jupyter-sample/mnist_train.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Simple MNIST convnet\n",
    "Tweaked version of the code from https://keras.io/examples/vision/mnist_convnet/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext tensorboard\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "import datetime, os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "import boto3\n",
    "from botocore.config import Config\n",
    "from botocore.exceptions import ClientError\n",
    "\n",
    "boto_config = Config(\n",
    "    region_name = 'us-east-1',\n",
    "    signature_version = 's3v4',\n",
    ")\n",
    "\n",
    "# Set up S3 client to interact with Filebase \n",
    "s3 = boto3.client('s3',\n",
    "                  endpoint_url='https://s3.filebase.com',\n",
    "                  aws_access_key_id=os.getenv('AWS_ACCESS_KEY_ID'),\n",
    "                  aws_secret_access_key=os.getenv('AWS_SECRET_ACCESS_KEY'),\n",
    "                  config=boto_config)\n",
    "\n",
    "def download_file(bucket, obj, filename):\n",
    "    s3.download_file(bucket, obj, filename)\n",
    "    \n",
    "def upload_file(filename, obj):\n",
    "    # All uploads should go to designated project bucket\n",
    "    root_bucket = os.getenv('ROOT_BUCKET')\n",
    "    project_dir = os.getenv('PROJECT_DIR')\n",
    "    s3.upload_file(filename, root_bucket, project_dir + '/' + obj)\n",
    "    \n",
    "def upload_model(model_path):\n",
    "    root_bucket = os.getenv('ROOT_BUCKET')\n",
    "    for root, dirs, files in os.walk(model_path):\n",
    "        for file in files:\n",
    "            full_path = os.path.join(root, file)\n",
    "            upload_file(full_path, full_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Training Data (from Filebase) and Preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_BUCKET = 'dataset-3b5d3834-72a8-49f5-9e1a-40f5cd8a77c5'\n",
    "DATA_OBJECT = 'MNIST/mnist.npz'\n",
    "download_file(DATA_BUCKET, DATA_OBJECT, 'mnist.npz')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# the data, split between train and test sets\n",
    "with np.load('mnist.npz', allow_pickle=True) as f:\n",
    "    x_train, y_train = f['x_train'], f['y_train']\n",
    "    x_test, y_test = f['x_test'], f['y_test']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train shape: (60000, 28, 28, 1)\n",
      "60000 train samples\n",
      "10000 test samples\n"
     ]
    }
   ],
   "source": [
    "# Model / data parameters\n",
    "num_classes = 10\n",
    "input_shape = (28, 28, 1)\n",
    "\n",
    "# Scale images to the [0, 1] range\n",
    "x_train = x_train.astype(\"float32\") / 255\n",
    "x_test = x_test.astype(\"float32\") / 255\n",
    "# Make sure images have shape (28, 28, 1)\n",
    "x_train = np.expand_dims(x_train, -1)\n",
    "x_test = np.expand_dims(x_test, -1)\n",
    "print(\"x_train shape:\", x_train.shape)\n",
    "print(x_train.shape[0], \"train samples\")\n",
    "print(x_test.shape[0], \"test samples\")\n",
    "\n",
    "# convert class vectors to binary class matrices\n",
    "y_train = tf.keras.utils.to_categorical(y_train, num_classes)\n",
    "y_test = tf.keras.utils.to_categorical(y_test, num_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 25 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot some samples\n",
    "def plot_examples():\n",
    "    fig = plt.figure(figsize=(8, 8))\n",
    "    rows, columns = 5, 5\n",
    "    for i in range(1, columns*rows +1):\n",
    "        ax = fig.add_subplot(rows, columns, i)\n",
    "        ax.set_axis_off()\n",
    "        plt.imshow(x_train[i], cmap=plt.get_cmap('gray'), interpolation='nearest')\n",
    "    plt.show()\n",
    "\n",
    "plot_examples()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-10-18 02:26:39.867347: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE4.1 SSE4.2 AVX AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d (Conv2D)              (None, 26, 26, 32)        320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 10)                16010     \n",
      "=================================================================\n",
      "Total params: 34,826\n",
      "Trainable params: 34,826\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def create_model():\n",
    "    return tf.keras.Sequential([\n",
    "        tf.keras.Input(shape=input_shape),\n",
    "        tf.keras.layers.Conv2D(32, kernel_size=(3, 3), activation=\"relu\"),\n",
    "        tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),\n",
    "        tf.keras.layers.Conv2D(64, kernel_size=(3, 3), activation=\"relu\"),\n",
    "        tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),\n",
    "        tf.keras.layers.Flatten(),\n",
    "        tf.keras.layers.Dropout(0.5),\n",
    "        tf.keras.layers.Dense(num_classes, activation=\"softmax\"),\n",
    "    ])\n",
    "\n",
    "model = create_model()\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Remove any old training logs\n",
    "!rm -rf ./logs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_model(model):\n",
    "    batch_size = 8\n",
    "    epochs = 7\n",
    "    model.compile(loss=\"categorical_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "\n",
    "    logdir = os.path.join(\"./logs\", datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\"))\n",
    "    tensorboard_callback = tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)\n",
    "\n",
    "    model.fit(x_train,\n",
    "              y_train,\n",
    "              batch_size=batch_size,\n",
    "              epochs=epochs,\n",
    "              validation_split=0.3,\n",
    "              callbacks=[tensorboard_callback])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture\n",
