gapbs_speedup.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "import common\n",
    "import importlib\n",
    "from matplotlib.lines import Line2D\n",
    "from matplotlib.patches import Patch\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import re\n",
    "\n",
    "pd.options.display.max_columns = None # always display all columns in a Dataframe\n",
    "pd.options.display.max_rows = None # same for the rows"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "OKY, the idea here is to compare the execution of the graph algorithms between those provided by the GAP.BS and those provided by the library, where available. Stinger comes with its own implementation of BFS, SSSP, WCC and PR; LLAMA of PR and BFS, same for GraphOne but the PR is so different that it makes no sense to compare it.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Data to load:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = common.import_gfe(\"view_graphalytics_inserts\").query(\"cluster == 'stones2'\")\n",
    "data[\"build_frequency_secs\"] = data[\"build_frequency_secs\"].fillna(0)\n",
    "# data[:5] # just to see what's there"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compute the summaries for each graph and algorithm:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "summaries = data.groupby([\"library\", \"compiler_family\", \"graph\", \"build_frequency_secs\", \"num_threads_read\", \"num_threads_write\", \"algorithm\"]).agg(completion_time=(\"median_secs\", \"median\"), count=(\"median_secs\", \"count\"))\n",
    "summaries = summaries.unstack(\"algorithm\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get the list of available executions:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>num_threads</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>library</th>\n",
       "      <th>compiler_family</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>g1_v4-bw-sp-ignore-build</th>\n",
       "      <th>gcc</th>\n",
       "      <td>3, 6, 12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>g1_v4-ref-ignore-build</th>\n",
       "      <th>gcc</th>\n",
       "      <td>3, 6, 12, 20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>livegraph_ro</th>\n",
       "      <th>gcc</th>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>llama6</th>\n",
       "      <th>gcc</th>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>llama6-ref</th>\n",
       "      <th>gcc</th>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>stinger5</th>\n",
       "      <th>gcc</th>\n",
       "      <td>20, 40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>stinger5-ref</th>\n",
       "      <th>gcc</th>\n",
       "      <td>20, 40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-dv.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-lcc-dv.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-lcc.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-ta-dv.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-ta-lcc-dv.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-ta-lcc.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo-ta.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teseo.9</th>\n",
       "      <th>gcc</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                           num_threads\n",
       "library                  compiler_family              \n",
       "g1_v4-bw-sp-ignore-build gcc                  3, 6, 12\n",
       "g1_v4-ref-ignore-build   gcc              3, 6, 12, 20\n",
       "livegraph_ro             gcc                        20\n",
       "llama6                   gcc                        16\n",
       "llama6-ref               gcc                        16\n",
       "stinger5                 gcc                    20, 40\n",
       "stinger5-ref             gcc                    20, 40\n",
       "teseo-dv.9               gcc                        40\n",
       "teseo-lcc-dv.9           gcc                        40\n",
