whats_new_numpy_1.8

{
 "metadata": {
  "name": "What's new in numpy 1.8"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "What's new in NumPy 1.8\n=======================\n\nRecent NumPy releases\n---------------------\n* 1.5.0 - August 2010\n    * 1.5.1 - February 2011 (bug fixes)\n* 1.6.0 \u2013 May 2011\n    * 1.6.1, 1.6.2 \u2013 July 2011, August 2012 (bug fixes)\n* 1.7.0 \u2013 February 2013\n    * 1.7.1, 1.7.2 \u2013 April 2013, December 2013 (bug fixes)\n* 1.8.0 \u2013 October 2013"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "What numpy are you running?\n---------------------------"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "import numpy as np\nnp.__version__",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 1,
       "text": "'1.8.0'"
      }
     ],
     "prompt_number": 1
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "From the release notes\n----------------------\n\n**Python versions supported**\n\n* This release supports Python 2.6 - 2.7 and 3.2 \u2013 3.3.\n* Support for Python 2.4 \u2013 2.5 has been dropped.\n\n**Highlights**\n\n* New, gufuncs for linear algebra.\n* New, in-place fancy indexing with the `.at` method.\n* New, `partition` function, partial sorting via selection.\n* New, `nanmean`, `nanvar`, and `nanstd` functions.\n* New, `full` and `full_like` functions for initialized arrays.\n\n**Highlight from 1.7.0**\n\n* New, reduction ufuncs generalize `axis=` parameter.\n"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`full` and `full_like`\n----------------------"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.ones((2, 4)) * 5",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 2,
       "text": "array([[ 5.,  5.,  5.,  5.],\n       [ 5.,  5.,  5.,  5.]])"
      }
     ],
     "prompt_number": 2
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.empty((2, 4))\na.fill(5)\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 3,
       "text": "array([[ 5.,  5.,  5.,  5.],\n       [ 5.,  5.,  5.,  5.]])"
      }
     ],
     "prompt_number": 3
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.full((2, 4), fill_value=5)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 4,
       "text": "array([[ 5.,  5.,  5.,  5.],\n       [ 5.,  5.,  5.,  5.]])"
      }
     ],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.full_like(a, fill_value=5)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 5,
       "text": "array([[ 5.,  5.,  5.,  5.],\n       [ 5.,  5.,  5.,  5.]])"
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`nanmean`, `nanstd` and `nanvar`\n--------------------------------"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Setup**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.arange(5, dtype=np.double)\na[2] = np.nan\na\n",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 6,
       "text": "array([  0.,   1.,  nan,   3.,   4.])"
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.mean(a)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 7,
       "text": "nan"
      }
     ],
     "prompt_number": 7
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.mean(a[~np.isnan(a)])",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 8,
       "text": "2.0"
      }
     ],
     "prompt_number": 8
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.nanmean(a)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 9,
       "text": "2.0"
      }
     ],
     "prompt_number": 9
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`.at`\n-----"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Setup**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.zeros(3)\nidx = [0, 0, 1, 2, 2 ,2]\na\n",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 10,
       "text": "array([ 0.,  0.,  0.])"
      }
     ],
     "prompt_number": 10
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a[idx] += 1\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 11,
       "text": "array([ 1.,  1.,  1.])"
      }
     ],
     "prompt_number": 11
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a[:] = 0\na += np.bincount(idx)\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 12,
       "text": "array([ 2.,  1.,  3.])"
      }
     ],
     "prompt_number": 12
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a[:] = 0\nnp.add.at(a, idx, 1)\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 13,
       "text": "array([ 2.,  1.,  3.])"
      }
     ],
     "prompt_number": 13
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a[:] = 1\nnp.multiply.at(a, idx, 2)\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 14,
       "text": "array([ 4.,  2.,  8.])"
      }
     ],
     "prompt_number": 14
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`partition`\n-----------"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Setup**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.arange(11)\nnp.random.shuffle(a)\na",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 15,
       "text": "array([ 7,  6,  2,  4,  3,  8,  1, 10,  5,  9,  0])"
      }
     ],
     "prompt_number": 15
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**\n"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.sort(a)[5]",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 16,
       "text": "5"
      }
     ],
     "prompt_number": 16
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.partition(a, 5)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 17,
       "text": "array([ 3,  1,  2,  0,  4,  5,  6,  7, 10,  9,  8])"
      }
     ],
     "prompt_number": 17
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.partition(a, 5)[5]",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 18,
       "text": "5"
      }
     ],
     "prompt_number": 18
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`partition` can be much faster than sorting..."
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.random.rand(1e7)",
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 19
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit np.sort(a)[:5]",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "1 loops, best of 3: 1.36 s per loop\n"
      }
     ],
     "prompt_number": 20
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit np.partition(a, np.arange(5))[:5]",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "10 loops, best of 3: 118 ms per loop\n"
      }
     ],
     "prompt_number": 21
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "np.allclose(np.sort(a)[:5], np.partition(a, np.arange(5))[:5])",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 22,
       "text": "True"
      }
     ],
     "prompt_number": 22
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "`solve` and several other linalg gufuncs\n----------------------------------------"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Setup**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.random.rand(1e6, 2, 2)",
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 30
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**\n"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%%timeit\nout = np.empty((1e6 ,2))\nfor j, mat in enumerate(a):\n    out[j] = np.linalg.solve(mat, [1, 1])",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "1 loops, best of 3: 29.7 s per loop\n"
      }
     ],
     "prompt_number": 24
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit np.linalg.solve(a, [1, 1])",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "1 loops, best of 3: 417 ms per loop\n"
      }
     ],
     "prompt_number": 25
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "Full list of available gufuncs:\n\n* `cholesky` - Cholesky decomposition\n* `det` - determinant\n* `eig` - eigenvalues and right eigenvectors\n* `eigh` - eigenvalues and right eigenvectors (hermitian/symmetric)\n* `eigvals` - eigenvalues\n* `eigvalsh` - eigenvalues (hermitian/symmetric)\n* `inv` - multiplicative matrix inverse\n* `slogdet` - sign and logarithm of determinant\n* `solve` - solve system of linear equations\n* `svd` - singular value decomposition"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "generalized `axis=` parameter for ufunc reduction\n-------------------------------------------------"
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Setup**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "a = np.random.rand(1000, 1000, 10)",
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 26
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**Before**\n"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit a.max(axis=-1).max(axis=-1)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "10 loops, best of 3: 67.1 ms per loop\n"
      }
     ],
     "prompt_number": 27
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit a.max(axis=-2).max(axis=-1)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "10 loops, best of 3: 33.2 ms per loop\n"
      }
     ],
     "prompt_number": 28
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": "**After**"
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "%timeit a.max(axis=(-1, -2))",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "100 loops, best of 3: 10.2 ms per loop\n"
      }
     ],
     "prompt_number": 29
    }
   ],
   "metadata": {}
  }
 ]
}