DistTrain_rpv.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Distributed training of ATLAS RPV CNN Classifier\n",
    "\n",
    "In this notebook, we extend the Train_rpv example to train distributed across nodes\n",
    "using ipyparallel and Horovod."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# System imports\n",
    "from __future__ import print_function\n",
    "from __future__ import division\n",
    "from __future__ import absolute_import\n",
    "\n",
    "# External imports\n",
    "import ipyparallel as ipp\n",
    "\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib notebook"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Connect to ipyparallel cluster"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)\n",
      "          24185321 interacti       sh sfarrell  R      12:47      8 nid000[63,72-75,77-79]\n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "squeue -u sfarrell"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Worker IDs: [0, 1, 2, 3, 4, 5, 6, 7]\n"
     ]
    }
   ],
   "source": [
    "# Cluster ID taken from job ID above\n",
    "job_id = 24185321\n",
    "cluster_id = 'cori_{}'.format(job_id)\n",
    "\n",
    "# Use default profile\n",
    "c = ipp.Client(timeout=60, cluster_id=cluster_id)\n",
    "print('Worker IDs:', c.ids)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup the workers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%px\n",
    "\n",
    "from __future__ import print_function\n",
    "from __future__ import division\n",
    "\n",
    "import os\n",
    "import socket\n",
    "\n",
    "import keras\n",
    "import horovod.keras as hvd\n",
    "\n",
    "from rpv import load_dataset, build_model, train_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[stdout:0] MPI rank 0, local rank 0, host nid00063\n",
      "[stdout:1] MPI rank 5, local rank 0, host nid00077\n",
      "[stdout:2] MPI rank 4, local rank 0, host nid00075\n",
      "[stdout:3] MPI rank 6, local rank 0, host nid00078\n",
      "[stdout:4] MPI rank 2, local rank 0, host nid00073\n",
      "[stdout:5] MPI rank 3, local rank 0, host nid00074\n",
      "[stdout:6] MPI rank 1, local rank 0, host nid00072\n",
      "[stdout:7] MPI rank 7, local rank 0, host nid00079\n"
     ]
    }
   ],
   "source": [
    "%%px\n",
    "\n",
    "# Initialize horovod\n",
    "hvd.init()\n",
    "print('MPI rank %i, local rank %i, host %s' %\n",
    "      (hvd.rank(), hvd.local_rank(), socket.gethostname()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[stdout:0] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:1] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:2] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:3] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:4] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:5] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:6] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n",
      "[stdout:7] \n",
      "train shape: (64000, 64, 64, 1)\n",
      "valid shape: (32000, 64, 64, 1)\n",
      "test shape:  (32000, 64, 64, 1)\n"
     ]
    }
   ],
   "source": [
    "%%px\n",
    "\n",
    "# Data config\n",
    "n_train = 64000 #412416\n",
    "n_valid = 32000 #137471\n",
    "n_test = 32000 #137471\n",
    "input_dir = '/global/cscratch1/sd/sfarrell/atlas-rpv-images'\n",
    "\n",
    "# Load the data files\n",
    "((train_input, train_labels, train_weights),\n",
    " (valid_input, valid_labels, valid_weights),\n",
    " (test_input, test_labels, test_weights)) = load_dataset(input_dir, n_train, n_valid, n_test)\n",
    "print('train shape:', train_input.shape)\n",
    "print('valid shape:', valid_input.shape)\n",
    "print('test shape: ', test_input.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build and train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[stdout:0] \n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         (None, 64, 64, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 64, 64, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 32, 32, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 32, 32, 32)        4640      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_5 (MaxPooling2 (None, 16, 16, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 16, 16, 64)        18496     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_6 (MaxPooling2 (None, 8, 8, 64)          0         \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 8, 8, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_2 (Flatten)          (None, 4096)              0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 128)               524416    \n",
      "_________________________________________________________________\n",
      "dropout_4 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 1)                 129       \n",
      "=================================================================\n",
      "Total params: 547,841\n",
      "Trainable params: 547,841\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "%%px\n",
    "\n",
    "# Model config\n",
    "conv_sizes = [16, 32, 64]\n",
    "fc_sizes = [128]\n",