    "%tensorboard --logdir ./logs --reload_interval 3 --reload_multifile True --host 0.0.0.0 --port 6006"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-10-18 02:27:22.183108: I tensorflow/core/profiler/lib/profiler_session.cc:131] Profiler session initializing.\n",
      "2021-10-18 02:27:22.183265: I tensorflow/core/profiler/lib/profiler_session.cc:146] Profiler session started.\n",
      "2021-10-18 02:27:22.203391: I tensorflow/core/profiler/lib/profiler_session.cc:164] Profiler session tear down.\n",
      "2021-10-18 02:27:23.358833: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 131712000 exceeds 10% of free system memory.\n",
      "2021-10-18 02:27:26.239044: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/7\n",
      "   1/5250 [..............................] - ETA: 2:04:43 - loss: 2.2707 - accuracy: 0.2500"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-10-18 02:27:27.965402: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 17981568 exceeds 10% of free system memory.\n",
      "2021-10-18 02:27:27.967583: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 17981568 exceeds 10% of free system memory.\n",
      "2021-10-18 02:27:27.988033: I tensorflow/core/profiler/lib/profiler_session.cc:131] Profiler session initializing.\n",
      "2021-10-18 02:27:27.988505: I tensorflow/core/profiler/lib/profiler_session.cc:146] Profiler session started.\n",
      "2021-10-18 02:27:28.001588: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 17981568 exceeds 10% of free system memory.\n",
      "2021-10-18 02:27:28.002649: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 17981568 exceeds 10% of free system memory.\n",
      "2021-10-18 02:27:28.021664: I tensorflow/core/profiler/lib/profiler_session.cc:66] Profiler session collecting data.\n",
      "2021-10-18 02:27:28.045961: I tensorflow/core/profiler/lib/profiler_session.cc:164] Profiler session tear down.\n",
      "2021-10-18 02:27:28.100846: I tensorflow/core/profiler/rpc/client/save_profile.cc:136] Creating directory: ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28\n",
      "\n",
      "2021-10-18 02:27:28.136562: I tensorflow/core/profiler/rpc/client/save_profile.cc:142] Dumped gzipped tool data for trace.json.gz to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.trace.json.gz\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  12/5250 [..............................] - ETA: 3:29 - loss: 2.2958 - accuracy: 0.1875"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-10-18 02:27:28.242887: I tensorflow/core/profiler/rpc/client/save_profile.cc:136] Creating directory: ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28\n",
      "\n",
      "2021-10-18 02:27:28.243272: I tensorflow/core/profiler/rpc/client/save_profile.cc:142] Dumped gzipped tool data for memory_profile.json.gz to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.memory_profile.json.gz\n",
      "2021-10-18 02:27:28.245169: I tensorflow/core/profiler/rpc/client/capture_profile.cc:251] Creating directory: ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28\n",
      "Dumped tool data for xplane.pb to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.xplane.pb\n",
      "Dumped tool data for overview_page.pb to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.overview_page.pb\n",
      "Dumped tool data for input_pipeline.pb to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.input_pipeline.pb\n",
      "Dumped tool data for tensorflow_stats.pb to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.tensorflow_stats.pb\n",
      "Dumped tool data for kernel_stats.pb to ./logs/20211018-022722/train/plugins/profile/2021_10_18_02_27_28/dac4f59cd70c.kernel_stats.pb\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5250/5250 [==============================] - 97s 18ms/step - loss: 0.1858 - accuracy: 0.9414 - val_loss: 0.0690 - val_accuracy: 0.9798\n",
      "Epoch 2/7\n",
      "5250/5250 [==============================] - 81s 16ms/step - loss: 0.0800 - accuracy: 0.9756 - val_loss: 0.0527 - val_accuracy: 0.9847\n",
      "Epoch 3/7\n",
      "5250/5250 [==============================] - 75s 14ms/step - loss: 0.0617 - accuracy: 0.9801 - val_loss: 0.0504 - val_accuracy: 0.9846\n",
      "Epoch 4/7\n",
      "5250/5250 [==============================] - 66s 13ms/step - loss: 0.0546 - accuracy: 0.9828 - val_loss: 0.0380 - val_accuracy: 0.9879\n",
      "Epoch 5/7\n",
      "5250/5250 [==============================] - 59s 11ms/step - loss: 0.0487 - accuracy: 0.9848 - val_loss: 0.0399 - val_accuracy: 0.9881\n",
      "Epoch 6/7\n",
      "5250/5250 [==============================] - 62s 12ms/step - loss: 0.0458 - accuracy: 0.9852 - val_loss: 0.0430 - val_accuracy: 0.9878\n",
      "Epoch 7/7\n",
      "5250/5250 [==============================] - 65s 12ms/step - loss: 0.0409 - accuracy: 0.9870 - val_loss: 0.0382 - val_accuracy: 0.9884\n"
     ]
    }
   ],
   "source": [
    "train_model(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss: 0.03167686238884926\n",
      "Test accuracy: 0.9889000058174133\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate(x_test, y_test, verbose=0)\n",
    "print(\"Test loss:\", score[0])\n",
    "print(\"Test accuracy:\", score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-10-18 02:38:06.264785: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: models/mnist/1/assets\n"
     ]
    }
   ],
   "source": [
    "saved_model_path = model.save(\"models/mnist/1\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Persist Model (and this notebook) to Filebase"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Upload model to Filebase\n",
    "upload_model('models/mnist')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Backup this notebook to Filebase (make sure to save first before running this cell)\n",
    "upload_file('mnist_train.ipynb', 'mnist_train.ipynb')"
   ]
  },
  {
   "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}