Read_Interactions.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ICP-CDN Interaction Preprocessing\n",
    "\n",
    "**Input:** Raw data table (hidden for future research purposes)\n",
    "\n",
    "**Output:** Interactions matrix of dimension: num_users x num_items\n",
    "* Go through raw data, keep track of which ICPs have used which CDNs\n",
    "* Nonzero entries represent a recorded interaction\n",
    "* Maintain *ICP-userindex* and *CDN-itemindex* mappings to remember which rows/columns correspond to which ICPs/CDNs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy.sparse as sp\n",
    "import csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Read in ICP-CDN interactions (icpdomaincdntotalamountofmedia)\n",
    "interactions_filepath = 'icpdomaincdntotalamountofmedia/all_interactions.txt'\n",
    "interactions_header = ['industry','icp','domain','cdn',\n",
    "                                      'totalsize','cdnsize','nocdnsize',\n",
    "                                      'ts']\n",
    "interactions_dtypes = {\n",
    "    'industry': str,\n",
    "    'icp': str,\n",
    "    'domain': str,\n",
    "    'cdn': str,\n",
    "    'totalsize': np.float64,\n",
    "    'cdnsize': np.float64,\n",
    "    'nocdnsize': np.float64,\n",
    "    'ts': str\n",
    "}\n",
    "interactions_df = pd.read_csv(interactions_filepath, \n",
    "                              sep=',', header=None, \n",
    "                              names=interactions_header,\n",
    "                              dtype=interactions_dtypes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>industry</th>\n",
       "      <th>icp</th>\n",
       "      <th>domain</th>\n",
       "      <th>cdn</th>\n",
       "      <th>totalsize</th>\n",
       "      <th>cdnsize</th>\n",
       "      <th>nocdnsize</th>\n",
       "      <th>ts</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0100</td>\n",
       "      <td>365jia.cn</td>\n",
       "      <td>365jia.cn</td>\n",
       "      <td>NaN</td>\n",
       "      <td>3.525360e+09</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.525360e+09</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0100</td>\n",
       "      <td>365jilin.com</td>\n",
       "      <td>365jilin.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>3.595730e+05</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.595730e+05</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0100</td>\n",
       "      <td>wfcmw.cn</td>\n",
       "      <td>wfcmw.cn</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.362390e+09</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.362390e+09</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.100ec.cn</td>\n",
       "      <td>www.100ec.cn</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2.940954e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.940954e+08</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.163.com</td>\n",
       "      <td>www.163.com</td>\n",
       "      <td>002</td>\n",
       "      <td>8.273679e+08</td>\n",
       "      <td>8.273679e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.163.com</td>\n",
       "      <td>www.163.com</td>\n",
       "      <td>004</td>\n",
       "      <td>3.701383e+08</td>\n",
       "      <td>3.701383e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.21cbh.com</td>\n",
       "      <td>www.21cbh.com</td>\n",
       "      <td>002</td>\n",
       "      <td>2.036111e+10</td>\n",
       "      <td>2.036111e+10</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.21cn.com</td>\n",
       "      <td>www.21cn.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.151693e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.151693e+08</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.21cn.com</td>\n",
       "      <td>www.21cn.com</td>\n",
       "      <td>003</td>\n",
       "      <td>8.983205e+09</td>\n",
       "      <td>8.983205e+09</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.2500sz.com</td>\n",
       "      <td>www.2500sz.com</td>\n",
       "      <td>001</td>\n",
       "      <td>1.621454e+10</td>\n",
       "      <td>1.621454e+10</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.265g.com</td>\n",
