cs109.ipynb

{
 "metadata": {
  "name": "",
  "signature": "sha256:98cf6883d0157d5abb6e699b55c0c43cf1718201d7c363928fc7dc2afad66454"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "You've Got Mail: CS109 Project"
     ]
    },
    {
     "cell_type": "heading",
     "level": 4,
     "metadata": {},
     "source": [
      "Robbie Gibson, Anson Kahng, Lisa Wang, Willy Xiao"
     ]
    },
    {
     "cell_type": "heading",
     "level": 4,
     "metadata": {},
     "source": [
      "Our github URL: https://github.com/willyxiao/cs109-hcs"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "<img src=\"https://dl.dropboxusercontent.com/u/80118923/mail.jpg\">"
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Overview and Motivation"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Harvard students send over 800,000 emails almost every year, reflecting information about their activities, classes, moods, interests, etc. There are over 7,000 lists and over 5 million emails hosted on the Harvard Computer Society's servers - thus serving as an incredible amount of data concerning student life. We are interested in exploring the ways Harvard students communicate with each other through email, and we decided to use the public HCS archives in order to analyze trends in campus-wide communication. This is an incredible amount of data about Harvard students, with patterns in social interaction and student habits hidden within the millions of emails. Our motivations are to better understand the interactions within an email list so that students can tailor their emails to predict responses and communicate more effectively.\n",
      "\n",
      "Thus, our goals in this project are to 1) understand the network created through an email list and see if we can determine the hierarchy of the organization using that list, i.e. can we figure out who is an important member of the club based on how many responses they get to their emails and 2) predict whether a student will get a response given an email and a list. "
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Related Work"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We're all quite interested in machine learning and text analysis, and a couple members have dabbled in part-time research projects dealing with similar topics. However, these endeavors haven't been very in-depth, and we're hoping to explore a more comprehensive view of the subject in this project. \n",
      "\n",
      "Through the lectures on creating classifiers for the use of predictions, we realized we could do something similar with the emails we have, training and testing our classifier on the lists we have, and using it to help students predict responses. "
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Initial Questions"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We initially wanted to consider how email predicts student interests, i.e. careers, events, etc. However, we realized that what we really wanted to analyze was how students communicate. This includes understanding the networks created on an email list, in particular visualizing clusters created by email responses and response times. We wanted to know if we could determine the hierarchy of the club based on these networks, i.e. who was president that year. We also wanted to create a classifier for emails that get a response and emails that don't. We wanted to know if there was a way to predict the response time of an email sent over a particular list.  "
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Data"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Check Point 1: Data Acquisition:\n",
      "---\n",
      "\n",
      "The easy part about the data is that it all lives in one place and is well-formatted through the archival format of an open-source mailing list program called mailman. The difficult part was how we were going to process and act on the data.\n",
      "\n",
      "The format of is as follows\n",
      "\n",
      "> list-name.mbox\n",
      "\n",
      "> Here\u2019s a link to more about the .mbox format: http://en.wikipedia.org/wiki/Mbox\n",
      "\n",
      "> Here\u2019s a python module that processes it: https://docs.python.org/2/library/mailbox.html\n",
      "\n",
      "We stood on the shoulders of giants.\n",
      "\n",
      "All emails spanning back to 2001 (including indexing, publicly available html pages, and attachments) made up about 4 terabytes of data that lived on an HCS production server that is constantly running and serving the school. Because of the load of the server, we obviously could not process the data on the HCS servers (especially because the archives are also constantly updating).\n",
      "\n",
      "The solution: Willy happened to have access to $2,000 / month of credits for an unnamed hosting platform. He was able to spin up a server along with an instance of NFS (network-file-server for block storage) and take a data-cut of all of the archival data. After filtering out the attachments and indexing, we were left with 250 GB of raw text that needed to be transferred over a network to our new servers! Yay! :)\n",
      "\n",
      "The following solution worked quite well.\n",
      "\n",
      "1. ssh into our server:\n",
      "> ssh root@208.43.122.13\n",
      "\n",
      "2. open a new instance of a screen we can come back to later:\n",
      "> screen \n",
      "\n",
      "3. \trun the following bash file:\n",
      "> rsync -vrh --exclude=attachments/* -e \"ssh\" --log-file=rsync.log --progress --stats hcs@trajan.hcs.harvard.edu:/nfs/mailman/archives/ archives/\n",
      "\n",
      "This took roughly 3 days to run..."
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Check Point 2: Basic Data-Processing\n",
      "---\n",
      "\n",
      "On the HCS servers, running an \u201cls\u201d command in the archive directory takes roughly 1 minute to return. After filtering the data and copying the data to our hawt-chrome-bare-metal-server it takes roughly 2-3 seconds to run the \u201cls\u201d command. Yay!\n",
      "\n",
      "But some lists, like the \u201caaa-open.mbox\u201d list contains millions of lines of text. To try a proof-of-concept on Checkpoint 2, we tried to run a simple script that iterated over each of the .mbox files and counted how many emails there were in total."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "#!/usr/bin/python\n",
      "\n",
      "import glob\n",
      "import mailbox\n",
      "import sys\n",
      "\n",
      "total_len = 0\n",
      "\n",
      "for f in glob.glob('/mnt/archives/*\u2019'):\n",
      "  m = mailbox.mbox(f)\n",
      "  f_len = m.__len__()\n",
      "  total_len += f_len\n",
      "  print \"Length of \" + f + \": \" + str(f_len)\n",
      "\n",
      "# print total_len"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "This job took about 10 hours or so to run - but we got results! Roughly 5 million emails! :)\n",
      "\n",
      "So obviously this wasn\u2019t going to work in the long-run. With an $O(n)$ algorithm with a small constant taking 10 hours to run, there\u2019s no way we\u2019d be able to do anything more complicated or\u2026.an $O(n*log(n))$ algorithm, no way!\n",
      "\n",
      "So, we used the rest of my $2,000 / month on buying more servers! Yay! We bought 4 more straight up bare-metal-servers on LAN connected to our NFS server. The startup script for each of those is here:"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "> apt-get install vim python-pip nfs-client dtach\n",
      "> mount -t nfs4 -o hard,intr nfswdc0101d.######## /mnt\n",
      "\n",
      "> echo \"set expandtab tabstop=2\n",
      "> set nu rnu\n",
      "> syntax on\n",
      "> set hlsearch\" >> ~/.vimrc\n",
      "\n",
      "> echo \"alias g='git'\n",
      "> alias gs='git status'\n",
      "> alias gb='git branch'\n",
      "> alias gci='git commit'\n",
      "> alias gco='git checkout'\n",
      "> alias gpl='git pull'\n",
      "> alias gph='git push'\" >> ~/.bash_profile\n",
      "\n",
      "> echo \"ssh-rsa ####### root@staging2\n",
      "\" >> ~/.ssh/authorized_keys\n",
      "\n",
      "> pip install mincemeat\n",
      "> pip install mailbox"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Then we decided it probably wasn\u2019t worth it to set up hadoop and use mrjob, so we looked for a light-weight map-reduce framework, we found it in what\u2019s called mincemeat.py: link here https://github.com/michaelfairley/mincemeatpy.\n",
      "\n",
      "Then we wrote our handler on top of it called mrjob.py (commenting out some things you can't actually execute in an ipython notebook):"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# import mincemeat\n",
      "# import glob\n",
      "# import logging\n",
      "# import os\n",
      "# import signal\n",
      "# import sys\n",
      "# import time\n",
      "# import pickle\n",
      "# import fabric.api\n",
      "# import fabric.tasks\n",
      "# import fabric.network\n",
      "\n",
      "class MrJob:\n",
      "  password = \"hello\"\n",
      "  server_ip = \"10.102.75.2\"\n",
      "  client_ips = [\"10.102.75.4\", \"10.102.75.6\", \"10.102.75.8\", \"208.43.122.12\"]\n",
      "\n",
      "  def __init__(self, mapfn, reducefn, name=None, test=False, email='willy@chenxiao.us'):\n",
      "    self.start_time = time.strftime(\"%Y%m%d%H%M%S\")\n",
      "\n",
      "    # use generic file-name if there isn't one already\n",
      "    if name is None:\n",
      "      self.name = \"mrjob\" + self.start_time\n",
      "    else:\n",
      "      self.name = name + self.start_time\n",
      "\n",
      "    # command to kick off the clients\n",
      "    self.command = \"dtach -n /tmp/\" + self.name + \" /usr/local/bin/mincemeat.py -p hello 10.102.75.2\"\n",
      "\n",
      "    self.mapfn = mapfn\n",
      "    self.reducefn = reducefn\n",
      "\n",
      "    glob_prefix = \"/mnt\"\n",
      "    glob_suffix = \"*.mbox\"\n",
      "    glob_body = \"subset\" if test else \"archives\"\n",
      "    glob_str = \"/\".join([glob_prefix, glob_body, glob_suffix])\n",
      "\n",
      "    # initialize data to run mapper and reducer off of\n",
      "    self.data = dict(enumerate(glob.glob(glob_str)))\n",
      "\n",
      "  def run(self):\n",
      "    pid = os.fork()\n",
      "\n",
      "    # start clients\n",
      "    if pid == 0:\n",
      "      time.sleep(5)\n",
      "      print \"Starting clients...\"\n",
      "      self.start_clients()\n",
      "\n",
      "    # start server\n",
      "    else:\n",
      "      print \"Starting server...\"\n",
      "      self.results = self.start_server()\n",
      "      return self.results\n",
      "\n",
      "  def start_clients(self):\n",
      "    fabric.tasks.execute(lambda: fabric.api.run(self.command), hosts=[\"root@\" + ip for ip in self.client_ips])\n",
      "    fabric.network.disconnect_all()\n",
      "\n",
      "  def start_server(self):\n",
      "    # make sure outfiles and logs exist\n",
      "    logname = self.name + \".log\"\n",
      "    logging.basicConfig(filename=logname,level=logging.INFO)\n",
      "    outname = self.name + \".out\"\n",
      "    outfile = open(outname, \"w\")\n",
      "\n",
      "    # start mincemeat and save the results!\n",
      "    s = mincemeat.Server()\n",
      "    s.datasource = self.data\n",
      "    s.mapfn = self.mapfn\n",
      "    s.reducefn = self.reducefn\n",
      "    result = s.run_server(password=self.password)\n",
      "    pickle.dump(result, outfile)\n",
      "    return result"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 14
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Now we had a running map-reduce cluster of 5 nodes with 1 master and 4 workers. Our same counting script now takes 2-3 hours, definitely something much more manageable. So now, we move to exploratory analysis!"