       "teseo-lcc.9              gcc                        40\n",
       "teseo-ta-dv.9            gcc                        40\n",
       "teseo-ta-lcc-dv.9        gcc                        40\n",
       "teseo-ta-lcc.9           gcc                        40\n",
       "teseo-ta.9               gcc                        40\n",
       "teseo.9                  gcc                        40"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fn_join_threads = lambda grp: \", \".join(map(str, grp.unique()))\n",
    "summaries.reset_index().groupby([\"library\", \"compiler_family\"]).agg(num_threads=(\"num_threads_write\", fn_join_threads))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Prepare the data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>algorithm</th>\n",
       "      <th>bfs</th>\n",
       "      <th>cdlp</th>\n",
       "      <th>lcc</th>\n",
       "      <th>pagerank</th>\n",
       "      <th>sssp</th>\n",
       "      <th>wcc</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>library</th>\n",
       "      <th>graph</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"7\" valign=\"top\">stinger</th>\n",
       "      <th>dota-league</th>\n",
       "      <td>2.021748</td>\n",
       "      <td>0.992872</td>\n",
       "      <td>1.000247</td>\n",
       "      <td>1.160999</td>\n",
       "      <td>4.824633</td>\n",
       "      <td>0.324122</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-22</th>\n",
       "      <td>3.452096</td>\n",
       "      <td>1.005264</td>\n",
       "      <td>1.001150</td>\n",
       "      <td>9.298210</td>\n",
       "      <td>8.026528</td>\n",
       "      <td>0.446463</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-24</th>\n",
       "      <td>8.824797</td>\n",
       "      <td>1.001041</td>\n",
       "      <td>NaN</td>\n",
       "      <td>15.082731</td>\n",
       "      <td>9.983933</td>\n",
       "      <td>0.612912</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-26</th>\n",
       "      <td>20.197229</td>\n",
       "      <td>0.978307</td>\n",
       "      <td>NaN</td>\n",
       "      <td>18.232644</td>\n",
       "      <td>9.997364</td>\n",
       "      <td>0.841455</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-22</th>\n",
       "      <td>3.853086</td>\n",
       "      <td>0.986295</td>\n",
       "      <td>1.010869</td>\n",
       "      <td>1.202704</td>\n",
       "      <td>5.610110</td>\n",
       "      <td>0.500772</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-24</th>\n",
       "      <td>11.512104</td>\n",
       "      <td>0.990429</td>\n",
       "      <td>0.997856</td>\n",
       "      <td>1.425896</td>\n",
       "      <td>6.997913</td>\n",
       "      <td>0.740911</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-26</th>\n",
       "      <td>21.278009</td>\n",
       "      <td>0.996586</td>\n",
       "      <td>1.005026</td>\n",
       "      <td>1.567614</td>\n",
       "      <td>6.995387</td>\n",
       "      <td>0.779794</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">llama</th>\n",
       "      <th>dota-league</th>\n",
       "      <td>0.946996</td>\n",
       "      <td>0.572936</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.462707</td>\n",
       "      <td>1.494911</td>\n",
       "      <td>2.669909</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-22</th>\n",
       "      <td>1.011812</td>\n",
       "      <td>0.770541</td>\n",
       "      <td>NaN</td>\n",
       "      <td>10.556589</td>\n",
       "      <td>0.877886</td>\n",
       "      <td>4.724101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-24</th>\n",
       "      <td>1.487265</td>\n",
       "      <td>0.658144</td>\n",
       "      <td>NaN</td>\n",
       "      <td>21.262629</td>\n",
       "      <td>0.455884</td>\n",
       "      <td>14.893422</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-22</th>\n",
       "      <td>0.913087</td>\n",