       "      <td>www.265g.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2.695444e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.695444e+08</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.315che.com</td>\n",
       "      <td>www.315che.com</td>\n",
       "      <td>002</td>\n",
       "      <td>3.201521e+09</td>\n",
       "      <td>3.201521e+09</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.364000.com</td>\n",
       "      <td>www.364000.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2.688717e+08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.688717e+08</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.3dmgame.com</td>\n",
       "      <td>bbs.3dmgame.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.172880e+05</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.172880e+05</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>0100</td>\n",
       "      <td>www.3dmgame.com</td>\n",
       "      <td>img01.3dmgame.com</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.386610e+07</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.386610e+07</td>\n",
       "      <td>201501</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   industry              icp             domain  cdn     totalsize  \\\n",
       "0      0100        365jia.cn          365jia.cn  NaN  3.525360e+09   \n",
       "1      0100     365jilin.com       365jilin.com  NaN  3.595730e+05   \n",
       "2      0100         wfcmw.cn           wfcmw.cn  NaN  1.362390e+09   \n",
       "3      0100     www.100ec.cn       www.100ec.cn  NaN  2.940954e+08   \n",
       "4      0100      www.163.com        www.163.com  002  8.273679e+08   \n",
       "5      0100      www.163.com        www.163.com  004  3.701383e+08   \n",
       "6      0100    www.21cbh.com      www.21cbh.com  002  2.036111e+10   \n",
       "7      0100     www.21cn.com       www.21cn.com  NaN  1.151693e+08   \n",
       "8      0100     www.21cn.com       www.21cn.com  003  8.983205e+09   \n",
       "9      0100   www.2500sz.com     www.2500sz.com  001  1.621454e+10   \n",
       "10     0100     www.265g.com       www.265g.com  NaN  2.695444e+08   \n",
       "11     0100   www.315che.com     www.315che.com  002  3.201521e+09   \n",
       "12     0100   www.364000.com     www.364000.com  NaN  2.688717e+08   \n",
       "13     0100  www.3dmgame.com    bbs.3dmgame.com  NaN  1.172880e+05   \n",
       "14     0100  www.3dmgame.com  img01.3dmgame.com  NaN  1.386610e+07   \n",
       "\n",
       "         cdnsize     nocdnsize      ts  \n",
       "0   0.000000e+00  3.525360e+09  201501  \n",
       "1   0.000000e+00  3.595730e+05  201501  \n",
       "2   0.000000e+00  1.362390e+09  201501  \n",
       "3   0.000000e+00  2.940954e+08  201501  \n",
       "4   8.273679e+08  0.000000e+00  201501  \n",
       "5   3.701383e+08  0.000000e+00  201501  \n",
       "6   2.036111e+10  0.000000e+00  201501  \n",
       "7   0.000000e+00  1.151693e+08  201501  \n",
       "8   8.983205e+09  0.000000e+00  201501  \n",
       "9   1.621454e+10  0.000000e+00  201501  \n",
       "10  0.000000e+00  2.695444e+08  201501  \n",
       "11  3.201521e+09  0.000000e+00  201501  \n",
       "12  0.000000e+00  2.688717e+08  201501  \n",
       "13  0.000000e+00  1.172880e+05  201501  \n",
       "14  0.000000e+00  1.386610e+07  201501  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "interactions_df.head(15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([False], dtype=bool)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "interactions_df.duplicated().unique()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Looks good, no duplicate data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# # Read in ICP Statistics\n",
    "# icp_statistics_filepath = 'icpstatistic/all_icp_statistics.txt'\n",
    "# icp_statistics_header = ['icp', 'textnum', 'imagenum', 'videonum', 'unknownnum',\\\n",
    "#                         'textbytes','imagebytes','unknownbytes','createtime', 'ts']\n",
    "# icp_statistics_dtypes = {\n",
    "#     'icp': str,\n",
    "#     'textnum': np.int64,\n",
    "#     'imagenum': np.int64,\n",
    "#     'videonum': np.int64,\n",
    "#     'unknownnum': np.int64,\n",
    "#     'textbytes': np.float64,\n",
    "#     'imagebytes': np.float64,\n",