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Exploratory Analysis"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Now that we had every set up, we could do tons of interesting exploratory analysis on our data fairly quickly. For instance, here's a script we ran for finding a distribution of when emails were sent (again with non-working parts in ipython commented out):\n",
      "\n",
      "Exploration 1: Email Sent Times\n",
      "---"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# import mrjob\n",
      "\n",
      "def mapfn(k, v):\n",
      "  import mailbox\n",
      "  import email.utils\n",
      "  import time\n",
      "  import datetime\n",
      "\n",
      "  mbox = mailbox.mbox(v)\n",
      "  for msg in mbox:\n",
      "    try:\n",
      "      email_date = datetime.datetime.fromtimestamp(time.mktime(email.utils.parsedate(msg['Date'])))\n",
      "        \n",
      "      # mincemeat yields a k, vs\n",
      "      yield (email_date.year, email_date.month, email_date.weekday(), email_date.hour), 1\n",
      "    except:\n",
      "      yield None, 1\n",
      "\n",
      "def reducefn(k, vs):\n",
      "  return sum(vs)\n",
      "\n",
      "# if __name__ == '__main__':\n",
      "#   job = mrjob.MrJob(mapfn, reducefn, name=\"dayhour\", test=False)\n",
      "#   print job.run()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 22
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "The following cells can be executed with the dayhour20131106001106.out file sent to our TF, Tudor."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%matplotlib inline  \n",
      "import matplotlib.pyplot as plt\n",
      "import pickle\n",
      "import seaborn as sns\n",
      "import numpy as np"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 32
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "FILE_NAME = \"dayhour20141106001106.out\"\n",
      "results = pickle.load(open(FILE_NAME, 'r'))"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 11
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# initialize dict to avoid O(n^2) concatenation\n",
      "weekdays = {weekday: 0 for weekday in xrange(7)}\n",
      "hours = {hour: 0 for hour in xrange(24)}\n",
      "months = {month: 0 for month in xrange(1, 13)}\n",
      "years = {year: 0 for year in xrange(2000, 2015)}\n",
      "\n",
      "# store all the results\n",
      "for t in results:\n",
      "    if t is None:\n",
      "        continue\n",
      "    year, month, weekday, hour = t\n",
      "    \n",
      "    # some dates are messed up formatting, just skip them\n",
      "    if year < 2000:\n",
      "        continue\n",
      "        \n",
      "    weekdays[weekday] += results[t]\n",
      "    hours[hour] += results[t]\n",
      "    months[month] += results[t]\n",
      "    years[year] += results[t]\n",
      "\n",
      "plt.figure()\n",
      "plt.bar(years.keys(), years.values(), color=\"#29abbb\")\n",
      "plt.xlabel(\"Year\")\n",
      "plt.ylabel(\"Number of Emails\")\n",
      "plt.title(\"Number of Emails vs. Year\")\n",
      "plt.show()\n",
      "\n",
      "plt.figure()\n",
      "plt.bar(months.keys(), months.values(), color=\"#29abbb\", align=\"center\")\n",
      "plt.xlabel(\"Month\")\n",
      "plt.ylabel(\"Number of Emails\")\n",
      "plt.title(\"Number of Emails vs. Month\")\n",
      "plt.xticks(months.keys(), ['January', 'February', 'March', 'April', 'May', 'June', 'July', 'August', 'September', 'October', 'November', 'December'], rotation=\"vertical\")\n",
      "plt.show()\n",
      "\n",
      "plt.figure()    \n",
      "plt.bar(weekdays.keys(), weekdays.values(),color=\"#29abbb\", align=\"center\")\n",
      "plt.xlabel(\"Day of the Week\")\n",
      "plt.ylabel(\"Number of Emails\")\n",
      "plt.title(\"Number of Emails vs. Day of the Week\")\n",
      "plt.xticks(weekdays.keys(), ['Sunday', 'Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday'], rotation=\"vertical\")\n",
      "plt.show()\n",
      "\n",
      "plt.figure()\n",
      "plt.bar(hours.keys(), hours.values(), color=\"#29abbb\")\n",
      "plt.xlabel(\"Hour\")\n",
      "plt.ylabel(\"Number of Emails\")\n",
      "plt.title(\"Number of Emails vs. Hour\")\n",
      "plt.xlim(1, 24)\n",
      "plt.show()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAhMAAAFwCAYAAAACK/lNAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10a02fb50>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAgcAAAGcCAYAAABN6AMmAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10aac9450>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAg0AAAGfCAYAAADcXuFBAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x109fa0610>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAgcAAAFwCAYAAAAsbtjfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3X+YnWV54PFvSEJAE4HABBMIBmnu3abLitBN3KoVq0hU\nFnDXTfAHZpfo5YortLYuoF0QYaPsttClXtK6IBC2RNKyIio/EqF2bbcaQEXbuHtDcZAMkMRM+FUU\nwzj7x/scOGecHycz58yZzPl+rmuuec/zvs9z7ifnTM593vd5nhckSZIkSZIkSZIkSZIkSZIkSZIk\nSZIk6UURcV1EXNLB5782Ivoj4ludiqHEcVVE/H7ZPjEiHulkPNJ0MavTAUjTQUT0AgcCR2fms6Xs\n/cB7MvONbXjKwfIz6SLi9cCbgUWZ+bNh9v874Brg2briQSAy8/FWxpKZH2ple60SEZ8EjsnMM4eU\n/wL4lcx8qCOBSU0yOZBaZz/gXODTk/R8M1rRSETsl5m/2IsqrwB6h0sM6vxNZv7mBEPbl01q4jaO\n11AalcmB1BqDwB8A/ykiPpeZT9bvjIglwEPArNp/4hHxDeCGzLymfNv+APBt4N8Du4D3Af8EuBiY\nA3wsM9fXNXtYRGwCXgN8B3hfZv64tP1PgT8Gjgd2Av85M/+87LsO+CnVh/xvAqcCdw+JdxHwJ8Br\ngX7gssy8OiLWAp8FZkfE08AfZObFw/x7jJi4lLMsny39OxrYCHwcuA74DWAL8G8z84ly/J8Dr6M6\nM3M/8KHM3FrXl0cy8z8P8zznAR8BXgY8CpydmUP7uQK4heosyGApewfwycx8VUQsBz4HLC3/Zn+W\nmb87Ut+a6X/dc88BLgP+bSnaCJyXmT8v74e1mfn6uuNfOOvQzGsoTcR+nQ5AmkbuBb4B/F6Txw+9\nNLCc6sNvPrCB6sPieOAY4L3AZyPiJeXYGcB7gE8BhwHfA/4MICJeCmwG/ifQA5wBfC4ifrXuud4F\nXJKZc4G/GSa2LwI/BhYC7wTWRcQbM/Ma4D8Af5uZ80ZIDJrp978G3kSV/JwC3A6cDyyg+n/pnLrj\nvwb8SunLd2r9rGvrl76lR8Q/AT4M/Hpmvgx4C9A79LjM/DbwjyWWmnfXPcd/B67IzIOAV1K9JuMS\nEUMThk9QveavKj/Lgd/fiybHeg2lcfPMgdQ6g8CFwN9ExH8fR/0fZeb1ABGxkerD41OZuQfYHBE/\np/qQ/H45/quZ+dfl+E8AT0bEkVTf9l9oC/heRPwvqm+onyplt2Tm3wJk5nP1QUTEYqpv8G/NzJ8D\n90fE1VTf9P+S5i5nvCYidtc9/klmLq17/MeZubM83zeB7Zl5f3n8Jeo+rDPzurrYLgbOjYh5mfl0\nKR4ungGqsy2/FhG7amdURrCB6oP26xExD3gr8NGy7+fA0og4LDN/QnVmp1mrIuKUUfa/G/iPpd1a\n3/6U6j3UjBFfQ2miTA6kFsrMv4+Ir1J9C/7hXlbfXrf909LeziFlc8v2ILCt7nn/MSL6gUVUp5pX\nDPlwngWsH67uMBYB/Zn5j3VlPwZ+vfmu8K36U+LDGNrX+sc/o/QzImYC/4Xq7EUPULuufhjwNCPI\nzAcj4reBT1IlCHcCH83Mx4Y5/Ebg/0TEh6jOaNyXmbVZD2upEqofRsSPgIsz82uj9KveTZn5vvqC\ncmmgZhHwcN3jH5eyZoz1GkoT4mUFqfUuoho/cERdWe2D9iV1ZS+fwHPMABbXHkTEXKrLEX1UHzJ/\nlZmH1P3My8wPN9n2o8D80mbNUbT3w2iksxHvprqe/qZyav/oYY4fdvBfZm4oCcoryjGXjXDcD6k+\npN9anu/Gun0PZua7M7On1P+LiDiwif4MjtKnmkeBJXWPjyplUL1fXnivRMRE3ivSXvPMgdRimfkP\nEXET1cyF75eynRHRB5wZEZ8H1lCNJZiIt0XEa4F7gEuoxgH0RcTXgM9ExHuBm8qxxwFPZ+b/ZYwP\nrcx8JCL+D/DpiPg9qnEBZ1F9cE62ucBzQH8ZS7FuyP4ZDNOfiAjgSKpr8c9RnY0Yrd83Ar8NrKC6\nxFBr573AneUMzpNUH/rNzApo5tLLBuD3I+Ke8vhC4IayfT/VGY9XAf+P6gzI3rYvjZtnDqT2+BTV\nN7/6b7UfAD4G/ARYRuMgsuEG1o02HW6QatDcRVQzG15NNWiRci3+LVQDEfuAx6imV+4/ynMN9S6q\nb7WPAv8LuLBupP9Y9QeBfxkRTw/5OWGMOvXbtcfrqb7V9wF/B/ztKMfWtzOHqs87qfp/GHDBKM+/\ngWrU/12Z2V9XfjLwd2VmxhXAGbXr+6VPrx2lP8P9G9WXXUo1iPX75efeUkZmJtV76OtUycE3h+ln\nR9a5UHdoe/ZZrhneC2zLzH8VEfOpvs28gmr08Kq6KUsXUH1DGQDOycxNpfwEqmlOBwC3Zea5pXwO\n1X8ex1P9B7k6Mx8u+9ZQDegCuHTIFDBJkjSCyThzcC6wlRez3POBzZkZwF3lMRGxDFhN9Y1qJdXU\nq1rychXVnN+lVCOHV5bytcCuUn4F5ZpiSUAupJoatBy4KCIObmsvJUmaJtqaHJRpVW8DrubFsxSn\nArUpVtcDp5ft04ANmbknM3uBB6lGXC8E5mXmlnLc+ro69W3dzIvTn04GNmXmE+WsxGaqhEOSJI2h\n3WcOrqC6xlo/gOfwzKxNW9oOHF62F9E4Gnob1WjvoeV9vDgK/AjgEYDMfJ5qnveho7QlSZLG0Lbk\noCz+sSMzv8sIYxvKcqUOqpEkaQpp51TG3wBOjYi3UQ0kfFlE3ABsj4iXZ+bj5ZLBjnJ8H3Xztqmm\nIW0r5UcOU16rcxTwaETMAg7KzF1lytiJdXUWM8a643v2PD84a9bMcXRTkqR904wZM4b98t625CAz\nP051MxUi4g3A72XmmRHxX6nmeF9Wft9SqtwK3BgRl1NdAlgKbMnMwYh4qtwgZQtwJnBlXZ01wLeo\nVlC7q5RvoloL/mCqsxYnAeeNFu/u3c+Otlsj6OmZx86dIy5Up0nga9B5vgad52vQWpO5zkHt8sFn\ngJMiIoHfKo8pd1nbSDWz4XaqO6jV6pxNNajxAeDBzLyjlF8DHBoRD1AtYHJ+aaufalGYe6gSiotr\n0yUlSdLoXGWr2LHjKcc+jIPZeuf5GnSer0Hn+RqMz4IFLxs2D3CFREmS1MDkQJIkNTA5kCRJDUwO\nJElSA5MDSZLUwORAkiQ1MDmQJEkNTA4kSVIDkwNJktTA5ECSJDUwOZAkSQ1MDiRJUoO23bJZkrR3\nBgYG6O19aEJtLFnySmbOnNmiiNStTA4kaYro7X2IVV+9k9kLFo6r/p4dj7HxlJM55pilLY5M3cbk\nQJKmkNkLFjJn0ZGdDkNdzjEHkiSpgcmBJElq4GUFSZqAgYEBMpP+/mfG3YaDCDXVmBxI0gQ4iFDT\nkcmBJE2Qgwg13TjmQJIkNTA5kCRJDUwOJElSA5MDSZLUwORAkiQ1MDmQJEkNTA4kSVIDkwNJktTA\n5ECSJDUwOZAkSQ3atnxyRBwA/BUwB9gf+HJmXhARnwTeD+wsh348M28vdS4AzgIGgHMyc1MpPwG4\nDjgAuC0zzy3lc4D1wPHALmB1Zj5c9q0BPlGe49LMXN+uvkratwwMDNDb+9C463ujJE13bUsOMvNn\nEfHGzHw2ImYBfx0RrwMGgcsz8/L64yNiGbAaWAYcAXw9IpZm5iBwFbA2M7dExG0RsTIz7wDWArsy\nc2lErAYuA86IiPnAhcAJpfn7IuLWzHyiXf2VtO+YyM2SvFGSukFbb7yUmc+Wzf2BmcDu8njGMIef\nBmzIzD1Ab0Q8CKyIiIeBeZm5pRy3HjgduAM4FbiolN8MfLZsnwxsqiUDEbEZWAl8sVV9k7Rv82ZJ\n0sjaOuYgIvaLiO8B24