src=\"static_images/icons/noun-advance-2109145.png\" alt=\">\" style=\"width:50px\"/> Step 1: Create transfer learning architecture\n",
    "\n",
    "Open the file `src/transfer.py` and complete the `get_model_transfer_learning` function. When you are done, execute this test:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m============================= test session starts ==============================\u001b[0m\n",
      "platform linux -- Python 3.7.6, pytest-7.1.1, pluggy-1.0.0 -- /home/xye/anaconda3/envs/udacity/bin/python\n",
      "cachedir: .pytest_cache\n",
      "rootdir: /home/xye/landmark\n",
      "plugins: anyio-3.6.2\n",
      "collected 1 item                                                               \u001b[0m\n",
      "\n",
      "src/transfer.py::test_get_model_transfer_learning \u001b[32mPASSED\u001b[0m\u001b[32m                 [100%]\u001b[0m\n",
      "\n",
      "\u001b[33m=============================== warnings summary ===============================\u001b[0m\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "  /home/xye/anaconda3/envs/udacity/lib/python3.7/site-packages/matplotlib/__init__.py:152: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.\n",
      "    if LooseVersion(module.__version__) < minver:\n",
      "\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "  /home/xye/anaconda3/envs/udacity/lib/python3.7/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.\n",
      "    other = LooseVersion(other)\n",
      "\n",
      "src/transfer.py::test_get_model_transfer_learning\n",
      "  /home/xye/anaconda3/envs/udacity/lib/python3.7/site-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at  /pytorch/c10/core/TensorImpl.h:1156.)\n",
      "    return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)\n",
      "\n",
      "-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html\n",
      "\u001b[33m======================== \u001b[32m1 passed\u001b[0m, \u001b[33m\u001b[1m11 warnings\u001b[0m\u001b[33m in 1.47s\u001b[0m\u001b[33m ========================\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "!pytest -vv src/transfer.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "## <img src=\"static_images/icons/noun-advance-2109145.png\" alt=\">\" style=\"width:50px\"/> Step 2: Train, validation and test\n",
    "\n",
    "Let's train our transfer learning model! Let's start defining the hyperparameters:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "batch_size = 64  # size of the minibatch for stochastic gradient descent (or Adam)\n",
    "valid_size = 0.2  # fraction of the training data to reserve for validation\n",
    "num_epochs = 25  # number of epochs for training\n",
    "num_classes = 50  # number of classes. Do not change this\n",
    "learning_rate = 0.003  # Learning rate for SGD (or Adam)\n",
    "opt = 'adam'      # optimizer. 'sgd' or 'adam'\n",
    "weight_decay = 0.0 # regularization. Increase this to combat overfitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1200x800 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from src.data import get_data_loaders\n",
    "from src.optimization import get_optimizer, get_loss\n",
    "from src.train import optimize\n",
    "from src.transfer import get_model_transfer_learning\n",
    "\n",
    "# Get a model using get_model_transfer_learning. Use one of the names reported here:\n",
    "# https://pytorch.org/vision/0.10/models.html\n",
    "# For example, if you want to load ResNet 18, use \"resnet18\"\n",
    "# NOTE: use the hyperparameters defined in the previous cell, do NOT copy/paste the\n",
    "# values\n",
    "model_transfer = get_model_transfer_learning(model_name=\"resnet18\", n_classes=num_classes) # YOUR CODE HERE\n",
    "\n",
    "# train the model\n",
    "data_loaders = get_data_loaders(batch_size=batch_size)\n",
    "optimizer = get_optimizer(\n",
    "    model_transfer,\n",
    "    learning_rate=learning_rate,\n",
    "    optimizer=opt,\n",
    "    weight_decay=weight_decay,\n",
    ")\n",
    "loss = get_loss()\n",
    "\n",
    "optimize(\n",
    "    data_loaders,\n",
    "    model_transfer,\n",
    "    optimizer,\n",
    "    loss,\n",
    "    n_epochs=num_epochs,\n",
    "    save_path=\"checkpoints/model_transfer.pt\",\n",