    "#     'unknownbytes': np.float64,\n",
    "#     'createtime': str,\n",
    "#     'ts': str\n",
    "# }\n",
    "# icp_statistics_df = pd.read_csv(icp_statistics_filepath,\n",
    "#                                sep=',', header=None,\n",
    "#                                names=icp_statistics_header,\n",
    "#                                dtype=icp_statistics_dtypes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# icp_statistics_df.head()\n",
    "# # Column numbers don't match! Look into this later"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num users (ICPs): 1753 | Num items (CDNs): 39\n",
      "['365jia.cn' '365jilin.com' 'wfcmw.cn' ..., 'www.metroer.com'\n",
      " 'www.xy280.com' 'www.yccar.com']\n",
      "['002' '004' '003' '001' '203' '013' '202' '005' '204' '201' '009' '007'\n",
      " '006' '008' '101' '104' '111' '102' '105' '014' '103' '106' '113' '110'\n",
      " '121' '116' '109' '125' '114' '108' '124' '107' '011' '119' '112' '117'\n",
      " '115' '122' '123']\n"
     ]
    }
   ],
   "source": [
    "# Create lists of unique users (ICPs), items (CDNs)\n",
    "icp_list = interactions_df.icp.unique()\n",
    "cdn_list = interactions_df.cdn.unique()[1:] # remove NaN\n",
    "n_users = icp_list.shape[0]\n",
    "n_items = cdn_list.shape[0]\n",
    "\n",
    "print \"Num users (ICPs): \" + str(n_users) + \" | Num items (CDNs): \" + str(n_items)\n",
    "print icp_list\n",
    "print cdn_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['365jia.cn', '365jilin.com', 'wfcmw.cn', '1000eb.com', 'duoshuo.com', 'gai001.com', 'nut.com.cn', 'panasonic.cn', 'segmentfault.com', 't.cn', 'w.cn', 'wallstreetcn.com', '5173tuan.com', 'tuanweihui.com', 'new.sdo.com', 'lengxiaohua.cn', 'kdweibo.com', 'youyuan.com', '5lux.com', 'bab.la', 'shuaijiao.com', 'tsinghua.edu.cn']\n"
     ]
    }
   ],
   "source": [
    "# Check for duplicates with or without \"www.__url__\"\n",
    "no_www_list = [url for url in icp_list if \"www.\" not in url]\n",
    "print no_www_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n",
      "(array([], dtype=int64),)\n"
     ]
    }
   ],
   "source": [
    "for element in no_www_list:\n",
    "    with_www = \"www.\" + element\n",
    "    print np.where(icp_list == with_www)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Confirmed: no \"www\" duplicates - yay"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Create interactions matrix + user-row, item-col mappings\n",
    "interactions = np.zeros((n_users, n_items))\n",
    "\n",
    "# user-row\n",
    "uidx_to_icp = {} # User index to ICP\n",
    "icp_to_uidx = {} # ICP to user index\n",
    "for (index, icp) in enumerate(icp_list):\n",
    "    uidx_to_icp[index] = icp\n",
    "    icp_to_uidx[icp] = index\n",
    "    \n",
    "# item-col\n",
    "iidx_to_cdn = {} # Item index to CDN\n",
    "cdn_to_iidx = {} # CDN to item index\n",
    "for (index, cdn) in enumerate(cdn_list):\n",
    "    iidx_to_cdn[index] = cdn\n",
    "    cdn_to_iidx[cdn] = index\n",
    "\n",
    "# Read in entries\n",
    "for entry in interactions_df.itertuples():\n",
    "    icp = entry[2] # 2nd element\n",
    "    cdn = entry[4] # 4th element\n",
    "    if cdn > 0: # Remove NaN\n",
    "        row = icp_to_uidx[icp]\n",
    "        col = cdn_to_iidx[cdn]\n",
    "        interactions[row, col] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Matrix shape: (1753, 39)\n",
      "Interaction pairs: 2242\n"
     ]
    }
   ],
   "source": [
    "# Convert to compressed sparse-row matrix, check shape, num. interactions\n",
    "interactions = sp.csr_matrix(interactions)\n",
    "print \"Matrix shape: \" + str(interactions.shape)\n",
    "print \"Interaction pairs: \" + str(interactions.getnnz())\n",
    "# 2242 interaction pairs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1151, 39)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check number of ICPs that have used CDNs (non-empty rows)\n",
    "interactions[interactions.getnnz(1) > 0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.950054579073\n"
     ]
    }
   ],
   "source": [
    "# Check sparsity of remaining ICP-CDN matrix (if we remove empty rows)\n",
    "sparsity = (1151*39 - 2242.)/(1151*39.)\n",