G/zMy/L7s+EhH3R8Q1EXFwKVsEbKurvo3qDMLQ8r5STvn9CEBmPg88GRGH\njtKWJEkaQ7vPHPwCOC4iDgLujIgTqS4RfKoccgnwh1SXBzrqkENewqxZXkMcj56eeZ0Ooev5Guyd\n3bvnTqj+/PlzX/g3n2hb9e21sq1u1K39boe2Jgc1mflkRHwN+PXM/EatPCKuBr5SHvYBi+uqHUn1\njb+vbA8tr9U5Cni0jGs4KDN3RUQfcGJdncXA3aPFuHv3s6Pt1gh6euaxc+fTnQ6jq/ka7L3+/mcm\nXL/2bz7Rturba2Vb3ca/g9Zq22WFiDisdskgIg4ETgK+GxEvrzvsHcAPyvatVIMJ94+Io4GlwJbM\nfBx4KiJWRMQM4Ezgy3V11pTtdwJ3le1NwFsi4uCIOKQ8951t6agkSdNMO88cLASuj4j9qJKQGzLz\nrohYHxHHUc1a+BHwQYDM3BoRG4GtwPPA2WWmAsDZVFMZD6SaynhHKb8GuCEiHqCaynhGaas/Ii4B\n7inHXexMBUmSmtPOqYw/oFp/YGj5+0apsw5YN0z5fcCxw5Q/B6waoa1rgWv3ImRJkqasia7PAc2v\n0TEpYw4kSdLETGR9Dti7NTpMDiRpmnIlyOlnstbnMDmQpGnKlSA1XiYHkjSNuRKkxsO7MkqSpAae\nOZAkqU321XEfJgeSJLXJvjruw+RAkqQ22hfHfTjmQJIkNTA5kCRJDUwOJElSA5MDSZLUwORAkiQ1\nMDmQJEkNTA4kSVIDkwNJktTA5ECSJDVwhURJwMTXgIfOrQMvqbVMDiQBE1sDHtq7DryJizS5TA4k\nvWCqrgE/lRMXaToyOZC0T5iqiYs0HZkcSJJUeAmrYnIgSVLhJayKyYEkSXW8hOU6B5IkaQjPHEiS\n9mkDAwNkJv39z4y7jekwTqCVTA4kSfs0xwm0nsmBJGmf5ziB1nLMgSRJatC2MwcRcQDwV8AcYH/g\ny5l5QUTMB24CXgH0Aqsy84lS5wLgLGAAOCczN5XyE4DrgAOA2zLz3FI+B1gPHA/sAlZn5sNl3xrg\nEyWcSzNzfbv6KknSdNK2MweZ+TPgjZl5HPDPgTdGxOuA84HNmRnAXeUxEbEMWA0sA1YCn4uIGaW5\nq4C1mbkUWBoRK0v5WmBXKb8CuKy0NR+4EFhefi6KiIPb1VdJkqaTtl5WyMxny+b+wExgN3AqcH0p\nvx44vWyfBmzIzD2Z2Qs8CKyIiIXAvMzcUo5bX1envq2bgTeV7ZOBTZn5RDkrsZkq4ZAkSWNo64DE\niNgP+A5wDHBVZv59RByemdvLIduBw8v2IuBbddW3AUcAe8p2TV8pp/x+BCAzn4+IJyPi0NLWtmHa\nkjRJJroMrVPLpM5pa3KQmb8AjouIg4A7I+KNQ/YPRsRgO2OQ1BkTmV7m1DKpsyZlKmNmPhkRXwNO\nALZHxMsz8/FyyWBHOawPWFxX7Uiqb/x9ZXtoea3OUcCjETELOCgzd0VEH3BiXZ3FwN2jxXjIIS9h\n1iy/pYxHT8+8TofQ9VrxGuzePXfCbcyfP/eFWHbvnjuh6WVD22p1bFOlrfr2pnI/p7JueQ1aHdto\n2jlb4TDg+cx8IiIOBE4CLgZuBdZQDR5cA9xSqtwK3BgRl1NdAlgKbClnF56KiBXAFuBM4Mq6Omuo\nLke8k2qAI8AmYF0ZhDijPPd5o8W7e/ezo+3WCHp65rFz59OdDqOrteo1mMjqcvVt1GKZaHutbGtf\niW0q93Mq65bXoNWxjaadAxIXAndHxPeAbwNfycy7gM8AJ0VEAr9VHpOZW4GNwFbgduDszKxdcjgb\nuBp4AHgwM+8o5dcAh0bEA8BvU2Y+ZGY/cAlwD1VCcXFtuqQkSRpd284cZOYPqNYfGFreD7x5hDrr\ngHXDlN8HHDtM+XPAqhHauha4du+iliRJrpAoSZIamBxIkqQGJgeSJKmByYEkSWpgciBJkhqYHEiS\npAYmB5IkqYHJgSRJamByIEmSGpgcSJKkBiYHkiSpgcmBJElqYHIgSZIatO2ujJIkjWRgYIDe3ofG\nXX/Jklcyc+bMFkakeiYHkqRJ19v7EKu+eiezFyzc67p7djzGxlNO5phjlrYhMoHJgSSpQ2YvWMic\nRUd2OgwNwzEHkiSpgcmBJElqYHIgSZIamBxIkqQGJgeSJKmByYEkSWpgciBJkhqYHEiSpAYmB5Ik\nqcGYKyRGxOXAxcA/An8JnAB8MDNvaHNskqQpYqL3QgDvh7AvaWb55Ddn5kcj4u1AH3AGcBtgciBJ\nXWIi90IA74ewr9mbeyu8AfhSZvZFxC/aFZAkaWryXgjdo5nkYEdE/AnwVuDTETEb8LyQNAUMDAyQ\nmfT3PzOu+p7mlTScZpKDdwPvAa7LzN0RsQS4opnGI2IxsB5YAAwCn8/MKyPik8D7gZ3l0I9n5u2l\nzgXAWcAAcE5mbirlJwDXAQcAt2XmuaV8TnmO44FdwOrMfLjsWwN8ojzHpZm5vpm4pX2Ft72V1A5j\nJgeZuYO6ZCAze4Frm2x/D/A7mfm9iJgL3BcRm6kShcsz8/L6gyNiGbAaWAYcAXw9IpZm5iBwFbA2\nM7dExG0RsTIz7wDWArsyc2lErAYuA86IiPnAhVQDKCnPfWtmPtFk7NI+wVO9klptxOQgIu4Zpd5g\nZi4fq/HMfBx4vGw/ExE/pPrQB5gxTJXTgA2ZuQfojYgHgRUR8TAwLzO3lOPWA6cDdwCnAheV8puB\nz5btk4FNtWSgJCUrgS+OFbckSd1stDMHH2vlE5XLEa8GvgW8FvhIRLwPuBf43fIhvqjsr9lGlUzs\nKds1fbyYZBwBPAKQmc9HxJMRcWhpa9swbUmSpFGMmBxk5jda9STlksJfAOeWMwhXAZ8quy8B/pDq\n8oAkSeqw0S4r/NfM/E8R8efD7B7MzFXNPEGZ3XAz8D8z8xZ4YRxDbf/VwFfKwz5gcV31I6m+8feV\n7aHltTpHAY9GxCzgoMzcFRF9wIl1dRYDd48U5yGHvIRZsxy1PR49PfM6HULX2r177oTqz58/94XX\nb6Jttbq9boytW/o5lWPrln6OZbTLCt8sv782zL7BZoKIiBnANcDWzPyjuvKFmflYefgO4Adl+1bg\nxrIq4xHAUmBLZg5GxFMRsQLYApwJXFlXZw3V5Yh3AneV8k3Auog4mGp8w0nAeSPFunv3s810SUP0\n9Mxj586nOx1G1xrvFMb6+rXXb6Jttbq9boytW/o5lWPrln6OZbTLCl8pv6+bQByvBd4LfD8ivlvK\nPg68KyKOo0oyfgR8sDzX1ojYCGwFngfOLjMVAM6mmsp4INVUxjtK+TXADRHxANVUxjNKW/0RcQlQ\nG1h5sTMVJEkaWzP3VphNte7Aq6g+mAcBMvOssepm5l8z/M2dbh+lzjpg3TDl9wHHDlP+HDDsJY7M\nvJbmp11KkiSauyvjnwK/AfwrIIHlwE/bGZQkSeqcZpKD5cC/A3Zn5qepLhX8s3YGJUmSOqeZ5OCn\n5br/QES8NDOfpFoOWZIkTUPN3FthV1mK+A7g9oj4CY2LC0mSpGmkmTMHb8/MfuD3gf9BtVbAv2lr\nVJIkqWOaufHSQN3vG9oekSRJ6qhmpjK+nmpq4a/UHT+YmY47kCRpGmpmzMEXqBYu+g4w0N5wJElS\npzWTHOzOzOHuryBJkqahZpKDmyLiQ8BNwM9qhZnpzQgkSZqGmpmt8DjVLZV/AjxTfrzTjiRJ01Qz\nZw4+Dfwm8N3azAVJkjR9NZMcbMvMe9seiSRJmhKaSQ7uiojL+OUxB1vbFpUkSeqYZpKDM6lu0zz0\ntshHtz4cSZLUac2skLhkEuKQJElTxIizFSLiNXXbhw3Zd0o7g5IkSZ0z2lTGq+q2Nw/Zd0kbYpEk\nSVNAM+scSJKkLmJyIEmSGow2IPFlEfE2YEbdNrXHbY9MkiR1xGjJwSPAx4bZBvhx2yKSJEkdNWJy\nkJknTmIckiRpinDMgSRJamByIEmSGpgcSJKkBqOtkLih/P7tyQtHkiR12mhnDo4tv9dMRiCSJGlq\nGG0q4z0R8STwkojYOWTfYGYuaGNck+4f/uGBCdVfsuSVzJw5s0XRSJLUOaNNZfz3EXE+8JfAW6kW\nP5q2Vn31TmYvWDiuunt2PMbGU07mmGOWtjgqSZIm36i3bM7M7RGxIjOfHk/jEbEYWA8sAAaBz2fm\nlRExH7gJeAXQC6zKzCdKnQuAs4AB4JzM3FTKTwCuAw4AbsvMc0v5nPIcxwO7gNWZ+XDZtwb4RAnn\n0sxcP1KssxcsZM6iI8fTTUmSppVmZiscEBFfjIiflJ8bI6Knyfb3AL+Tmb8GvAb4cET8KnA+sDkz\nA7irPCYilgGrgWXASuBzEVE7Y3EVsDYzlwJLI2JlKV8L7CrlVwCXlbbmAxcCy8vPRRFxcJNxS5LU\ntZpJDv4USOBVwHHAA6VsTJn5eGZ+r2w/A/wQOAI4Fbi+HHY9cHrZPg3YkJl7MrMXeBBYERELgXmZ\nuaUct76uTn1bNwNvKtsnA5sy84lyVmIzVcIhSZJGMeplheKYzPzXdY8vioj79/aJImIJ8Grg28Dh\nmbm97NoOHF62FwHfqqu2jSqZ2FO2a/pKOeX3IwCZ+XxEPBkRh5a2tg3TliRJGkUzycGMiHjhwzwi\nDmcvBydGxFyqb/XnZubTEfHCvswcjIjBvWlvKpo/fy49PfM6HUZHdGu/p4Ldu+dOqH79+3aibbW6\nvW6MrVv6OZVj65Z+jqWZ5OAPgO9ExNeokoK3UcYINCMiZlMlBjdk5i2leHtEvDwzHy+XDHaU8j5g\ncV31I6m+8feV7aHltTpHAY9GxCzgoMzcFRF9wIl1dRYDdzcb997q73+GnTvHNW5zn9bTM68r+z1V\n9Pc/M+H6tddvom21ur1ujK1b+jmVY+uWfo5lzDEHZYT/ycAPgPuBt2TmDc0EUQYTXgNszcw/qtt1\nKy8urrQGuKWu/IyI2D8ijgaWAlsy83HgqYhYUdo8E/jyMG29k2qAI8Am4C0RcXBEHAKcBNzZTNyS\nJHWzZs4ckJl/B/zdONp/LfBe4PsR8d1SdgHwGWBjRKylTGUsz7M1IjYCW4HngbMzs3bJ4WyqqYwH\nUk1lvKOUXwPcEBEPUE1lPKO01R8RlwD3lOMurk2XlCRJI2sqORivzPxrRj478eYR6qwD1g1Tfh8v\nLulcX/4cJbkYZt+1wLXNxitJkrwroyRJGmLU5CAi9ouIt01WMJIkqfNGTQ4y8xfAf5mkWCRJ0hTQ\nzGWF70bEirZHIkmSpoRmBiT+OvC+MhugNslyMDOXty8sSZLUKc0kB+e0PQpJkjRljJkcZOY3ACKi\nJzN3tj0iSZLUUWOOOSirEj4MfLc8/hcR8fm2RyZJkjqimQGJV1DdT2EnQGbeA7yunUFJkqTOaSY5\n2D8z/35I2c/bEYwkSeq8ZpKDn0XEC/d3jIhlwE/bF5IkSeqkZmYrrKO6m+GiiLgeWEl1MyVJkjQN\nNTNb4baI+H9Ut20GuDQzH2hvWNL0NTAwQG/vQ+Ouv2TJK5k5c2YLI5KkRs3elfHHwDeBQeBH7QtH\nmv56ex9i1VfvZPaChXtdd8+Ox9h4yskcc8zSNkQmSZUxk4OIeB2wgRfHGRwQEe/KzL9pa2TSNDZ7\nwULmLDqy02FI0rCaGZD4OeC9mRmZGcB7SpkkSZqGmkkOBjPzr2oPMvObbYxHkiR1WDPJweaIeC9A\nRMyIiPcAm9obliRJ6pQRxxxERP19FD4aEVeX7f2BXcDH2hmYJEnqjNEGJP6LSYtCkiRNGSMmB5nZ\nO4lxSJKkKaKZqYyvp1ol8Vfqjh/MzAXtDEySJHVGM4sgfQH4OPAdYKC94UiSpE5rJjnYnZl/3vZI\nJEnSlNBMcnBTRHwIuAn4Wa0wM59tW1SSJKljmlnn4HHgD4GfAM+Un6fbGZQkSeqcZs4cfBr4TeC7\nmemYA0mSprlmkoNtmXlv2yORJElTQjPJwV0RcRm/POZga9uikiRJHdNMcnAmMAisGlJ+9FgVI+IL\nwNuBHZl5bCn7JPB+oLY888cz8/ay7wLgLKopk+dk5qZSfgJwHXAAcFtmnlvK5wDrgeOplnRenZkP\nl31rgE+U57g0M9c30VdJkrremMlBZi6ZQPvXAn9M9QFeMwhcnpmX1x8YEcuA1cAy4Ajg6xGxNDMH\ngauAtZm5JSJui4iVmXkHsBbYlZlLI2I1cBlwRkTMBy4ETijN3xcRt2bmExPoiyRJXaGZFRKXDVfe\nzGWFzPxmRCwZZteMYcpOAzZk5h6gNyIeBFZExMPAvMzcUo5bD5wO3AGcClxUym8GPlu2TwY21ZKB\niNgMrAS+OFbMkiR1u2YuK9xG9W0fqtP6hwMP08RlhVF8JCLeB9wL/G75EF8EfKvumG1UZxD2lO2a\nvlJO+f0IQGY+HxFPRsShpa1tw7QlSZLGsNeXFSLiTcDbJvCcVwGfKtuXUK2hsHYC7UmSpBZq5sxB\ng8y8KyL+YLxPmJk7atsRcTXwlfKwD1hcd+iRVN/4+8r20PJanaOARyNiFnBQZu6KiD7gxLo6i4G7\nxxtzM+bPn0tPz7x2PsWU1a39Hq/du+dOqH79e20qtWVsE2+vW/o5lWPrln6OZW/HHOwHLAf2H29g\nEbEwMx8rD98B/KBs3wrcGBGXU10CWApsyczBiHgqIlYAW6hmT1xZV2cN1eWIdwJ3lfJNwLqIOJhq\nfMNJwHnjjbkZ/f3PsHNn9y0c2dMzryv7PRH9/c9MuH7t33wqtWVsE2+vW/o5lWPrln6OZW/HHDwP\nPEj1gTymiNgAvAE4LCIeoRo8eGJEHFfa/BHwQagGOEbERmBreZ6zy0wFgLOppjIeSDWV8Y5Sfg1w\nQ0Q8QDWV8YzSVn9EXALcU4672JkKkiQ1p61TGTPzXcMUf2GU49cB64Ypvw84dpjy5/jl9Rdq+66l\nmkopSZL2wojJwUhTGGtcIVGSpOlptDMH9ZcT6s0DDgFmtiUiSZLUUSMmB8NMYXwp8LvAh4HLh6sj\nSZL2fc3MVphFNSDwPKqzCcdnZl+7A5MkSZ0x2piDGVTTBi8C7gPemJk5WYFJkqTOGO3MwfeBlwIX\nUy1zPKt+kKIDEiVJmp5GSw7mUQ1I/OQI+ydybwVJkjRFNT0gUZIkdYf9Oh2AJEmaWkwOJElSA5MD\nSZLUwOQnt/P6AAALPElEQVRAkiQ1MDmQJEkNTA4kSVIDkwNJktTA5ECSJDUwOZAkSQ1MDiRJUgOT\nA0mS1MDkQJIkNRjtrowap4GBAXp7Hxp3/SVLXsnMmTNbGJEkSc0zOWiD3t6HWPXVO5m9YOFe192z\n4zE2nnIyxxyztA2RSZI0NpODNpm9YCFzFh3Z6TAkSdprjjmQJEkNTA4kSVIDkwNJktTA5ECSJDUw\nOZAkSQ1MDiRJUoO2TmWMiC8Abwd2ZOaxpWw+cBPwCqAXWJWZT5R9FwBnAQPAOZm5qZSfAFwHHADc\nlpnnlvI5wHrgeGAXsDozHy771gCfKKFcmpnr29lXSZKmi3afObgWWDmk7Hxgc2YGcFd5TEQsA1YD\ny0qdz0XEjFLnKmBtZi4FlkZErc21wK5SfgVwWWlrPnAhsLz8XBQRB7eni5IkTS9tTQ4y85vA7iHF\npwLXl+3rgdPL9mnAhszck5m9wIPAiohYCMzLzC3luPV1derbuhl4U9k+GdiUmU+UsxKb+eUkRZIk\nDaMTYw4Oz8ztZXs7cHjZXgRsqztuG3DEMOV9pZzy+xGAzHweeDIiDh2lLUmSNIaODkjMzEFgsJMx\nSJKkRp24t8L2iHh5Zj5eLhnsKOV9wOK6446k+sbfV7aHltfqHAU8GhGzgIMyc1dE9AEn1tVZDNzd\n8p7UmT9/Lj098wDYvXtuy9raF+xLsU4FrXx/TKW2jG3i7XVLP6dybN3Sz7F0Ijm4FVhDNXhwDXBL\nXfmNEXE51SWApcCWzByMiKciYgWwBTgTuHJIW98C3kk1wBFgE7CuDEKcAZwEnNfOTvX3P8POnU+/\nsN2qtqa6np55+0ys4zXRW3BD4224W/n+mEptGdvE2+uWfk7l2Lqln2Np91TGDcAbgMMi4hGqGQSf\nATZGxFrKVEaAzNwaERuBrcDzwNnlsgPA2VRTGQ+kmsp4Rym/BrghIh6gmsp4RmmrPyIuAe4px11c\nmy4p7a2J3IIbvA23pH1PW5ODzHzXCLvePMLx64B1w5TfBxw7TPlzlORimH3XUk2llCbMW3BL6iau\nkChJkhqYHEiSpAYmB5IkqYHJgSRJamByIEmSGpgcSJKkBiYHkiSpgcmBJElqYHIgSZIamBxIkqQG\nJgeSJKmByYEkSWpgciBJkhqYHEiSpAYmB5IkqYHJgSRJamByIEmSGpgcSJKkBiYHkiSpgcmBJElq\nYHIgSZIamBxIkqQGJgeSJKmByYEkSWpgciBJkhqYHEiSpAazOh2ARjcwMEBv70MTamPJklcyc+bM\nFkUkSZruTA6muN7eh1j11TuZvWDhuOrv2fEYG085mWOOWdriyCRJ05XJwT5g9oKFzFl0ZKfDkCR1