    "    interactive_tracking=True\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"static_images/icons/noun-question-mark-869751.png\" alt=\"?\" style=\"width:25px\"/> __Question:__ Outline the steps you took to get to your final CNN architecture and your reasoning at each step.  Describe why you think the architecture is suitable for the current problem."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"static_images/icons/noun-answer-3361020.png\" alt=\">\" style=\"width:25px\"/>  __Answer:__ I decided to use ResNet18 for the base of my model, since it performs fairly well on ImageNet and is not too large of a model. Also, since ResNet18 was trained for the ImageNet task, it is a good model to use for this landmark classificaiton task, since both ImageNet and this landmark task use images of natural scenes."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now play with the hyperparameters and see which performance you can get on the validation set. You should get at least 60% for a passing grade, but a good model choice and a good training strategy could get you up to 80% or so. Let's see how close you can get!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "## <img src=\"static_images/icons/noun-advance-2109145.png\" alt=\">\" style=\"width:50px\"/> Step 3: Test the Model\n",
    "\n",
    "Try out your model on the test dataset of landmark images. Use the code cell below to calculate and print the test loss and accuracy.  Ensure that your test accuracy is greater than 60% and matches more or less what you got on the validation set (otherwise you're overfitting!)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Testing: 100%|██████████████████████████████| 1250/1250 [00:17<00:00, 72.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Loss: 1.073693\n",
      "\n",
      "\n",
      "Test Accuracy: 74% (928/1250)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1.0736930375644518"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch\n",
    "from src.train import one_epoch_test\n",
    "from src.transfer import get_model_transfer_learning\n",
    "\n",
    "model_transfer = get_model_transfer_learning(\"resnet18\", n_classes=num_classes)\n",
    "# Load saved weights\n",
    "model_transfer.load_state_dict(torch.load('checkpoints/model_transfer.pt'))\n",
    "\n",
    "one_epoch_test(data_loaders['test'], model_transfer, loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "## <img src=\"static_images/icons/noun-advance-2109145.png\" alt=\">\" style=\"width:50px\"/> Step 4: Export using torchscript\n",
    "\n",
    "Now, just like we did with our original model, we export the best fit model using torchscript so that it can be used in our application:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'data_loaders' 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/tmp/ipykernel_41603/1716410941.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;31m# First let's get the class names from our data loaders\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m \u001b[0mclass_names\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdata_loaders\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"train\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclasses\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      6\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m \u001b[0;31m# Then let's move the model_transfer to the CPU\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 'data_loaders' is not defined"
     ]
    }
   ],
   "source": [
    "from src.predictor import Predictor\n",
    "from src.helpers import compute_mean_and_std\n",
    "\n",
    "# First let's get the class names from our data loaders\n",
    "class_names = data_loaders[\"train\"].dataset.classes\n",
    "\n",
    "# Then let's move the model_transfer to the CPU\n",
    "# (we don't need GPU for inference)\n",
    "model_transfer = model_transfer.cpu()\n",
    "# Let's make sure we use the right weights by loading the\n",
    "# best weights we have found during training\n",
    "# NOTE: remember to use map_location='cpu' so the weights\n",
    "# are loaded on the CPU (and not the GPU)\n",
    "model_transfer.load_state_dict(\n",
    "    torch.load(\"checkpoints/model_transfer.pt\", map_location=\"cpu\")\n",
    ")\n",
    "\n",
    "# Let's wrap our model using the predictor class\n",
    "mean, std = compute_mean_and_std()\n",
    "predictor_export = Predictor(model_transfer, class_names, mean, std)\n",
    "\n",
    "# Export using torch.jit.script\n",
    "scripted_predictor = torch.jit.script(predictor_export)\n",
    "torch.jit.save(scripted_predictor, \"checkpoints/transfer_exported.pt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from src.predictor import predictor_test\n",
    "from src.helpers import plot_confusion_matrix\n",
    "\n",
    "model_reloaded = torch.jit.load(\"checkpoints/transfer_exported.pt\")\n",
    "\n",
    "pred, truth = predictor_test(data_loaders['test'], model_reloaded)\n",
    "\n",
    "plot_confusion_matrix(pred, truth)"
   ]
  },
  {
   "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": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {},
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}