    "print sparsity"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Remove ICPs that haven't used CDNs from interactions matrix\n",
    "rows_interacted = interactions.getnnz(1).astype(bool) # Boolean list of if row (ICP) has used a CDN\n",
    "icp_list_remaining = icp_list[rows_interacted] # Remaining list of ICPs that have used CDNs\n",
    "interactions_remaining = interactions[rows_interacted] # Remaining array, only includes ICPs that have used CDNs\n",
    "num_users_remaining = interactions_remaining.shape[0]\n",
    "\n",
    "assert (False not in interactions_remaining.getnnz(1)) # No empty rows remaining"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Redo user-row mappings using remaining ICP list\n",
    "uidx_to_icp_remaining = {}\n",
    "icp_to_uidx_remaining = {}\n",
    "for (index, icp) in enumerate(icp_list_remaining):\n",
    "    uidx_to_icp_remaining[index] = icp\n",
    "    icp_to_uidx_remaining[icp] = index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# True if ICP has used a CDN (included in final matrix)\n",
    "# False if ICP has NOT used a CDN (excluded from final matrix)\n",
    "def has_used_cdn(icp_name):\n",
    "    return icp_name in icp_to_uidx_remaining"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Use for testing: sort .csv by ICP name, plug in some names, see if result is correct\n",
    "has_used_cdn('www.nipic.com')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of CDNs: 0 | Count: 0\n",
      "Number of CDNs: 1 | Count: 567\n",
      "Number of CDNs: 2 | Count: 314\n",
      "Number of CDNs: 3 | Count: 163\n",
      "Number of CDNs: 4 | Count: 58\n",
      "Number of CDNs: 5 | Count: 26\n",
      "Number of CDNs: 6 | Count: 9\n",
      "Number of CDNs: 7 | Count: 6\n",
      "Number of CDNs: 8 | Count: 1\n",
      "Number of CDNs: 9 | Count: 0\n",
      "Number of CDNs: 10 | Count: 1\n",
      "Number of CDNs: 11 | Count: 1\n",
      "Number of CDNs: 12 | Count: 2\n",
      "Number of CDNs: 13 | Count: 2\n",
      "Number of CDNs: 14 | Count: 0\n",
      "Number of CDNs: 15 | Count: 0\n",
      "Number of CDNs: 16 | Count: 0\n",
      "Number of CDNs: 17 | Count: 0\n",
      "Number of CDNs: 18 | Count: 0\n",
      "Number of CDNs: 19 | Count: 0\n",
      "Number of CDNs: 20 | Count: 0\n",
      "Number of CDNs: 21 | Count: 1\n"
     ]
    }
   ],
   "source": [
    "# Just out of curiosity: See how many different CDNs each ICP has used (num. different interactions)\n",
    "max_num_cdns = max(interactions_remaining.getnnz(1))\n",
    "CDN_counts = np.zeros(max_num_cdns+1).astype(np.int32)\n",
    "\n",
    "for num in range(0, max_num_cdns+1):\n",
    "    CDN_counts[num] = interactions_remaining[interactions_remaining.getnnz(1) == num].shape[0]\n",
    "\n",
    "for num_cdns, count in enumerate(CDN_counts):\n",
    "    print \"Number of CDNs: \" + str(num_cdns) + \" | \" + \"Count: \" + str(count)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'www.gov.cn'"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# How did 1 ICP use 21 different CDN providers?? Let's find out who it was\n",
    "row = np.argmax(interactions_remaining.getnnz(1))\n",
    "uidx_to_icp_remaining[row]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Of course it was the government"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/lucashu/anaconda/envs/recsys/lib/python2.7/site-packages/lightfm/_lightfm_fast.py:9: UserWarning: LightFM was compiled without OpenMP support. Only a single thread will be used.\n",
      "  warnings.warn('LightFM was compiled without OpenMP support. '\n"
     ]
    }
   ],
   "source": [
    "# Let's try training a simple implicit feedback CF model on all our interaction data\n",
    "from lightfm import LightFM\n",
    "import lightfm.evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Code from Ethan Rosenthal's LightFM for Sketchfab tutorial\n",
    "def print_log(row, header=False, spacing=12):\n",
    "    top = ''\n",
    "    middle = ''\n",
    "    bottom = ''\n",
    "    for r in row:\n",
    "        top += '+{}'.format('-'*spacing)\n",
    "        if isinstance(r, str):\n",
    "            middle += '| {0:^{1}} '.format(r, spacing-2)\n",
    "        elif isinstance(r, int):\n",
    "            middle += '| {0:^{1}} '.format(r, spacing-2)\n",
    "        elif (isinstance(r, float)\n",