\nCcccSJKkBh07cxARvcBTwACwJzOXR8R84CbgFUAvsCoznyjHXwCcVY4/JzM3lfITgOuAA4DbMvPc\nUj4HWA8cD+wCVmfmw5PVP0mS9lWdPHMwCJyYma/OzOWl7Hxgc2YGcFd5TEQsA1YDy4CVwOciYkap\ncxWwNjOXAksjYmUpXwvsKuVXAJdNRqckSdrXdfqywowhj08Fri/b1wOnl+3TgA2ZuScze4EHgRUR\nsRCYl5lbynHr6+rUt3Uz8KbWhy9J0vTT6TMHX4+IeyPiA6Xs8MzcXra3A4eX7UXAtrq624Ajhinv\nK+WU348AZObzwJPlsoUkSRpFJ2crvDYzH4uIHmBzRPzf+p2ZORgRgx2Kba/Nnz+Xnp55AOzePXfK\ntDW0vXZoZ9tTQatfg6n0/mjne83Y9r69bunnVI6tW/o5lo4lB5n5WPm9MyK+BCwHtkfEyzPz8XLJ\nYEc5vA9YXFf9SKozBn1le2h5rc5RwKMRMQs4KDP729Wf/v5n2Lnz6Re2p0pbQ9trtZ6eeW1re6po\n9Wswld4f7XyvGdvet9ct/ZzKsXVLP8fSkcsKEfGSiJhXtl8KvAX4AXArsKYctga4pWzfCpwREftH\nxNHAUmBLZj4OPBURK8oAxTOBL9fVqbX1TqoBjpIkaQydOnNwOPCliKjF8GeZuSki7gU2RsRaylRG\ngMzcGhEbga3A88DZmVm75HA21VTGA6mmMt5Ryq8BboiIB6imMp4xGR2TJGlf15HkIDN/BBw3THk/\n8OYR6qwD1g1Tfh9w7DDlz1GSC3Wfid6TwvtRSOpmLp+saWki96TwfhSSup3JgaYt70khSePT6UWQ\nJEnSFGNyIEmSGpgcSJKkBiYHkiSpgcmBJElqYHIgSZIaOJWxi0x0YSBwcSBJ6gYmB11kIgsDgYsD\nSVK3MDnoMi4MJEkai2MOJElSA5MDSZLUwORAkiQ1cMyBpgRnUkjS1GFyoCnBmRSSNHWYHGjcBgYG\nyEz6+58ZV/2h3/SdSSFJU4PJgcZtIt/2/aYvSVOXyYEmxG/7kjT9OFtBkiQ1MDmQJEkNTA4kSVID\nkwNJktTA5ECSJDUwOZAkSQ1MDiRJUgOTA0mS1MDkQJIkNTA5kCRJDab18skRsRL4I2AmcHVmXtbh\nkCRJmvKm7ZmDiJgJfBZYCSwD3hURv9rZqCRJmvqmbXIALAcezMzezNwDfBE4rcMxSZI05U3n5OAI\n4JG6x9tKmSRJGsV0HnMwuDcH79nx2LifqKr7z1vSXivbGq69bunnRNrrlti6pZ+tbs+/0fHZl2Lr\nln52pYh4TUTcUff4gog4r5MxSZK0L5jOZw7uBZZGxBLgUWA18K6ORiRJ0j5g2o45yMzngf8I3Als\nBW7KzB92NipJkiRJkiRJkiRJkiRJkqanGZ0OQPuuiOgFngIGgD2ZubyzEU1/EfEF4O3Ajsw8tpTN\nB24CXgH0Aqsy84mOBTnNjfAafBJ4P7CzHHZBZt4xfAuaiIhYDKwHFlCtZ/P5zLzSv4PWmrazFTQp\nBoETM/PVJgaT5lqq+4XUOx/YnJkB3FUeq32Gew0GgcvL38KrTQzaag/wO5n5a8BrgA+X++b4d9BC\nJgeaKM8+TaLM/Cawe0jxqcD1Zft64PRJDarLjPAagH8LkyIzH8/M75XtZ4AfUi2N799BC5kcaCIG\nga9HxL0R8YFOB9PFDs/M7WV7O3B4J4PpYh+JiPsj4pqIOLjTwXSDssjdq4Fv499BS5kcaCJem5mv\nBt5KdWrv9Z0OqNtl5iB7eV8RtcRVwNHAccBjwB92NpzpLyLmAjcD52bm0/X7/DuYOJMDjVtmPlZ+\n7wS+RHWbbE2+7RHxcoCIWAjs6HA8XSczd2TmYPlQuhr/FtoqImZTJQY3ZOYtpdi/gxYyOdC4RMRL\nImJe2X4p8BbgB52NqmvdCqwp22uAW0Y5Vm1QPoxq3oF/C20TETOAa4CtmflHdbv8O2ghB9BoXCLi\naKqzBVDdwOvPMvPTHQypK0TEBuANwGFU11UvBL4MbASOwilcbTfMa3ARcCLVJYVB4EfAB+uuf6uF\nIuJ1wP8Gvs+Llw4uALbg34EkSZIkSZIkSZIkSZIkSZIkSZIkSZIkdZ2I6I2IZUPK7o2IN3QqJkl7\nxxUSJbXacGvat2yt+4iY2Yp2JI1sVqcDkDQtDbv6akQcDvwJ8MpyzH/LzBvKvl8AczPz2aGPy/bF\nwNuB26lWJZTUJiYHklptBvAXEfGzurIo5VcC38/Md5Sb5NwXEfdl5tYm2n02M72hkTQJTA4ktdog\n8G/qP/Aj4p6y+SbgdwAy8/GIuA34LaCZ5OD6VgcqaXiOOZA02WYM2a6NRRig/J8UEQcMU++ZNscl\nqTA5kDSZvg58AKBcVngrcHfZ9yBQu2zw7skPTVKNyYGkyTIInAO8KiLuBzYB52XmD8v+jwJ/GhH3\nUt0OeXBIXUmSJEmSJEmSJEmSJEmSJEmSJEmSJEmSJEmSJEnT2f8HBeABQV177PQAAAAASUVORK5C\nYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10a7bd090>"
       ]
      }
     ],
     "prompt_number": 16
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Exploration 2: Subject-lines:\n",
      "---\n",
      "\n",
      "Running this map-reduce script."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# import mrjob\n",
      "\n",
      "def mapfn(k, v):\n",
      "  import re\n",
      "  import mailbox\n",
      "  pattern = \"^(\\[.+\\]|\\s|Re:|Fwd:)+\"\n",
      "\n",
      "  mbox = mailbox.mbox(v)\n",
      "  subjects = set()\n",
      "  for msg in mbox:\n",
      "    try:\n",
      "      subject = re.sub(\"\\W\", \"\", re.sub(pattern, \"\", msg['Subject'])).lower()\n",
      "      yield subject, 1\n",
      "    except:\n",
      "      yield None, 1\n",
      "\n",
      "def reducefn(k, vs):\n",
      "  return sum(vs)\n",
      "\n",
      "# if __name__ == '__main__':\n",
      "#   job = mrjob.MrJob(mapfn, reducefn, name=\"nsubjects\", test=False)\n",
      "#   print job.run()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 24
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Produced the following results (you can't actually execute because even the compressed file is too large for upload / running in an iptyhon environment). If you want you can go to this link via curl or wget http://208.43.122.13/nsubjects20141204005425.out and then run the code in your python console in terminal:"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "FILE_NAME = \"nsubjects20141204005425.out\"\n",
      "# results = sorted(pickle.load(open(FILE_NAME, 'r')).items(), key=lambda x: x[1])\n",
      "# results = results.reverse()\n",
      "\n",
      "# len(results) produces 1,745,282 unique subject lines\n",
      "# results[:100] produces:\n",
      "# ('', 41243)\n",
      "# ('nosubject', 39161)\n",
      "# ('thisistheonlythreadtowhichyoureallowedtopost', 17751)\n",
      "# ('tonight', 7973)\n",
      "# ('dinner', 6771)\n",
      "# ('stabilityrunonmrc35380', 6282)\n",
      "# ('meeting', 6018)\n",
      "# ('meetingtonight', 5626)\n",
      "# ('thankyou', 4751)\n",
      "# ('meetingtomorrow', 4626)\n",
      "# ('tomorrow', 4567)\n",
      "# ('question', 3909)\n",
      "# ('happybirthday', 3850)\n",
      "# ('happybirthdayto', 3433)\n",
      "# ('help', 3420)\n",
      "# ('swerror', 3356)\n",
      "# ('boardmeeting', 3186)\n",
      "# ('seniorsale', 3067)\n",
      "# ('housingday', 3040)\n",
      "# ('important', 2863)\n",
      "# ('dinnertonight', 2783)\n",
      "# ('party', 2671)\n",
      "# ('housing', 2578)\n",
      "# ('update', 2415)\n",
      "# ('postering', 2326)\n",
      "# ('congratulations', 2320)\n",
      "# ('meetingtoday', 2201)\n",
      "# ('reminder', 2179)\n",
      "# ('today', 2148)\n",
      "# ('minutes', 2117)\n",
      "# ('thanks', 2113)\n",
      "# ('wow', 1967)\n",
      "# ('spikecheckrunonmrc35380', 1964)\n",
      "# ('nowisagreattimetoinvestintheoverseaspropertybonanza', 1933)\n",
      "# ('practice', 1914)\n",
      "# ('hi', 1864)\n",
      "# ('hello', 1815)\n",
      "# ('updates', 1787)\n",
      "# ('test', 1764)\n",
      "# ('elections', 1731)\n",
      "# ('tshirts', 1699)\n",
      "# ('whosup', 1696)\n",
      "# ('omg', 1632)\n",
      "# ('secretsanta', 1579)\n",
      "# ('website', 1577)\n",
      "# ('movienight', 1556)\n",
      "# ('harvardyale', 1525)\n",
      "# ('retreat', 1513)\n",
      "# ('springbreak', 1504)\n",
      "# ('welcomeback', 1483)\n",
      "# ('practicetonight', 1469)\n",
      "# ('practicetoday', 1454)\n",
      "# ('congrats', 1402)\n",
      "# ('activitiesfair', 1395)\n",
      "# ('thisweek', 1392)\n",
      "# ('hey', 1378)\n",
      "# ('fyi', 1373)\n",
      "# ('brunch', 1369)\n",
      "# ('meetingthisweek', 1364)\n",
      "# ('thisweekend', 1363)\n",
      "# ('also', 1352)\n",
      "# ('practicetomorrow', 1334)\n",
      "# ('freshmen', 1333)\n",
      "# ('lol', 1331)\n",
      "# ('rehearsal', 1312)\n",
      "# ('studybreak', 1307)\n",
      "# ('lunch', 1303)\n",
      "# ('happythanksgiving', 1302)\n",
      "# ('meetingminutes', 1299)\n",
      "# ('merrychristmas', 1289)\n",
      "# ('rehearsaltomorrow', 1273)\n",
      "# ('urgent', 1267)\n",
      "# ('theneverendingthread', 1265)\n",
      "# ('posteringtomorrow', 1259)\n",
      "# ('primalscream', 1252)\n",
      "# ('thegame', 1250)\n",
      "# ('rehearsaltonight', 1247)\n",
      "# ('meetingnotes', 1247)\n",
      "# ('food', 1243)\n",
      "# ('posters', 1218)\n",
      "# ('happynewyear', 1202)\n",
      "# ('dinnertomorrow', 1200)\n",
      "# ('meetingtime', 1172)\n",
      "# ('pleaseread', 1148)\n",
      "# ('4amclub', 1110)\n",
      "# ('agenda', 1108)\n",
      "# ('riverrun', 1098)\n",
      "# ('yay', 1092)\n",
      "# ('so', 1065)\n",
      "# ('thoughts', 1063)\n",
      "# ('guys', 1061)\n",
      "# ('friday', 1056)\n",
      "# ('houses', 1053)\n",
      "# ('rfnoiserunonmrc35380', 1052)\n",
      "# ('prefrosh', 996)\n",
      "# ('tickets', 993)\n",
      "# ('partytonight', 989)\n",
      "# ('yo', 988)\n",
      "# ('heyguys', 963)\n",
      "# ('amazingnewmlmscriptsmassmailersdownlinebuilders', 934)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 31
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "plt.figure()\n",
      "plt.title(\"Percent of Emails Covered by Subjects\")\n",