    "optimizer = 'Adam'\n",
    "lr = 0.001 * hvd.size()\n",
    "dropout=0.2\n",
    "\n",
    "# Training config\n",
    "batch_size = 128\n",
    "n_epochs = 4\n",
    "\n",
    "# Build the model\n",
    "model = build_model(train_input.shape[1:],\n",
    "                    conv_sizes=conv_sizes, fc_sizes=fc_sizes,\n",
    "                    dropout=dropout,\n",
    "                    optimizer=optimizer, lr=lr,\n",
    "                    use_horovod=True)\n",
    "if hvd.rank() == 0:\n",
    "    model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%px\n",
    "\n",
    "# Train the model\n",
    "history = train_model(model, train_input=train_input, train_labels=train_labels,\n",
    "                      valid_input=valid_input, valid_labels=valid_labels,\n",
    "                      batch_size=batch_size, n_epochs=n_epochs,\n",
    "                      use_horovod=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Can I get worker-local variables out?\n",
    "epochs = c[0].get('history.epoch')\n",
    "histories = c[:].get('history.history')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(9,4))\n",
    "\n",
    "# Plot the loss\n",
    "plt.subplot(121)\n",
    "plt.plot(epochs, histories[0]['val_loss'], label='Validation')\n",
    "for i, h in enumerate(histories):\n",
    "    plt.plot(epochs, h['loss'])#, label='Train %i' % i)\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Loss')\n",
    "plt.legend(loc=0)\n",
    "\n",
    "# Plot the accuracy\n",
    "plt.subplot(122)\n",
    "plt.plot(epochs, histories[0]['val_acc'], label='Validation')\n",
    "for i, h in enumerate(histories):\n",
    "    plt.plot(epochs, h['acc'])#, label='Train %i' % i)\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.legend(loc=0)\n",
    "\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluate on the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn import metrics\n",
    "\n",
    "def summarize_metrics(labels, outputs, threshold=0.5, weights=None):\n",
    "    preds = outputs > threshold\n",
    "    #print('Metrics summaries with threshold of %.3f' % threshold)\n",
    "    print('Accuracy:   %.4f' % metrics.accuracy_score(labels, preds, sample_weight=weights))\n",
    "    print('Purity:     %.4f' % metrics.precision_score(labels, preds, sample_weight=weights))\n",
    "    print('Efficiency: %.4f' % metrics.recall_score(labels, preds, sample_weight=weights))\n",
    "\n",
    "def draw_roc(labels, outputs, weights=None, ax=None):\n",
    "    fpr, tpr, _ = metrics.roc_curve(labels, outputs, sample_weight=weights)\n",
    "    auc = metrics.roc_auc_score(labels, outputs, sample_weight=weights)\n",
    "    if ax is None:\n",
    "        fig, ax = plt.subplots()\n",
    "    ax.plot(fpr, tpr, label='CNN, AUC=%.3f' % auc)\n",
    "    ax.plot([0, 1], [0, 1], '--', label='Random')\n",
    "    ax.set_xlabel('False positive rate')\n",
    "    ax.set_ylabel('True positive rate')\n",
    "    ax.legend(loc=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%px\n",
    "\n",
    "test_output = model.predict(test_input)\n",
    "test_output = test_output.squeeze(-1)\n",
    "score = model.evaluate(test_input, test_labels, verbose=2)\n",
    "print('Test loss:', score[0])\n",
    "print('Test accuracy:', score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_output = c[0].get('test_output')\n",
    "test_labels = c[0].get('test_labels')\n",
    "test_weights = c[0].get('test_weights')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Unweighted metrics\n",
      "Accuracy:   0.9834\n",
      "Purity:     0.9802\n",
      "Efficiency: 0.9813\n",
      "\n",
      "Weighted metrics\n",
      "Accuracy:   0.9672\n",
      "Purity:     0.0017\n",
      "Efficiency: 0.9813\n"
     ]
    }
   ],
   "source": [
    "# Unweighted results\n",
    "print('Unweighted metrics')\n",
    "summarize_metrics(test_labels, test_output)\n",
    "print()\n",
    "\n",
    "# Weighted results\n",
    "print('Weighted metrics')\n",
    "summarize_metrics(test_labels, test_output, weights=test_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(1, 2, figsize=(9,4))\n",
    "draw_roc(test_labels, test_output, ax=axs[0])\n",
    "draw_roc(test_labels, test_output, ax=axs[1], weights=test_weights)\n",
    "axs[0].set_xlim([0, 0.001])\n",
    "axs[0].set_title('Unweighted')\n",
    "axs[1].set_xlim([0, 0.001])\n",
    "axs[1].set_title('Weighted');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize the model output\n",
    "plt.figure()\n",
    "\n",
    "# Select real/fake sample model outputs\n",
    "real_idx = test_labels > 0.5\n",
    "fake_idx = (real_idx == False)\n",
    "\n",
    "binning=dict(bins=100, range=(0, 1), log=True)\n",
    "plt.hist(test_output[fake_idx], weights=test_weights[fake_idx], label='Fake', **binning)\n",
    "plt.hist(test_output[real_idx], weights=test_weights[real_idx], label='Real', **binning)\n",
    "plt.legend(loc=0);"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tensorflow-intel(cpu)/1.13.1-py36",
   "language": "python",
   "name": "tensorflow_intel_1.13.1_py36"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}