       "      <td>0.946757</td>\n",
       "      <td>0.980745</td>\n",
       "      <td>2.143191</td>\n",
       "      <td>0.595346</td>\n",
       "      <td>8.051645</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-24</th>\n",
       "      <td>1.281126</td>\n",
       "      <td>0.696811</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2.261215</td>\n",
       "      <td>0.371342</td>\n",
       "      <td>20.498434</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"7\" valign=\"top\">graphone</th>\n",
       "      <th>dota-league</th>\n",
       "      <td>1.080176</td>\n",
       "      <td>1.053629</td>\n",
       "      <td>1.002889</td>\n",
       "      <td>1.143461</td>\n",
       "      <td>1.066358</td>\n",
       "      <td>0.841244</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-22</th>\n",
       "      <td>1.170737</td>\n",
       "      <td>1.162486</td>\n",
       "      <td>0.999213</td>\n",
       "      <td>3.155735</td>\n",
       "      <td>0.986472</td>\n",
       "      <td>0.810322</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-24</th>\n",
       "      <td>1.655716</td>\n",
       "      <td>1.388251</td>\n",
       "      <td>NaN</td>\n",
       "      <td>8.280799</td>\n",
       "      <td>1.014122</td>\n",
       "      <td>0.961226</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>graph500-26</th>\n",
       "      <td>1.318149</td>\n",
       "      <td>0.987389</td>\n",
       "      <td>NaN</td>\n",
       "      <td>14.282759</td>\n",
       "      <td>0.986895</td>\n",
       "      <td>1.175021</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-22</th>\n",
       "      <td>1.066395</td>\n",
       "      <td>1.496286</td>\n",
       "      <td>1.006629</td>\n",
       "      <td>1.306351</td>\n",
       "      <td>1.034623</td>\n",
       "      <td>0.318515</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-24</th>\n",
       "      <td>1.185394</td>\n",
       "      <td>0.669239</td>\n",
       "      <td>1.010907</td>\n",
       "      <td>1.583424</td>\n",
       "      <td>1.049177</td>\n",
       "      <td>0.158576</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>uniform-26</th>\n",
       "      <td>1.422569</td>\n",
       "      <td>1.000060</td>\n",
       "      <td>1.016527</td>\n",
       "      <td>1.681277</td>\n",
       "      <td>1.008271</td>\n",
       "      <td>0.108896</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "algorithm                   bfs      cdlp       lcc   pagerank      sssp  \\\n",
       "library  graph                                                             \n",
       "stinger  dota-league   2.021748  0.992872  1.000247   1.160999  4.824633   \n",
       "         graph500-22   3.452096  1.005264  1.001150   9.298210  8.026528   \n",
       "         graph500-24   8.824797  1.001041       NaN  15.082731  9.983933   \n",
       "         graph500-26  20.197229  0.978307       NaN  18.232644  9.997364   \n",
       "         uniform-22    3.853086  0.986295  1.010869   1.202704  5.610110   \n",
       "         uniform-24   11.512104  0.990429  0.997856   1.425896  6.997913   \n",
       "         uniform-26   21.278009  0.996586  1.005026   1.567614  6.995387   \n",
       "llama    dota-league   0.946996  0.572936       NaN   1.462707  1.494911   \n",
       "         graph500-22   1.011812  0.770541       NaN  10.556589  0.877886   \n",
       "         graph500-24   1.487265  0.658144       NaN  21.262629  0.455884   \n",
       "         uniform-22    0.913087  0.946757  0.980745   2.143191  0.595346   \n",
       "         uniform-24    1.281126  0.696811       NaN   2.261215  0.371342   \n",
       "graphone dota-league   1.080176  1.053629  1.002889   1.143461  1.066358   \n",
       "         graph500-22   1.170737  1.162486  0.999213   3.155735  0.986472   \n",
       "         graph500-24   1.655716  1.388251       NaN   8.280799  1.014122   \n",
