gpu_benchmark.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tested on i7-6800K with GTX 1080"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import time\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset shape: (13454, 784)\n",
      "Non-zeros rate: 0.16339869714795388\n",
      "Classes balance: 0.4692284822357663 / 0.5307715177642337\n"
     ]
    }
   ],
   "source": [
    "from sklearn.datasets import fetch_mldata\n",
    "from sklearn.preprocessing import scale\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import roc_auc_score, accuracy_score\n",
    "\n",
    "\n",
    "mnist = fetch_mldata('MNIST original', data_home='./tmp')\n",
    "\n",
    "# only binary classification supported\n",
    "mask = (mnist['target'] == 3) + (mnist['target'] == 5)\n",
    "\n",
    "X_all = scale(mnist['data'][mask].astype(float))\n",
    "y_all = (mnist['target'][mask]==3)*1\n",
    "\n",
    "# make it more sparse\n",
    "X_all = X_all * (np.random.uniform(0, 1, X_all.shape) > 0.8)\n",
    "\n",
    "print('Dataset shape: {}'.format(X_all.shape))\n",
    "print('Non-zeros rate: {}'.format(np.mean(X_all != 0)))\n",
    "print('Classes balance: {} / {}'.format(np.mean(y_all==0), np.mean(y_all==1)))\n",
    "\n",
    "X_tr, X_te, y_tr, y_te = train_test_split(X_all, y_all, random_state=42, test_size=0.3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Dense data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from tffm import TFFMClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU only / dense data\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 100/100 [00:20<00:00,  4.77epoch/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.91726529601189\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "model = TFFMClassifier(\n",
    "    order=3,\n",
    "    rank=10, \n",
    "    optimizer=tf.train.AdamOptimizer(learning_rate=0.01), \n",
    "    n_epochs=100, \n",
    "    batch_size=-1,\n",
    "    init_std=0.001,\n",
    "    reg=0.001,\n",
    "    input_type='dense',\n",
    "    session_config=tf.ConfigProto(log_device_placement=True, device_count = {'GPU': 0}),\n",
    "    seed=42\n",
    ")\n",
    "print('CPU only / dense data')\n",
    "model.fit(X_tr, y_tr, show_progress=True)\n",
    "predictions = model.predict(X_te)\n",
    "print('accuracy: {}'.format(accuracy_score(y_te, predictions)))\n",
    "model.destroy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "in stderr: main_block/order_2/Pow/y: (Const): /job:localhost/replica:0/task:0/cpu:0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GPU / dense data\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 100/100 [00:06<00:00, 15.26epoch/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.91726529601189\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "model = TFFMClassifier(\n",
    "    order=3,\n",
    "    rank=10, \n",
    "    optimizer=tf.train.AdamOptimizer(learning_rate=0.01), \n",
    "    n_epochs=100, \n",
    "    batch_size=-1,\n",
    "    init_std=0.001,\n",
    "    reg=0.001,\n",
    "    input_type='dense',\n",
    "    session_config=tf.ConfigProto(log_device_placement=True),\n",
    "    seed=42\n",
    ")\n",
    "print('GPU / dense data')\n",
    "model.fit(X_tr, y_tr, show_progress=True)\n",
    "predictions = model.predict(X_te)\n",
    "print('accuracy: {}'.format(accuracy_score(y_te, predictions)))\n",
    "model.destroy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "in stderr: main_block/order_2/mul_1/x: (Const): /job:localhost/replica:0/task:0/gpu:0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sparse example"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import scipy.sparse as sp\n",
    "# only CRS format supported\n",
    "X_tr_sparse = sp.csr_matrix(X_tr)\n",
    "X_te_sparse = sp.csr_matrix(X_te)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU only / sparse data\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 100/100 [00:29<00:00,  3.49epoch/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.91726529601189\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "model = TFFMClassifier(\n",
    "    order=3,\n",
