NN prediction benchmark

torch_vs_keras_vs_tf.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**update**: Install Intel-optimized tensorflow according to https://github.com/tensorflow/tensorflow/issues/17028#issuecomment-366548366. Now TF is as fast as PyTorch.\n",
    "\n",
    "**NN prediction timing (2-layer MLP)**\n",
    "\n",
    "|PyTorch|Keras (TF)|TensorFlow|\n",
    "|------|------|------|\n",
    "|133 ms|379 ms|131 ms|"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Platform info"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- System: AWS Ubuntu 16.04 base AMI (ami-66506c1c)\n",
    "- Instance type: c5.large, CPU-only\n",
    "\n",
    "Software installation:\n",
    "- Python: Miniconda Python 3.6 https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh\n",
    "- TensorFlow: \n",
    "```conda install -c intel tensorflow```\n",
    "- Keras: ```pip install keras```\n",
    "- PyTorch: ```conda install pytorch```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Architecture:          x86_64\r\n",
      "CPU op-mode(s):        32-bit, 64-bit\r\n",
      "Byte Order:            Little Endian\r\n",
      "CPU(s):                2\r\n",
      "On-line CPU(s) list:   0,1\r\n",
      "Thread(s) per core:    2\r\n",
      "Core(s) per socket:    1\r\n",
      "Socket(s):             1\r\n",
      "NUMA node(s):          1\r\n",
      "Vendor ID:             GenuineIntel\r\n",
      "CPU family:            6\r\n",
      "Model:                 85\r\n",
      "Model name:            Intel(R) Xeon(R) Platinum 8124M CPU @ 3.00GHz\r\n",
      "Stepping:              3\r\n",
      "CPU MHz:               3000.000\r\n",
      "BogoMIPS:              6000.00\r\n",
      "Hypervisor vendor:     KVM\r\n",
      "Virtualization type:   full\r\n",
      "L1d cache:             32K\r\n",
      "L1i cache:             32K\r\n",
      "L2 cache:              1024K\r\n",
      "L3 cache:              25344K\r\n",
      "NUMA node0 CPU(s):     0,1\r\n",
      "Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single kaiser fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f rdseed adx smap clflushopt clwb avx512cd xsaveopt xsavec xgetbv1 ida arat\r\n"
     ]
    }
   ],
   "source": [
    "!lscpu"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_sample = 20000\n",
    "n_feature = 500\n",
    "X = np.random.rand(n_sample, n_feature)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# use same number of hidden layers for different frameworks\n",
    "H1, H2 = 400, 400 "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Pytorch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'0.3.0'"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch\n",
    "from torch.autograd import Variable\n",
    "torch.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = torch.nn.Sequential(\n",
    "    torch.nn.Linear(n_feature, H1),\n",
    "    torch.nn.ReLU(),\n",
    "    torch.nn.Linear(H1, H2),\n",
    "    torch.nn.ReLU(),\n",
    "    torch.nn.Linear(H2, 1)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_torch = Variable(torch.from_numpy(X).type(torch.FloatTensor))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "133 ms ± 1.18 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)\n"
     ]
    }
   ],
   "source": [
    "%timeit model(X_torch).data.numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Keras (TF-backend)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'2.1.4'"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import keras\n",
    "from keras import models, layers\n",
    "keras.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "network = models.Sequential()\n",
    "network.add(layers.Dense(H1, activation='relu', input_shape=(n_feature,)))\n",
    "network.add(layers.Dense(H2, activation='relu'))\n",
    "network.add(layers.Dense(1, activation=None))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "379 ms ± 560 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit network.predict(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# TensorFlow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'1.3.1'"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.contrib.layers import fully_connected\n",
    "tf.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.reset_default_graph()\n",
    "\n",
    "X_tf = tf.placeholder(tf.float32, shape=(None, n_feature), name=\"X\")\n",
    "\n",
    "hidden1 = fully_connected(X_tf, H1, scope=\"hidden1\") \n",
    "hidden2 = fully_connected(hidden1, H2, scope=\"hidden2\") \n",
    "out = fully_connected(hidden2, 1, scope=\"outputs\", activation_fn=None)\n",
    "\n",
    "init = tf.global_variables_initializer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "131 ms ± 124 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess: \n",
    "    init.run()\n",
    "    %timeit out.eval(feed_dict={X_tf: X})"
   ]
  }
 ],
 "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": "3.6.3"
  },
  "toc": {
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}