       "         graph500-26   1.318149  0.987389       NaN  14.282759  0.986895   \n",
       "         uniform-22    1.066395  1.496286  1.006629   1.306351  1.034623   \n",
       "         uniform-24    1.185394  0.669239  1.010907   1.583424  1.049177   \n",
       "         uniform-26    1.422569  1.000060  1.016527   1.681277  1.008271   \n",
       "\n",
       "algorithm                   wcc  \n",
       "library  graph                   \n",
       "stinger  dota-league   0.324122  \n",
       "         graph500-22   0.446463  \n",
       "         graph500-24   0.612912  \n",
       "         graph500-26   0.841455  \n",
       "         uniform-22    0.500772  \n",
       "         uniform-24    0.740911  \n",
       "         uniform-26    0.779794  \n",
       "llama    dota-league   2.669909  \n",
       "         graph500-22   4.724101  \n",
       "         graph500-24  14.893422  \n",
       "         uniform-22    8.051645  \n",
       "         uniform-24   20.498434  \n",
       "graphone dota-league   0.841244  \n",
       "         graph500-22   0.810322  \n",
       "         graph500-24   0.961226  \n",
       "         graph500-26   1.175021  \n",
       "         uniform-22    0.318515  \n",
       "         uniform-24    0.158576  \n",
       "         uniform-26    0.108896  "
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Stinger\n",
    "stinger_native = summaries.loc[(\"stinger5\", \"gcc\", slice(None), slice(None), slice(None), 40), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "stinger_gapbs = summaries.loc[(\"stinger5-ref\", \"gcc\", slice(None), slice(None), slice(None), 40), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "stinger = pd.concat( [stinger_native / stinger_gapbs], names=['library'], keys=['stinger'] )\n",
    "# LLAMA\n",
    "llama_native = summaries.loc[(\"llama6\", \"gcc\", slice(None), 10.0, slice(None), 16), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "llama_ref = summaries.loc[(\"llama6-ref\", \"gcc\", slice(None), 10.0, slice(None), 16), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "llama = pd.concat( [llama_native / llama_ref] , names=['library'], keys=[\"llama\"] )\n",
    "# GraphOne\n",
    "g1_native = summaries.loc[(\"g1_v4-bw-sp-ignore-build\", \"gcc\", slice(None), 0, slice(None), 12), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "g1_gapbs = summaries.loc[(\"g1_v4-ref-ignore-build\", \"gcc\", slice(None), 0, slice(None), 12), \"completion_time\"].droplevel([0,1,3,4,5])\n",
    "g1 = pd.concat( [g1_native / g1_gapbs], names=['library'], keys=['graphone'] )\n",
    "\n",
    "# Final data\n",
    "speedup = pd.concat( [stinger, llama, g1 ])\n",
    "speedup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Final plot for the paper:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f6a2f6585b0>"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x244.8 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "pt2inc = 1/72\n",
    "fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(8, 3.4))\n",
    "\n",
    "plt.title(\"Speed-up of the GAP.BS algorithms\")\n",
    "plt.ylabel(\"Speed-up\")\n",
    "plt.ylim([0, 30])\n",
    "\n",
    "# Grid lines\n",
    "ax.set_axisbelow(True)\n",
    "plt.grid(b=True, which='both', color='0.80', linestyle='--')\n",
    "\n",
    "bar_width = 1\n",
    "gap_minor = 0.2\n",
    "# BFS\n",
    "start_bfs = 0;\n",
    "plt.bar(start_bfs + 0, speedup.loc[(\"stinger\", \"graph500-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"blue\")\n",
    "plt.bar(start_bfs + 0 + bar_width, speedup.loc[(\"stinger\", \"uniform-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"white\", hatch=\"////\", edgecolor=\"blue\")\n",
    "plt.bar(start_bfs + 2 * bar_width + gap_minor, speedup.loc[(\"llama\", \"graph500-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"orange\")\n",
    "plt.bar(start_bfs + 3 * bar_width + gap_minor, speedup.loc[(\"llama\", \"uniform-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"white\", hatch='////', edgecolor=\"orange\")\n",