    "    rank=10, \n",
    "    optimizer=tf.train.AdamOptimizer(learning_rate=0.01), \n",
    "    n_epochs=100, \n",
    "    batch_size=-1,\n",
    "    init_std=0.001,\n",
    "    reg=0.001,\n",
    "    input_type='sparse',\n",
    "    session_config=tf.ConfigProto(log_device_placement=True, device_count = {'GPU': 0}),\n",
    "    seed=42\n",
    ")\n",
    "print('CPU only / sparse data')\n",
    "model.fit(X_tr_sparse, y_tr, show_progress=True)\n",
    "predictions = model.predict(X_te_sparse)\n",
    "print('accuracy: {}'.format(accuracy_score(y_te, predictions)))\n",
    "model.destroy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GPU / sparse data\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 100/100 [00:11<00:00,  8.90epoch/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.91726529601189\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "model = TFFMClassifier(\n",
    "    order=3,\n",
    "    rank=10, \n",
    "    optimizer=tf.train.AdamOptimizer(learning_rate=0.01), \n",
    "    n_epochs=100, \n",
    "    batch_size=-1,\n",
    "    init_std=0.001,\n",
    "    reg=0.001,\n",
    "    input_type='sparse',\n",
    "    session_config=tf.ConfigProto(log_device_placement=True),\n",
    "    seed=42\n",
    ")\n",
    "print('GPU / sparse data')\n",
    "model.fit(X_tr_sparse, y_tr, show_progress=True)\n",
    "predictions = model.predict(X_te_sparse)\n",
    "print('accuracy: {}'.format(accuracy_score(y_te, predictions)))\n",
    "model.destroy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# n_features/time complexity"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 9/9 [04:17<00:00, 29.19s/it]\n"
     ]
    }
   ],
   "source": [
    "n_features = X_all.shape[1]\n",
    "used_features = range(100, 1000, 100)\n",
    "n_repeats = 5\n",
    "elapsed_mean = {'gpu':[], 'cpu':[]}\n",
    "elapsed_std = {'gpu':[], 'cpu':[]}\n",
    "model_title = ''\n",
    "\n",
    "for cur_n_feats in tqdm(used_features):\n",
    "    for mode, config in [('gpu', None), ('cpu', tf.ConfigProto(device_count = {'GPU': 0}))]:\n",
    "        \n",
    "        time_observation = []\n",
    "        for _ in range(n_repeats):\n",
    "            active_features = np.random.choice(range(n_features), size=cur_n_feats)\n",
    "            model = TFFMClassifier(\n",
    "                        order=5, \n",
    "                        rank=50,\n",
    "                        optimizer=tf.train.AdamOptimizer(learning_rate=0.001), \n",
    "                        n_epochs=1, \n",
    "                        batch_size=-1,\n",
    "                        init_std=0.01,\n",
    "                        session_config=config,\n",
    "                        input_type='dense'\n",
    "            )\n",
    "            model_title = model.__str__()\n",
    "            # manually initialize model without calling .fit()\n",
    "            model.core.set_num_features(cur_n_feats)\n",
    "            model.core.build_graph()\n",
    "            model.initialize_session()\n",
    "\n",
    "            start_time = time.time()\n",
    "            predictions = model.decision_function(X_all[:, active_features])\n",
    "            end_time = time.time()\n",
    "            model.destroy()\n",
    "            time_observation.append(end_time - start_time)\n",
    "        elapsed_mean[mode].append(np.mean(time_observation))\n",
    "        elapsed_std[mode].append(np.std(time_observation))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAicAAAGHCAYAAABrpPKuAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSExRciIiIiIx5X8AKvGqWpdWgd0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f237b38bef0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%pylab inline\n",
    "errorbar(used_features, elapsed_mean['cpu'], yerr=elapsed_std['cpu'], label='CPU')\n",
    "errorbar(used_features, elapsed_mean['gpu'], yerr=elapsed_std['gpu'], label='GPU')\n",
    "xlim(0, 1000)\n",
    "title('CPU/GPU times')\n",
    "xlabel('n_features')\n",
    "ylabel('test time')\n",
    "legend(loc=\"upper left\")\n",
    "grid()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}