precision_and_speed.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## This demo shows how computing the same operation using different NumPy floating point precisions can have drastically different execution times.  \n",
    "\n",
    "The following data was generated on an Intel Core Intel Core i7-8650U 3.90 GHz quad-core processor with AVX2. The `@` symbol is the NumPy matrix multiplication operator. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "n = 500"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "395 ms ± 2.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "a = np.random.random((n,n)).astype(dtype=np.float16)\n",
    "b = np.random.random((n,n)).astype(dtype=np.float16)\n",
    "\n",
    "float16_time = %timeit -o a @ b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.78 ms ± 93.6 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)\n"
     ]
    }
   ],
   "source": [
    "a = np.random.random((n,n)).astype(dtype=np.float32)\n",
    "b = np.random.random((n,n)).astype(dtype=np.float32)\n",
    "\n",
    "float32_time = %timeit -o a @ b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.71 ms ± 169 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
     ]
    }
   ],
   "source": [
    "a = np.random.random((n,n)).astype(dtype=np.float64)\n",
    "b = np.random.random((n,n)).astype(dtype=np.float64)\n",
    "\n",
    "float64_time = %timeit -o a @ b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "float32 speed up is 221.5 times faster than float16\n",
      "float64 speed up is 83.9 times faster than float16\n",
      "float32 speed up is 2.6 times faster than float64\n"
     ]
    }
   ],
   "source": [
    "print('float32 speed up is {0:.1f} times faster than float16'.format(float16_time.average/float32_time.average))\n",
    "print('float64 speed up is {0:.1f} times faster than float16'.format(float16_time.average/float64_time.average))\n",
    "print('float32 speed up is {0:.1f} times faster than float64'.format(float64_time.average/float32_time.average))"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}