    "plt.bar(start_bfs + 4 * bar_width + 2* gap_minor, speedup.loc[(\"graphone\", \"graph500-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"green\")\n",
    "plt.bar(start_bfs + 5 * bar_width + 2* gap_minor, speedup.loc[(\"graphone\", \"uniform-24\"), \"bfs\"], bar_width, align=\"edge\", color=\"white\", hatch='////', edgecolor=\"green\")\n",
    "\n",
    "# PageRank\n",
    "start_pagerank = start_bfs + 6 * bar_width + 4* gap_minor\n",
    "plt.bar(start_pagerank, speedup.loc[(\"stinger\", \"graph500-24\"), \"pagerank\"], bar_width, align=\"edge\", color=\"blue\")\n",
    "plt.bar(start_pagerank + 1 * bar_width, speedup.loc[(\"stinger\", \"uniform-24\"), \"pagerank\"], bar_width, align=\"edge\", color=\"white\", hatch=\"////\", edgecolor=\"blue\")\n",
    "plt.bar(start_pagerank + 2 * bar_width + gap_minor, speedup.loc[(\"llama\", \"graph500-24\"), \"pagerank\"], bar_width, align=\"edge\", color=\"orange\")\n",
    "plt.bar(start_pagerank + 3 * bar_width + gap_minor, speedup.loc[(\"llama\", \"uniform-24\"), \"pagerank\"], bar_width, align=\"edge\", color=\"white\", hatch='////', edgecolor=\"orange\")\n",
    "\n",
    "start_sssp = start_pagerank + 4 * bar_width + 4 *gap_minor\n",
    "plt.bar(start_sssp, speedup.loc[(\"stinger\", \"graph500-24\"), \"sssp\"], bar_width, align=\"edge\", color=\"blue\")\n",
    "plt.bar(start_sssp + 1 * bar_width, speedup.loc[(\"stinger\", \"uniform-24\"), \"sssp\"], bar_width, align=\"edge\", color=\"white\", hatch=\"////\", edgecolor=\"blue\")\n",
    "plt.bar(start_sssp + 2 * bar_width + gap_minor, speedup.loc[(\"llama\", \"graph500-24\"), \"sssp\"], bar_width, align=\"edge\", color=\"orange\")\n",
    "plt.bar(start_sssp + 3 * bar_width + gap_minor, speedup.loc[(\"llama\", \"uniform-24\"), \"sssp\"], bar_width, align=\"edge\", color=\"white\", hatch='////', edgecolor=\"orange\")\n",
    "\n",
    "start_wcc = start_sssp + 4 * bar_width + 4 *gap_minor\n",
    "plt.bar(start_wcc, speedup.loc[(\"stinger\", \"graph500-24\"), \"wcc\"], bar_width, align=\"edge\", color=\"blue\")\n",
    "plt.bar(start_wcc + 1 * bar_width, speedup.loc[(\"stinger\", \"uniform-24\"), \"wcc\"], bar_width, align=\"edge\", color=\"white\", hatch=\"////\", edgecolor=\"blue\")\n",
    "\n",
    "# Groups\n",
    "plt.xticks([\n",
    "    start_bfs + bar_width * 3 + gap_minor, # BFS\n",
    "    start_pagerank + (start_sssp - start_pagerank - 3* gap_minor) / 2.0, # PageRank\n",
    "    start_sssp + (start_wcc - start_sssp - 3 * gap_minor) / 2.0, # SSSP\n",
    "    start_wcc + bar_width, # WCC\n",
    "], [\"BFS\", \"PageRank\", \"SSSP\", \"WCC\"])\n",
    "\n",
    "# add a label at the top of each bar with its value\n",
    "for rect in plt.gca().patches:\n",
    "    value = rect.get_height();\n",
    "    plt.gca().annotate(common.fmtlabel(value) + \" x\", xy=(rect.get_x() + rect.get_width() / 2, value * 1.02), xytext=(0,3), textcoords=\"offset points\", ha=\"center\", va=\"bottom\", rotation=\"90\")\n",
    "\n",
    "# reduce the axes by 10% to leave some space to the legend\n",
    "box = ax.get_position()\n",
    "reduce_by = 0.05 # 0.1 == 10%\n",
    "ax.set_position([box.x0, box.y0 + box.height * reduce_by, box.width, box.height * (1 - reduce_by)])\n",
    "    \n",
    "# insert the legend at the bottom of the plots\n",
    "fig.legend(handles=[\n",
    "    Patch(color=\"blue\", label=\"Stinger\"),\n",
    "    Patch(color=\"orange\", label=\"LLAMA\"), \n",
    "    Patch(color=\"green\", label=\"GraphOne\"),\n",
    "    Patch(color=\"white\", label=\"Graph:\"),\n",
    "    Patch(color=\"black\", label=\"graph500\"), \n",
    "    Patch(facecolor=\"white\", edgecolor=\"black\", hatch=\"////\", label=\"uniform\"), \n",
    "], loc=\"lower center\", ncol = 12)\n",
    "\n",
    "# plt.savefig(\"/home/dean/workspace/eclipse/paper_gfe/img/plot_graphalytics_gapbs.svg\", bbox_inches='tight')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}