    "              or isinstance(r, np.float32)\n",
    "              or isinstance(r, np.float64)):\n",
    "            middle += '| {0:^{1}.5f} '.format(r, spacing-2)\n",
    "        bottom += '+{}'.format('='*spacing)\n",
    "    top += '+'\n",
    "    middle += '|'\n",
    "    bottom += '+'\n",
    "    if header:\n",
    "        print(top)\n",
    "        print(middle)\n",
    "        print(bottom)\n",
    "    else:\n",
    "        print(middle)\n",
    "        print(top)\n",
    "        \n",
    "def patk_learning_curve(model, train,\n",
    "                       iterarray, user_features=None,\n",
    "                       item_features=None, k=5,\n",
    "                       **fit_params):\n",
    "    old_epoch = 0\n",
    "    train_patk = []\n",
    "    headers = ['Epoch', 'train p@5']\n",
    "    print_log(headers, header=True)\n",
    "    for epoch in iterarray:\n",
    "        more = epoch - old_epoch\n",
    "        model.fit_partial(train,\n",
    "                          user_features=user_features,\n",
    "                          item_features=item_features,\n",
    "                         epochs=more, **fit_params)\n",
    "        this_train = lightfm.evaluation.precision_at_k(model, train, train_interactions=None, k=k)\n",
    "        \n",
    "        train_patk.append(np.mean(this_train))\n",
    "        row = [epoch, train_patk[-1]]\n",
    "        print_log(row)\n",
    "    return model, train_patk"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+------------+------------+\n",
      "|   Epoch    | train p@5  |\n",
      "+============+============+\n",
      "|     10     |  0.34926   |\n",
      "+------------+------------+\n",
      "|     20     |  0.36507   |\n",
      "+------------+------------+\n",
      "|     30     |  0.37150   |\n",
      "+------------+------------+\n",
      "|     40     |  0.37480   |\n",
      "+------------+------------+\n",
      "|     50     |  0.37533   |\n",
      "+------------+------------+\n",
      "|     60     |  0.37550   |\n",
      "+------------+------------+\n",
      "|     70     |  0.37550   |\n",
      "+------------+------------+\n",
      "|     80     |  0.37550   |\n",
      "+------------+------------+\n",
      "|     90     |  0.37550   |\n",
      "+------------+------------+\n",
      "|    100     |  0.37550   |\n",
      "+------------+------------+\n"
     ]
    }
   ],
   "source": [
    "model = LightFM(loss='warp', random_state=2016)\n",
    "model.fit(interactions_remaining, epochs=0) # just initialize\n",
    "\n",
    "iterarray = range(10, 110, 10)\n",
    "model, train_patk = patk_learning_curve(\n",
    "    model, interactions_remaining, iterarray, k=5,\n",
    "    **{'num_threads':1}\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhEAAAF+CAYAAAA81EgAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nd4oAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11235a950>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Generate plot\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "sns.set_style('white')\n",
    "\n",
    "def plot_patk(iterarray, patk,\n",
    "              title, k=5):\n",
    "    plt.plot(iterarray, patk);\n",
    "    plt.title(title, fontsize=20);\n",
    "    plt.xlabel('Epochs', fontsize=24);\n",
    "    plt.ylabel('p@{}'.format(k), fontsize=24);\n",
    "    plt.xticks(fontsize=14);\n",
    "    plt.yticks(fontsize=14);\n",
    "    \n",
    "# Plot train on left\n",
    "plot_patk(iterarray, train_patk,\n",
    "         'Train', k=5)\n",
    "\n",
    "plt.show();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Not bad! Still need to try with separate test data though.\n",
    "\n",
    "TODO 6/28/17:\n",
    "Create train-test splits\n",
    "Implement user, item features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Initial save: interactions matrix, index mappings\n",
    "import pickle\n",
    "with open('20170629-interactions-mappings.pkl', 'wb') as output:\n",
    "    pickle.dump((interactions_remaining, \\\n",
    "                 iidx_to_cdn, cdn_to_iidx, \\\n",
    "                 uidx_to_icp_remaining, icp_to_uidx_remaining), \\\n",
    "                 output, -1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}