      "plt.plot(np.arange(0,1.00,.01), [0.23473338279209063, 0.29710048035380265, 0.34266579826290494, 0.37972140629650236, 0.41137483124355706, 0.4389550974869488, 0.4639508201670958, 0.4859073094004707, 0.5062030123833802, 0.5250228602977015, 0.542597612752308, 0.558279920750968, 0.5739631273462358, 0.5882955635207159, 0.6008421287969301, 0.6133886940731443, 0.6259345404720723, 0.6384811057482865, 0.6489310654192708, 0.6583409893764315, 0.6677509133335922, 0.6771608372907528, 0.6865702220899488, 0.6959801460471094, 0.7053900700042701, 0.7147999939614308, 0.7215653481019483, 0.7278382713014123, 0.7341115539395194, 0.7403848365776265, 0.7466581192157337, 0.7529314018538408, 0.7592046844919479, 0.7654776076914118, 0.7717508903295189, 0.778024172967626, 0.7842974556057332, 0.7905707382438403, 0.7968436614433043, 0.8031169440814114, 0.8093902267195185, 0.8156635093576257, 0.821213242007163, 0.8243498833262164, 0.8274863449259484, 0.830622986245002, 0.8337596275640555, 0.8368962688831091, 0.8400329102021626, 0.8431695515212162, 0.8463060131209482, 0.8494426544400018, 0.8525792957590553, 0.8557159370781089, 0.8588525783971624, 0.8619890399968945, 0.865125681315948, 0.8682623226350016, 0.8713989639540551, 0.8745356052731087, 0.8776722465921623, 0.8808087081918943, 0.8839453495109478, 0.8870819908300014, 0.890218632149055, 0.8933552734681085, 0.8964917350678404, 0.899628376386894, 0.9027650177059475, 0.9059016590250011, 0.9090383003440546, 0.9121749416631082, 0.9153114032628402, 0.9184480445818938, 0.9215846859009473, 0.9247213272200009, 0.9278579685390544, 0.9309944301387865, 0.93413107145784, 0.9372677127768936, 0.9404043540959471, 0.9435409954150007, 0.9466776367340543, 0.9498140983337863, 0.9529507396528398, 0.9560873809718934, 0.9592240222909469, 0.9623606636100005, 0.9654971252097324, 0.968633766528786, 0.9717704078478395, 0.9749070491668931, 0.9780436904859466, 0.9811803318050002, 0.9843167934047322, 0.9874534347237858, 0.9905900760428393, 0.9937267173618929, 0.9968633586809464, 1.0])\n",
      "plt.xlabel(\"Top x percent most popular subject lines\")\n",
      "plt.ylabel(\"Percent of emails covered\")\n",
      "plt.show()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAYkAAAEcCAYAAAAydkhNAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10aac9350>"
       ]
      }
     ],
     "prompt_number": 36
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "This means that roughly the top 10% of popular subject-lines account for 50% of the emails sent over the Harvard mailing lists. That makes sense! Generally these types of coverage plots usually results in some sort of log-increase. \n",
      "\n",
      "Also, we see that beyond the 40% mark of popular email lines, the plot looks fairly linear. That makess sense too, when we run: \n",
      "> np.mean([1 if x[1] == 1 else 0 for x in subjects])\n",
      "\n",
      "It produces $0.57230693950891609$. This means that roughly 60% of subject lines are used once, receive no replies, and then die off; they compromise about 60% of all traffic over the mailing lists since 2001."
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Network Analysis"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We looked at the SCAS (Small Claims Advisory Service) club and the Premed-Exec lists for the years 2013 and 2014.\n",
      "Using [d3](http://d3js.org/), a Javascript library, we were able to visualize the networks created.\n",
      "Each person on the list is represented by a node, and the edges represent responses.\n",
      "An email where person A responds to person B corresponds to an edge from A to B.\n",
      "Each edge is also weighted based on the time between responses.\n",
      "We used $\\frac{1}{\\sqrt{t}}$ as a weighting factor, where $t$ represents the time taken to respond.\n",
      "Subsequent emails between the same two people just add to the edge weight.\n",
      "We then sized the nodes by the total weight of all edges entering and leaving that node.\n",
      "In other words, the largest nodes are those for people who both respond a lot and also get responded to.\n"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Below is the Python code we actually used to create the graph data structure.\n",
      "The graph ended up being an adjacency list.\n",
      "The top level was an dictionary, where the keys were email addresses.\n",
      "Each email address mapped to another dictionary, again keyed by email address, where the value was the weight for that edge.\n",
      "To execute this, call the function `make_graph` with a mailbox file and a year, and it'll create the graph for that mailbox and emails sent in the given year.\n",
      "\n",
      "Again, it is likely that the actual .mbox files are too large to open in iPython, so you may have to call these functions from the terminal."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import mailbox\n",
      "import re\n",
      "import time\n",
      "import datetime\n",
      "import email.utils\n",
      "import pdb\n",
      "import math\n",
      "import logging\n",
      "\n",
      "def make_graph(mbox, year):\n",
      "  if type(mbox) is str:\n",
      "    mbox = mailbox.mbox(mbox)\n",
      "\n",
      "  year = int(year)\n",
      "\n",
      "  pattern = \"^(\\[.+\\]|\\s|Re:|Fwd:)+\"\n",
      "\n",
      "  graph = {}\n",
      "  mailing_threads = {}\n",
      "\n",
      "  for msg in mbox:\n",
      "    try:\n",
      "      thread = re.sub(pattern, \"\", msg[\"Subject\"])\n",
      "    except:\n",
      "      logging.error(\"Could not parse subject \" + str(msg['Subject']))\n",
      "      continue\n",
      "\n",
      "    try:\n",
      "      A_time = datetime.datetime.fromtimestamp(time.mktime(email.utils.parsedate(msg['Date'])))\n",
      "      A_email = email.utils.parseaddr(msg['From'])[1]\n",
      "    except:\n",
      "      logging.error(\"Could not parse date or email: \" + str(msg['Date']) + \", \" + msg['From'])\n",
      "      continue\n",
      "\n",
      "    # remove emails with the wrong date\n",
      "    if A_time.year == year:\n",
      "      if thread not in mailing_threads:\n",
      "        mailing_threads[thread] = []\n",
      "\n",
      "      for B_email, B_time in mailing_threads[thread]:\n",
      "        if A_time > B_time:\n",
      "          greater_time, lesser_time = A_time, B_time\n",
      "          greater_email, lesser_email = A_email, B_email\n",
      "        else:\n",
      "          greater_time, lesser_time = B_time, A_time\n",
      "          greater_email, lesser_email = B_email, A_email\n",
      "\n",
      "        if greater_email not in graph:\n",
      "          graph[greater_email] = {}\n",
      "\n",
      "        if lesser_email not in graph[greater_email]:\n",
      "          graph[greater_email][lesser_email] = new_graph_weight(greater_time - lesser_time)\n",
      "        else:\n",
      "          graph[greater_email][lesser_email] = update_graph_weight(graph[greater_email][lesser_email], greater_time - lesser_time)\n",
      "\n",
      "      mailing_threads[thread].append((A_email, A_time))\n",
      "\n",
      "  return graph\n",
      "\n",
      "def new_graph_weight(timedelta):\n",
      "  diff = timedelta.total_seconds()\n",
      "  if diff < 1:\n",
      "    logging.error(\"Time difference is \" + str(diff))\n",
      "    diff = 1\n",
      "  return 1/math.sqrt(diff)\n",
      "\n",
      "def update_graph_weight(old_weight, new_timedelta):\n",
      "  return old_weight + new_graph_weight(new_timedelta)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "This next block of code is javascript, not python, used to visualize the graphs. It's not executable here, but we figured if you're interested, this is how we did it!"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "// adapted from http://bl.ocks.org/mbostock/1153292\n",
      "\n",
      "// create graph of given list file in div with that id\n",
      "function create_graph(list) {\n",
      "  var width = 600,\n",
      "      height = 350;\n",
      "\n",
      "  var svg = d3.select(\"#\" + list + \" .network-graph\").append(\"svg\")\n",
      "      .attr(\"width\", width)\n",
      "      .attr(\"height\", height);\n",
      "\n",
      "  // marker def for arrowhead\n",
      "  svg.append(\"defs\")\n",
      "    .append(\"marker\")\n",
      "      .attr(\"id\", \"arrow\")\n",
      "      .attr(\"viewBox\", \"0 -5 10 10\")\n",
      "      .attr(\"refX\", 15)\n",
      "      .attr(\"refY\", -1.5)\n",
      "      .attr(\"markerWidth\", 6)\n",
      "      .attr(\"markerHeight\", 6)\n",
      "      .attr(\"orient\", \"auto\")\n",
      "    .append(\"path\")\n",
      "      .attr(\"d\", \"M0,-5L10,0L0,5\");\n",
      "\n",
      "  d3.json(list + \".json\", function(error, data) {\n",
      "    var graph_data = data;\n",
      "    var graph_nodes = {};\n",
      "    var graph_links = [];\n",
      "\n",
      "    // loop through keys in graph_data\n",
      "    for (var source in graph_data) {\n",
      "      for (var target in graph_data[source]) {\n",
      "        if (target != source) {\n",
      "          var link = {};\n",
      "          link.source = graph_nodes[source] || (graph_nodes[source] = {name: source, totalDegree: 0});\n",
      "          link.target = graph_nodes[target] || (graph_nodes[target] = {name: target, totalDegree: 0});\n",
      "          link.time = graph_data[source][target];\n",
      "\n",
      "          graph_nodes[source].totalDegree += link.time;\n",
      "          graph_nodes[target].totalDegree += link.time;\n",
      "\n",
      "          graph_links.push(link);\n",
      "        }\n",
      "      }\n",
      "    }\n",
      "\n",
      "    var time_extent = d3.extent(graph_links, function(d) {\n",
      "      return d.time;\n",
      "    });\n",
      "\n",
      "    total_time_extent = d3.extent(d3.values(graph_nodes), function(d) {\n",
      "      return d.totalDegree;\n",
      "    })\n",
      "\n",
      "    // link distance scale (edge weight)\n",
      "    var edgeWeight = d3.scale.linear()\n",
      "      .domain(time_extent)\n",
      "      .range([100, 300]);\n",
      "\n",
      "    // opacity scale based on time val (edge weight)\n",
      "    var opacityScale = d3.scale.linear()\n",
      "        .domain(time_extent)\n",
      "        .range([0.7, 1]);\n",
      "\n",
      "    // color scale also based on time val (edge weight)\n",
      "    var colorScale = d3.scale.linear()\n",
      "        .domain(time_extent)\n",
      "        .range([d3.rgb(200, 200, 200), d3.rgb(0, 0, 0)]);\n",
      "\n",
      "    // node radius scale based on total time\n",
      "    var radiusScale = d3.scale.pow(0.5)\n",
      "        .domain(total_time_extent)\n",
      "        .range([6, 12]);\n",
      "\n",
      "    var force = d3.layout.force()\n",
      "        .nodes(d3.values(graph_nodes))\n",
      "        .links(graph_links)\n",
      "        .size([width, height])\n",
      "        .linkDistance(function(d) {\n",
      "          return edgeWeight(d.time)\n",
      "        })\n",
      "        .charge(-300)\n",
      "        .gravity(0.4)\n",
      "        .on(\"tick\", function(d) {\n",
      "            path.attr(\"d\", linkArc);\n",
      "            circle.attr(\"transform\", transform);\n",
      "            text.attr(\"transform\", transform);\n",
      "        })\n",
      "        .start();\n",
      "\n",
      "    var path = svg.append(\"g\").selectAll(\"path\")\n",
      "        .data(force.links())\n",
      "      .enter().append(\"path\")\n",
      "        .attr(\"class\", function(d) {\n",
      "          var className = \"link\"\n",
      "          if (d.target.name == d.source.name) {\n",
      "            className += \" selfloop \" + fixEmail(d.target.name);\n",
      "          } else {\n",
      "            className += \" \" + fixEmail(d.target.name) + \" \" + fixEmail(d.source.name);\n",
      "          }\n",
      "\n",
      "          return className;\n",
      "        })\n",
      "        .attr(\"marker-end\", function(d) {\n",
      "          if (d.target.name == d.source.name) {\n",
      "            return null;\n",
      "          } else {\n",
      "            return \"url(#arrow)\"; \n",
      "          }\n",
      "        })\n",
      "        .style(\"stroke\", function(d) {\n",
      "          return colorScale(d.time);\n",
      "        })\n",
      "        .style(\"opacity\", function(d) {\n",
      "          return opacityScale(d.time);\n",
      "        });\n",
      "\n",
      "    var circle = svg.append(\"g\").selectAll(\"circle\")\n",
      "        .data(force.nodes())\n",
      "      .enter().append(\"circle\")\n",
      "        .attr(\"class\", function(d) {\n",
      "          return fixEmail(d.name);\n",
      "        })\n",
      "        .attr(\"r\", function(d) {\n",
      "          return radiusScale(d.totalDegree);\n",
      "        })\n",
      "        .call(force.drag);\n",
      "\n",
      "    var text = svg.append(\"g\").selectAll(\"text\")\n",
      "        .data(force.nodes())\n",
      "      .enter().append(\"text\")\n",
      "        .attr(\"class\", function(d) {\n",
      "          return fixEmail(d.name);\n",
      "        })\n",
      "        .style(\"visibility\", \"hidden\")\n",
      "        .attr(\"x\", 8)\n",
      "        .attr(\"y\", \".31em\")\n",
      "        .text(function(d) { return d.name; });\n",
      "\n",
      "    // add mouseover handler so emails only appear on mouseover\n",
      "    svg.selectAll(\"circle\")\n",
      "        .on(\"mouseover\", function(d) {\n",
      "          // get email of this user\n",
      "          var email = d3.select(this).attr(\"class\");\n",
      "          // set corresponding text visible\n",
      "          svg.select(\"text.\" + email).style(\"visibility\", \"visible\");\n",
      "          // set all paths going into or out of this user to max opacity and color\n",
      "          svg.selectAll(\"path.\" + email).style(\"opacity\", 1);\n",
      "          svg.selectAll(\"path.\" + email).style(\"stroke\", \"red\");\n",
      "\n",
      "        })\n",
      "        .on(\"mouseout\", function(d) {\n",
      "          var email = d3.select(this).attr(\"class\");\n",
      "          svg.select(\"text.\" + email).style(\"visibility\", \"hidden\");\n",
      "          svg.selectAll(\"path.\" + email).style(\"opacity\", function(d) {\n",
      "            return opacityScale(d.time);\n",
      "          });\n",
      "          svg.selectAll(\"path.\" + email).style(\"stroke\", function(d) {\n",
      "            return colorScale(d.time);\n",
      "          });\n",
      "        });\n",
      "\n",
      "    // select Willy and make him red\n",
      "    if (willy = svg.select(\"circle.wxiao-college-harvard-edu\")) {\n",
      "      willy.style(\"fill\", \"red\");\n",
      "    }\n",
      "\n",
      "    // select Robbie and make him blue\n",
      "    if (robbie = svg.select(\"circle.rkgibson2-gmail-com\")) {\n",
      "      robbie.style(\"fill\", \"blue\");\n",
      "    }\n",
      "\n",
      "    // select Robbie and make him blue\n",
      "    if (robbie = svg.select(\"circle.huihuifan-college-harvard-edu\")) {\n",
      "      robbie.style(\"fill\", \"green\");\n",
      "    }\n",
      "\n",
      "    // find node with max degree, and color it purple\n",
      "    maxDegree = d3.values(graph_nodes).reduce(function(prevElt, currElt) {\n",
      "      return (!prevElt || currElt.totalDegree > prevElt.totalDegree) ? currElt : prevElt;\n",
      "    }, null);\n",
      "    if (maxNode = svg.select(\"circle.\" + fixEmail(maxDegree.name))) {\n",
      "      maxNode.style(\"fill\", \"purple\");\n",
      "    }\n",
      "  });\n",
      "}\n",
      "\n",
      "// Use elliptical arc path segments to doubly-encode directionality.\n",
      "function linkArc(d) {\n",
      "  // console.log(d);\n",
      "  var dx = d.target.x - d.source.x,\n",
      "      dy = d.target.y - d.source.y,\n",
      "      dr = Math.sqrt(dx * dx + dy * dy);\n",
      "  if (dr == 0) {\n",
      "    return \"M\" + d.source.x + \",\" + d.source.y + \"A\" + \"10,10 0 1,1 \" + d.target.x + \",\" + (d.target.y - 0.1);\n",
      "  }\n",
      "  return \"M\" + d.source.x + \",\" + d.source.y + \"A\" + dr + \",\" + dr + \" 0 0,1 \" + d.target.x + \",\" + d.target.y;\n",
      "}\n",
      "\n",
      "function transform(d) {\n",
      "  return \"translate(\" + d.x + \",\" + d.y + \")\";\n",
      "}\n",
      "\n",
      "// function to replace @ signs in emails with hyphens for CSS class names\n",
      "function fixEmail(email) {\n",
      "  return email.replace(/[@\\.]/g, \"-\");\n",
      "}\n",
      "\n",
      "create_graph(\"scas-board-2013\");\n",
      "create_graph(\"scas-board-2014\");\n",
      "create_graph(\"premed-exec-2013\");\n",
      "create_graph(\"premed-exec-2014\");"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "These networks are much more interesting to pull apart and play around with on the webpage submitted later. Because we can't embed javascript code, we could only include screenshots here. \n",
      "\n",
      "Below, the red node is Willy, one of the team members, and the green node is Angela Fan, one of the CS 109 TFs.\n",
      "The purple node is the node with the maximum total weight of all the nodes in the graph.\n",
      "For the premed-exec list, the largest node in 2013 is the president of the Premedical Society at the time, Vivian Leung, and the largest node in 2014 is Lily Zhang, the treasurer (the current president is also one of the largest nodes).\n",
      "For the scas-board list, Jessica Chao was the executive officer last year and Anita Lo is the executive officer this year.\n",
      "They are both the largest nodes for their respective years.\n",
      "You can also see how Willy took on a larger role in the group this year; his circle got correspondingly larger."
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "<img src=\"https://dl.dropboxusercontent.com/u/80118923/scas13.png\">\n",
      "<img src=\"https://dl.dropboxusercontent.com/u/80118923/scas14.png\">\n",
      "<img src=\"https://dl.dropboxusercontent.com/u/80118923/premed13.png\">\n",
      "<img src=\"https://dl.dropboxusercontent.com/u/80118923/premed14.png\">"
     ]
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Prediction"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We also implemented a prediction portion of this project that, given the text of an email, predicts whether or not that email will elicit a response from the email list in question. In order to implement this, we go through every email in a mailing list's mbox file and extract the text of each email with a binary indicator variable corresponding to whether or not anyone responded to the email in question. From this, we proceed to extract features from the text either as a bag-of-words or n-gram format (with n=2 to reduce excessive computation in this code). Once we have these features extracted, we split the data from the list into a training set and a test set in a 60-40 fashion. We also create label vectors corresponding to the true label of each of the emails in these two sets of data. With these features and label vectors, we then train two different classifiers: a random forest and a linear SVM model. With each of these, we run grid_search with ten-fold cross-validation over a reasonable set of parameters in order to optimize the classifier for the current mailing list data. The output of our training function is a pair of classifiers and the corresponding dictionary of n-grams or individual words, which we then use to predict whether or not people will respond to an email that the user can input. Overall, it turns out to be extremely hard to predict exactly what would make people respond to your emails because that depends so much on specific events going on with the email list, the person sending it, or random other factors that we can't access through our data. Also, it turns out that our data files are too large for ipython to handle, but you can run the code at the very bottom of this file in the Terminal in Python if you want to see how people respond to your email.\n",
      "\n",
      "One quick note about results: when running this, the test set accuracy results are as follows:\n",
      "\n",
      "Bag of words:\n",
      "  Random forest: 0.81995133819951338, \n",
      "  SVM: 0.73479318734793186\n",
      "    \n",
      "N-grams (n=2):\n",
      "  Random forest: 0.81386861313868608,\n",
      "  SVM: 0.71654501216545008"
     ]
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "Initial imports and helper functions for both classifiers"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import mailbox\n",
      "import re\n",
      "import email.utils\n",
      "import pdb\n",
      "import math\n",
      "import logging\n",
      "import collections\n",
      "import numpy as np\n",
      "import time\n",
      "import datetime\n",
      "import sklearn\n",
      "import random\n",
      "from sklearn.linear_model import LinearRegression\n",
      "from sklearn.ensemble import RandomForestClassifier\n",
      "from sklearn.cross_validation import cross_val_score\n",
      "from sklearn.svm import SVC"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def get_body(msg):\n",
      "    body = None\n",
      "    # step through until not multipart; take that part as text\n",
      "    if msg.is_multipart():\n",
      "        for part in msg.walk():\n",
      "            if part.is_multipart():\n",
      "                for subpart in part.walk():\n",
      "                    if subpart.get_content_type() == 'text/plain':\n",
      "                        body = subpart.get_payload(decode=True)\n",
      "            elif part.get_content_type() == 'text/plain':\n",
      "                body = part.get_payload(decode=True)\n",
      "    # if not multipart to begin with, take payload as text\n",
      "    elif msg.get_content_type() == 'text/plain':\n",
      "        body = msg.get_payload(decode=True)\n",
      "    return body\n",
      "\n",
      "def count_exclamations(body):\n",
      "    return body.count('!')\n",
      "\n",
      "def strip_punctuation(body):\n",
      "    return re.sub(r'[^\\w\\s]', '', body)\n",
      "\n",
      "def delete_nums(body):\n",
      "    return re.sub(' \\d+', '', body)\n",
      "\n",
      "def replace_newlines(body):\n",
      "    return body.replace('\\n',' ')\n",
      "\n",
      "def process_ngrams(words, n):\n",
      "    ngrams = zip(*[words[i:] for i in range(n)])\n",
      "    return [''.join(x) for x in ngrams]"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 2
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def find_time(mbox):\n",
      "\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    pattern = \"^(\\[.+\\]|\\s|Re:|Fwd:)+\"\n",
      "\n",
      "    mailing_threads = {}\n",
      "    emails = []\n",
      "\n",
      "    for msg in mbox:\n",
      "        emails.append(0)\n",
      "\n",
      "        try:\n",
      "            thread = re.sub(pattern, \"\", msg[\"Subject\"])\n",
      "        except:\n",
      "            logging.error(\"Could not parse subject \" + str(msg['Subject']))\n",
      "            continue\n",
      "\n",
      "        try:\n",
      "            email_time = datetime.datetime.fromtimestamp(time.mktime(email.utils.parsedate(msg['Date'])))\n",
      "        except:\n",
      "            logging.error(\"Could not parse date or email: \" + str(msg['Date']) + \", \" + msg['From'])\n",
      "            continue\n",
      "\n",
      "        if thread not in mailing_threads:\n",
      "            mailing_threads[thread] = []\n",
      "        else:\n",
      "            previous_email = mailing_threads[thread][len(mailing_threads[thread]) - 1]\n",
      "            try:\n",
      "                emails[previous_email[0]] = (email_time - previous_email[1]).total_seconds()\n",
      "            except:\n",
      "                logging.error(\"Issue updating email time\")\n",
      "                continue\n",
      "\n",
      "        mailing_threads[thread].append((len(emails) - 1, email_time))\n",
      "\n",
      "    return emails"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def split_by_response(mbox):\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    pattern = \"^(\\[.+\\]|\\s|Re:|Fwd:)+\"\n",
      "    responded = []\n",
      "    no_response = []\n",
      "    mailing_threads = {}\n",
      "\n",
      "    for msg in mbox:\n",
      "        try:\n",
      "            thread = re.sub(pattern, \"\", msg[\"Subject\"])\n",
      "        except:\n",
      "            logging.error(\"Could not parse subject \" + str(msg['Subject']))\n",
      "            continue\n",
      "\n",
      "        try:\n",
      "            A_email = email.utils.parseaddr(msg['From'])[1]\n",
      "        except:\n",
      "            logging.error(\"Could not parse date or email: \" + str(msg['Date']) + \", \" + msg['From'])\n",
      "            continue\n",
      "\n",
      "        if thread not in mailing_threads:\n",
      "            mailing_threads[thread] = []\n",
      "\n",
      "        mailing_threads[thread].append(msg)\n",
      "\n",
      "    for i in mailing_threads:\n",
      "        if len(mailing_threads[i]) > 1:\n",
      "            responded.append(mailing_threads[i])\n",
      "        else:\n",
      "            no_response.append(mailing_threads[i])\n",
      "\n",
      "    responded = sum(responded, [])\n",
      "    no_response = sum(no_response, [])\n",
      "\n",
      "    split = (responded, no_response)\n",
      "\n",
      "    return split"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 10
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "Bag-of-words classifier"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Bag-of-words classifier. Trained a random forest and an SVM on these features in order to predict whether or not an email you send to a list will get a response. The method 'evaluate_classifiers' returns a label based on each of the two classifiers given the text of an email you would like to send to the desired list.\n",
      "\n",
      "For the SCAS list, we found that words such as \"meeting\", \"director\", \"service\", \"board\", \"comp\", \"phone\", \"committee\", \"sunday\" and \"contact\" were commonly found in emails with a response. Emails with words such as \"workgroup\", \"volunteer\", \"applications\", \"wednesday\", \"training\", \"apply\", and \"anyone\" were not likely to elicit a response. "
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def make_bow(mbox,num_words):\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    # define global bag of words\n",
      "    global_text = []\n",
      "\n",
      "    for msg in mbox:\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamations = count_exclamations(body)\n",
      "            # strip newlines, make lowercase\n",
      "            # strip punctuation, remove numbers on their own,\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            # split on spaces\n",
      "            split_body = body.split(' ')\n",
      "            global_text = global_text + split_body\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            continue\n",
      "\n",
      "    global_bow = collections.Counter(global_text)  # 14677 unique words\n",
      "    global_bow_top = global_bow.most_common(num_words)\n",
      "\n",
      "    global_bow_words = [x[0] for x in global_bow_top]\n",
      "\n",
      "    bow_list = []\n",
      "    for msg in mbox:\n",
      "        msg_word_dict = dict.fromkeys(global_bow_words,0)\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamaions = count_exclamations(body)\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            split_body = body.split(' ')\n",
      "            for word in split_body:\n",
      "                if word in msg_word_dict.keys():\n",
      "                    msg_word_dict[word] += 1\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            bow_list.append(msg_word_dict)  # should be all 0's if gets here\n",
      "            continue\n",
      "\n",
      "        bow_list.append(msg_word_dict)\n",
      "\n",
      "    bow_mat_list = []\n",
      "    for entry in bow_list:\n",
      "        bow_mat_list.append(entry.values())\n",
      "\n",
      "    # big matrix with feature list for each email\n",
      "    # rows = emails, columns = words\n",
      "    bow_mat = np.array(bow_mat_list)\n",
      "\n",
      "    # return matrix and list of words\n",
      "    return bow_mat, global_bow_words    \n",
      "\n",
      "\n",
      "def make_bow_given_dict(mbox,global_bow_words):\n",
      "\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    bow_list = []\n",
      "    for msg in mbox:\n",
      "        msg_word_dict = dict.fromkeys(global_bow_words,0)\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamaions = count_exclamations(body)\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            split_body = body.split(' ')\n",
      "            for word in split_body:\n",
      "                if word in msg_word_dict.keys():\n",
      "                    msg_word_dict[word] += 1\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            bow_list.append(msg_word_dict)  # should be all 0's if gets here\n",
      "            continue\n",
      "\n",
      "        bow_list.append(msg_word_dict)\n",
      "\n",
      "    bow_mat_list = []\n",
      "    for entry in bow_list:\n",
      "        bow_mat_list.append(entry.values())\n",
      "\n",
      "    # big matrix with feature list for each email\n",
      "    # rows = emails, columns = words\n",
      "    bow_mat = np.array(bow_mat_list)\n",
      "\n",
      "    # return matrix and list of words\n",
      "    return bow_mat\n",
      "\n",
      "def make_bow_given_dict_string(input_str,global_bow_words):\n",
      "\n",
      "    bow_list = []\n",
      "    msg_word_dict = dict.fromkeys(global_bow_words,0)\n",
      "\n",
      "    body = input_str\n",
      "    exclamaions = count_exclamations(body)\n",
      "    body = replace_newlines(body)\n",
      "    body = strip_punctuation(delete_nums(body)).lower()\n",
      "    split_body = body.split(' ')\n",
      "    for word in split_body:\n",
      "        if word in msg_word_dict.keys():\n",
      "            msg_word_dict[word] += 1\n",
      "\n",
      "    bow_list.append(msg_word_dict)\n",
      "\n",
      "    bow_mat_list = []\n",
      "    for entry in bow_list:\n",
      "        bow_mat_list.append(entry.values())\n",
      "\n",
      "    bow_mat = np.array(bow_mat_list)\n",
      "\n",
      "    return bow_mat"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def train_bow_classifier(mbox,num_words):\n",
      "\n",
      "    # split data\n",
      "    split_pct = 0.6  # test-train split percent\n",
      "    split = split_by_response(mbox)\n",
      "    num_pos = len(split[0])\n",
      "    num_neg = len(split[1])\n",
      "    pos_split_idx = int(num_pos * split_pct)\n",
      "    neg_split_idx = int(num_neg * split_pct)\n",
      "\n",
      "    # shuffle data - random.shuffle is in place\n",
      "    pos_data = split[0]\n",
      "    neg_data = split[1]\n",
      "    random.shuffle(pos_data)\n",
      "    random.shuffle(neg_data)\n",
      "\n",
      "    # split data\n",
      "    train_pos = pos_data[:pos_split_idx]\n",
      "    train_neg = neg_data[:neg_split_idx]\n",
      "    test_pos = pos_data[pos_split_idx:]\n",
      "    test_neg = neg_data[neg_split_idx:]\n",
      "    train_pos_labels = [1]*len(train_pos)\n",
      "    train_neg_labels = [0]*len(train_neg)\n",
      "    test_pos_labels = [1]*len(test_pos)\n",
      "    test_neg_labels = [0]*len(test_neg)\n",
      "\n",
      "    train_data = train_pos + train_neg\n",
      "    test_data = test_pos + test_neg\n",
      "    train_labels = train_pos_labels + train_neg_labels\n",
      "    test_labels = test_pos_labels + test_neg_labels\n",
      "\n",
      "    # get data matrices - TRAIN DATA\n",
      "    train_bow_mat, global_bow_words = make_bow(train_data, num_words)\n",
      "    train_times = find_time(train_data)\n",
      "    train_bool_responses = [1 if x > 0 else 0 for x in train_times]\n",
      "    train_times = [x if x > 0 else float('Inf') for x in train_times]\n",
      "\n",
      "    # get data matrices - TEST DATA\n",
      "    test_bow_mat = make_bow_given_dict(test_data, global_bow_words)\n",
      "    test_times = find_time(test_data)\n",
      "    test_bool_responses = [1 if x > 0 else 0 for x in test_times]\n",
      "    test_times = [x if x > 0 else float('Inf') for x in test_times]\n",
      "\n",
      "    # transform to np arrays\n",
      "    # train_bool_responses = np.array(train_bool_responses)\n",
      "    train_times = np.array(train_times)\n",
      "    # test_bool_responses = np.array(test_bool_responses)\n",
      "    test_times = np.array(test_times)\n",
      "    train_bool_responses = np.array(train_labels)\n",
      "    test_bool_responses = np.array(test_labels)\n",
      "\n",
      "    classifiers = []\n",
      "\n",
      "    # train, evaluate, and test random forest\n",
      "    rf_test_scores = []\n",
      "    k = np.arange(20)+1\n",
      "    parameters = {'n_estimators': k}\n",
      "    rf = RandomForestClassifier()\n",
      "    rf = sklearn.grid_search.GridSearchCV(rf, parameters, cv=10)\n",
      "    rf.fit(train_bow_mat,train_bool_responses)\n",
      "    rf_test_scores.append(rf.score(test_bow_mat,test_bool_responses))\n",
      "    print 'rf trained'\n",
      "\n",
      "    # train, evaluate, and test SVM\n",
      "    c = [1,10]\n",
      "    parameters = {'C': c}\n",
      "    svm_test_scores = []\n",
      "    svm = SVC(kernel='linear', probability=True)\n",
      "    svm = sklearn.grid_search.GridSearchCV(svm, parameters, cv=10)\n",
      "    svm.fit(train_bow_mat,train_bool_responses)\n",
      "    svm_test_scores.append(svm.score(test_bow_mat,test_bool_responses))\n",
      "    print 'svm trained'\n",
      "\n",
      "    # print test scores\n",
      "    print rf_test_scores\n",
      "    print svm_test_scores\n",
      "\n",
      "    classifiers.append(rf)\n",
      "    classifiers.append(svm)\n",
      "\n",
      "    return classifiers, global_bow_words\n",
      "\n",
      "\n",
      "def evaluate_bow_classifiers(classifiers,global_bow_words,input_message):\n",
      "    \n",
      "    # transform input message to matrix\n",
      "    input_bow_mat = make_bow_given_dict_string(input_message, global_bow_words)\n",
      "\n",
      "    for classifier in classifiers:\n",
      "        print classifier.predict(input_bow_mat)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 5
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "N-gram classifier"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "N-gram classifier. Trained a random forest and an SVM on these features in order to predict a binary response to any email. The method 'evaluate_classifiers' returns a label based on each of the two classifiers given the text of an email you would like to send to the desired list."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def make_ngram(mbox,num_words,n):\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    # define global ngram collection\n",
      "    global_ngrams = []\n",
      "\n",
      "    for msg in mbox:\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamations = count_exclamations(body)\n",
      "            # strip newlines, make lowercase\n",
      "            # strip punctuation, remove numbers on their own,\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            # split on spaces\n",
      "            split_body = body.split(' ')\n",
      "            n_grams = process_ngrams(split_body,n)\n",
      "            global_ngrams = global_ngrams + n_grams\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            continue\n",
      "\n",
      "    global_ngrams = collections.Counter(global_ngrams)\n",
      "    global_ngrams_top = global_ngrams.most_common(num_words)\n",
      "    global_ngrams_words = [x[0] for x in global_ngrams_top]\n",
      "\n",
      "    ngram_list = []\n",
      "    for msg in mbox:\n",
      "        msg_ngram_dict = dict.fromkeys(global_ngrams_words,0)\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamaions = count_exclamations(body)\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            split_body = body.split(' ')\n",
      "            n_grams = process_ngrams(split_body,n)\n",
      "\n",
      "            for ngram in n_grams:\n",
      "                if ngram in msg_ngram_dict.keys():\n",
      "                    msg_ngram_dict[ngram] += 1\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            ngram_list.append(msg_ngram_dict)  # should be all 0's if gets here\n",
      "            continue\n",
      "\n",
      "        ngram_list.append(msg_ngram_dict)\n",
      "\n",
      "    ngram_mat_list = []\n",
      "    for entry in ngram_list:\n",
      "        ngram_mat_list.append(entry.values())\n",
      "\n",
      "    # big matrix with feature list for each email\n",
      "    # rows = emails, columns = words\n",
      "    ngram_mat = np.array(ngram_mat_list)\n",
      "\n",
      "    # return matrix and list of words\n",
      "    return ngram_mat, global_ngrams_words   \n",
      "\n",
      "\n",
      "def make_ngram_given_dict(mbox,global_ngrams_words,n):\n",
      "\n",
      "    if type(mbox) is str:\n",
      "        mbox = mailbox.mbox(mbox)\n",
      "\n",
      "    ngram_list = []\n",
      "    for msg in mbox:\n",
      "        msg_ngram_dict = dict.fromkeys(global_ngrams_words,0)\n",
      "        try:\n",
      "            body = get_body(msg)\n",
      "            exclamaions = count_exclamations(body)\n",
      "            body = replace_newlines(body)\n",
      "            body = strip_punctuation(delete_nums(body)).lower()\n",
      "            split_body = body.split(' ')\n",
      "            ngrams = process_ngrams(split_body,n)\n",
      "            for ngram in ngrams:\n",
      "                if ngram in msg_ngram_dict.keys():\n",
      "                    msg_ngram_dict[ngram] += 1\n",
      "        except:\n",
      "            logging.error('Could not get body of email' + str(msg['Subject']))\n",
      "            ngram_list.append(msg_ngram_dict)  # should be all 0's if gets here\n",
      "            continue\n",
      "\n",
      "        ngram_list.append(msg_ngram_dict)\n",
      "\n",
      "    ngram_mat_list = []\n",
      "    for entry in ngram_list:\n",
      "        ngram_mat_list.append(entry.values())\n",
      "\n",
      "    # big matrix with feature list for each email\n",
      "    # rows = emails, columns = words\n",
      "    ngram_mat = np.array(ngram_mat_list)\n",
      "\n",
      "    # return matrix\n",
      "    return ngram_mat\n",
      "\n",
      "def make_ngram_given_dict_string(input_str,global_ngrams_words,n):\n",
      "\n",
      "    ngram_list = []\n",
      "    msg_ngram_dict = dict.fromkeys(global_ngrams_words,0)\n",
      "\n",
      "    body = input_str\n",
      "    exclamaions = count_exclamations(body)\n",
      "    body = replace_newlines(body)\n",
      "    body = strip_punctuation(delete_nums(body)).lower()\n",
      "    split_body = body.split(' ')\n",
      "    ngrams = process_ngrams(split_body,n)\n",
      "    for ngram in ngrams:\n",
      "        if ngram in msg_ngram_dict.keys():\n",
      "            msg_ngram_dict[ngram] += 1\n",
      "\n",
      "    ngram_list.append(msg_ngram_dict)\n",
      "\n",
      "    ngram_mat_list = []\n",
      "    for entry in ngram_list:\n",
      "        ngram_mat_list.append(entry.values())\n",
      "\n",
      "    ngram_mat = np.array(ngram_mat_list)\n",
      "\n",
      "    return ngram_mat "
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "def train_ngram_classifier(mbox,num_words,n):\n",
      "\n",
      "    # split data\n",
      "    split_pct = 0.6  # test-train split percent\n",
      "    split = split_by_response(mbox)\n",
      "    num_pos = len(split[0])\n",
      "    num_neg = len(split[1])\n",
      "    pos_split_idx = int(num_pos * split_pct)\n",
      "    neg_split_idx = int(num_neg * split_pct)\n",
      "\n",
      "    # shuffle data - random.shuffle is in place\n",
      "    pos_data = split[0]\n",
      "    neg_data = split[1]\n",
      "    random.shuffle(pos_data)\n",
      "    random.shuffle(neg_data)\n",
      "\n",
      "    # split data\n",
      "    train_pos = pos_data[:pos_split_idx]\n",
      "    train_neg = neg_data[:neg_split_idx]\n",
      "    test_pos = pos_data[pos_split_idx:]\n",
      "    test_neg = neg_data[neg_split_idx:]\n",
      "    train_pos_labels = [1]*len(train_pos)\n",
      "    train_neg_labels = [0]*len(train_neg)\n",
      "    test_pos_labels = [1]*len(test_pos)\n",
      "    test_neg_labels = [0]*len(test_neg)\n",
      "    \n",
      "    train_data = train_pos + train_neg\n",
      "    test_data = test_pos + test_neg\n",
      "    train_labels = train_pos_labels + train_neg_labels\n",
      "    test_labels = test_pos_labels + test_neg_labels\n",
      "\n",
      "    # get data matrices - TRAIN DATA\n",
      "    train_ngram_mat, global_ngrams_words = make_ngram(train_data, num_words, n)\n",
      "    train_times = find_time(train_data)\n",
      "    train_bool_responses = [1 if x > 0 else 0 for x in train_times]\n",
      "    train_times = [x if x > 0 else float('Inf') for x in train_times]\n",
      "\n",
      "    # get data matrices - TEST DATA\n",
      "    test_ngram_mat = make_ngram_given_dict(test_data, global_ngrams_words, n)\n",
      "    test_times = find_time(test_data)\n",
      "    test_bool_responses = [1 if x > 0 else 0 for x in test_times]\n",
      "    test_times = [x if x > 0 else float('Inf') for x in test_times]\n",
      "\n",
      "    # transform to np arrays\n",
      "    # train_bool_responses = np.array(train_bool_responses)\n",
      "    train_times = np.array(train_times)\n",
      "    # test_bool_responses = np.array(test_bool_responses)\n",
      "    test_times = np.array(test_times)\n",
      "\n",
      "    train_bool_responses = np.array(train_labels)\n",
      "    train_weights = train_bool_responses + 1  # should be 2's and 1's\n",
      "    test_bool_responses = np.array(test_labels)\n",
      "\n",
      "    classifiers = []\n",
      "\n",
      "    # train, evaluate, and test random forest\n",
      "    rf_test_scores = []\n",
      "    k = np.arange(20)+1\n",
      "    parameters = {'n_estimators': k}\n",
      "    rf = RandomForestClassifier()\n",
      "    rf = sklearn.grid_search.GridSearchCV(rf, parameters, cv=10)\n",
      "    rf.fit(train_ngram_mat,train_bool_responses)#,sample_weight=train_weights)\n",
      "    rf_test_scores.append(rf.score(test_ngram_mat,test_bool_responses))\n",
      "    print 'rf trained'\n",
      "\n",
      "    # train, evaluate, and test SVM\n",
      "    c = [1,10] #,100,1000]\n",
      "    parameters = {'C': c}\n",
      "    svm_test_scores = []\n",
      "    svm = SVC(kernel='linear', probability=True, class_weight='auto')\n",
      "    svm = sklearn.grid_search.GridSearchCV(svm, parameters, cv=10)\n",
      "    svm.fit(train_ngram_mat,train_bool_responses)\n",
      "    svm_test_scores.append(svm.score(test_ngram_mat,test_bool_responses))\n",
      "    print 'svm trained'\n",
      "\n",
      "    # print test scores\n",
      "    print rf_test_scores\n",
      "    print svm_test_scores\n",
      "\n",
      "    classifiers.append(rf)\n",
      "    classifiers.append(svm)\n",
      "\n",
      "    return classifiers, global_ngrams_words\n",
      "\n",
      "\n",
      "def evaluate_ngram_classifiers(classifiers,global_ngrams_words,input_message,n):\n",
      "    \n",
      "    input_bow_mat = make_ngram_given_dict_string(input_message, global_ngrams_words, n)\n",
      "\n",
      "    for classifier in classifiers:\n",
      "        print classifier.predict(input_bow_mat)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 7
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Predict your own email!"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "**Please don't run this in the ipython notebook; it will crash the notebook because the data file is too big. However, you can run it in your terminal (in Python) and it should predict your email. Also, this full functionality will be available on our website so you can test it there! Thanks!** \n",
      "\n",
      "The mbox file is located at 'http://208.43.122.13/scas-list.mbox'. Please download that and input the path to the file in the code provided. Then you can run the entire thing in the Python console. Sorry for the convoluted steps, but mbox does not play well with Python."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "n = 2  # 2-grams for performance\n",
      "num_bow_words = 300  # number of top words to keep in dictionary\n",
      "num_ngram_words = 300  # number of top 2-grams to keep in dictionary\n",
      "\n",
      "mbox = mailbox.mbox('PATH/TO/FILE/HERE')  # read above description!\n",
      "\n",
      "input_message = \"Enter your own email here. Welcome back, everyone, and please respond to this email! Just reply all and you'll be set. Don't forget. Thanks!\"\n",
      "\n",
      "# Get classifiers\n",
      "bow_classifiers, global_bow_words = train_bow_classifier(mbox,num_bow_words)\n",
      "ngram_classifiers, global_ngram_words = train_ngram_classifier(mbox,num_ngram_words,n)\n",
      "\n",
      "# Predict the response to your email\n",
      "evaluate_bow_classifiers(bow_classifiers,global_bow_words,input_message)\n",
      "evaluate_ngram_classifiers(ngram_classifiers,global_ngram_words,input_message,n)\n"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Final Analysis"
     ]
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "What We Learned"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Quite generally, we learned that Harvard students have steadily decreased in the number of emails sent since 2009. In addition, the most emails are sent in April, on Sundays, and in the late afternoon, peaking around 4 pm. In addition, roughly the top 10% of popular subject-lines account for 50% of the emails sent over the Harvard mailing lists\n",
      "\n",
      "Through our mailbox to graph method, we are able to turn the emails into a graph based on responses, which was able to accurately determine the highest status member of a club, i.e. President. Beyond that, it became harder to differentiate between say, the Vice President and the Treasurer, which makes sense, as their involvement would be about the same. In comparison, the President of any club is usually much more involved than any other member and receives much more email activity. \n",
      "\n",
      "In addition, we were able to fairly accurately predict whether or not an email sent over the SCAS list would receive a response. Because the typical email sent at Harvard is generally similar in style across similar lists, we can generalize this classifier to clubs similar in type to SCAS. However, it is hard to predict exactly due to multiple factors we cannot capture, such as the particular character of a person outside of their email or the workload or activity load of the students in a club at a particular time. But, for our data set, the bag of words model worked better than the n-gram model, which is usually not the case. We believe the reason for this is that the n-grams were too sparse to be indicative. With bag of words, the random forest method worked the best, with a test accuracy of 82%. "
     ]
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "How We Answered Our Questions"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Through the methods outlined in our data section, we were able to analyze the emails to visualize general trends. \n",
      "\n",
      "For the graph, we conducted some research on the best way to calculate weights, and found that $\\frac{1}{t^2}$ was the best method. We then created the graphs and checked if the largest nodes corresponded to the Presidents of the clubs we looked at. We found perfect matches and thus concluded that forming a network from an email list provided a way to inspect the hierarchy of a club or organization. \n",
      "\n",
      "For the predictor, we tried both bag of words, and n-grams, then for each we tried a random forest and SVM classifer. We used the SCAS list, one with high activity, and split the data set 60-40 for training and testing. We then calculated the accuracy of each method with each classifier, and found that the bag of words model with the random forest classifier had the highest accuracy"
     ]
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "How We Justify Our Answers"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We justify our answers based on the accuracy of our tests, i.e. whether our networks were able to present the hierarchy of a club accurately and the accuracy of our classifier on our testing set. In addition, intuition plays a role - do the email trends we found make sense, considering the life of an average student, and does it make sense that there are aspects of email communication we cannot capture, thus making it difficult for our classifier to be perfectly accurate. \n",
      "\n",
      "In conclusion, we believe we have found interesting information on the emails sent over Harvard lists, as well as the interactions among members of a list and a relevant model to help students construct their emails in order to best enable responses. "
     ]
    }
   ],
   "metadata": {}
  }
 ]
}