notebooks/plot_bias_amp.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "fed89b1d",
   "metadata": {},
   "source": [
    "## Plot bias amplification after fine-tuning\n",
    "\n",
    "Code for generating Figure 4, 6, 7, 8 and 9"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "98a95f66",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.lines import Line2D\n",
    "from matplotlib.patches import Patch\n",
    "from matplotlib.ticker import MaxNLocator\n",
    "import matplotlib.patches as mpatches\n",
    "\n",
    "matplotlib.rcParams['text.usetex'] = True\n",
    "plt.rc('font', family='serif', serif=['Times'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "f9851330",
   "metadata": {},
   "outputs": [],
   "source": [
    "models = [\n",
    "    '$\\\\mathrm{\\\\textsc{LXMERT_{180K}}}$', #0\n",
    "    '$\\\\mathrm{\\\\textsc{LXMERT_{180K}^{N}}}$', #4\n",
    "    '$\\\\mathrm{\\\\textsc{LXMERT_{3M}}}$', #1\n",
    "    '$\\\\mathrm{\\\\textsc{LXMERT_{3M}^{N}}}$', #5\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{4M}}}$', #2\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{4M}^{N-COCO}}}$', #6\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{4M}^{N-CC3M}}}$', #7\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{14M}}}$', #2\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{14M}^{N-COCO}}}$', #6\n",
    "    '$\\\\mathrm{\\\\textsc{ALBEF_{14M}^{N-CC3M}}}$', #7\n",
    "    '$\\\\mathrm{\\\\textsc{BLIP_{129M}}}$', #3\n",
    "    '$\\\\mathrm{\\\\textsc{BLIP_{129M}}^{N}}$' #8\n",
    "]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d0dea6e",
   "metadata": {},
   "source": [
    "# VQAv2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "9e8c3df2",
   "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>M</th>\n",
       "      <th>F</th>\n",
       "      <th>N</th>\n",
       "      <th>model</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>lxmert</th>\n",
       "      <td>-0.029297</td>\n",
       "      <td>-0.068376</td>\n",
       "      <td>0.004679</td>\n",
       "      <td>lxmert</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>lxmert_n</th>\n",
       "      <td>-0.019950</td>\n",
       "      <td>-0.051893</td>\n",
       "      <td>0.000668</td>\n",
       "      <td>lxmert_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concaplxmert</th>\n",
       "      <td>-0.020508</td>\n",
       "      <td>-0.064713</td>\n",
       "      <td>-0.037433</td>\n",
       "      <td>concaplxmert</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concaplxmert_n</th>\n",
       "      <td>-0.020787</td>\n",
       "      <td>-0.074481</td>\n",
       "      <td>-0.023619</td>\n",
       "      <td>concaplxmert_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef4m</th>\n",
       "      <td>-0.020787</td>\n",
       "      <td>-0.068071</td>\n",
       "      <td>0.000446</td>\n",
       "      <td>albef4m</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef4m_n</th>\n",
       "      <td>-0.025530</td>\n",
       "      <td>-0.074786</td>\n",
       "      <td>-0.006907</td>\n",
       "      <td>albef4m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concapalbef4m_n</th>\n",
       "      <td>-0.023996</td>\n",
       "      <td>-0.065934</td>\n",
       "      <td>-0.008467</td>\n",
       "      <td>concapalbef4m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef14m</th>\n",
       "      <td>-0.034738</td>\n",
       "      <td>-0.074786</td>\n",
       "      <td>0.000446</td>\n",
       "      <td>albef14m</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef14m_n</th>\n",
       "      <td>-0.027902</td>\n",
       "      <td>-0.071123</td>\n",
       "      <td>-0.000668</td>\n",
       "      <td>albef14m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concapalbef14m_n</th>\n",
       "      <td>-0.033901</td>\n",
       "      <td>-0.076007</td>\n",
       "      <td>-0.006907</td>\n",
       "      <td>concapalbef14m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blip</th>\n",
       "      <td>-0.019671</td>\n",
       "      <td>-0.016178</td>\n",
       "      <td>-0.034759</td>\n",
       "      <td>blip</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blip_n</th>\n",
       "      <td>-0.016323</td>\n",
       "      <td>-0.016789</td>\n",
       "      <td>-0.032531</td>\n",
       "      <td>blip_n</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                         M         F         N             model\n",
       "lxmert           -0.029297 -0.068376  0.004679            lxmert\n",
       "lxmert_n         -0.019950 -0.051893  0.000668          lxmert_n\n",
       "concaplxmert     -0.020508 -0.064713 -0.037433      concaplxmert\n",
       "concaplxmert_n   -0.020787 -0.074481 -0.023619    concaplxmert_n\n",
       "albef4m          -0.020787 -0.068071  0.000446           albef4m\n",
       "albef4m_n        -0.025530 -0.074786 -0.006907         albef4m_n\n",
       "concapalbef4m_n  -0.023996 -0.065934 -0.008467   concapalbef4m_n\n",
       "albef14m         -0.034738 -0.074786  0.000446          albef14m\n",
       "albef14m_n       -0.027902 -0.071123 -0.000668        albef14m_n\n",
       "concapalbef14m_n -0.033901 -0.076007 -0.006907  concapalbef14m_n\n",
       "blip             -0.019671 -0.016178 -0.034759              blip\n",
       "blip_n           -0.016323 -0.016789 -0.032531            blip_n"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lxmert=[-0.02929688,-0.06837607, 0.00467914]\n",
    "lxmert_n=[-0.01994978,-0.05189255,0.00066845]\n",
    "lxmert_3m=[-0.02050781,-0.06471306,-0.03743316]\n",
    "lxmert_3m_n=[-0.02078683,-0.07448107,-0.02361854]\n",
    "albef_4m=[-0.02078683,-0.06807082,0.00044563]\n",
    "albef_4m_coco=[-0.02553013,-0.07478632,-0.00690731]\n",
    "albef_4m_cc3m=[-0.02399554,-0.06593407,-0.00846702]\n",
    "albef_14m=[-0.03473772,-0.07478632,0.00044563]\n",
    "albef_14m_coco=[-0.02790179,-0.07112332,-0.00066845]\n",
    "albef_14m_cc3m=[-0.03390067,-0.07600733,-0.00690731]\n",
    "blip=[-0.01967076,-0.01617827,-0.03475936]\n",
    "blip_coco=[-0.01632254,-0.01678877,-0.03253119]\n",
    "\n",
    "# Visualize data\n",
    "data_ = [lxmert, lxmert_n,\n",
    "         lxmert_3m, lxmert_3m_n,\n",
    "         albef_4m, albef_4m_coco, albef_4m_cc3m,\n",
    "         albef_14m, albef_14m_coco, albef_14m_cc3m,\n",
    "         blip, blip_coco]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"albef4m_n\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"albef14m_n\", \"concapalbef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                  columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "085f9693",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "name=\"vqa_dba\"\n",
    "\n",
    "data_ = [\n",
    "         lxmert, lxmert_n,\n",
    "         albef_4m, albef_4m_coco,\n",
    "         albef_14m, albef_14m_coco,\n",
    "         blip, blip_coco]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"albef4m\", \"albef4m_n\", \n",
    "          \"albef14m\", \"albef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                  columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 7.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 5.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(5.5+bar_width, 7.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width, 6.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{180K}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{BLIP}_{129M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(df)))\n",
    "xs = [x[i] for i in range(3) for x in df.values]\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    ys = [x for x in df[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(df)/2), \n",
    "            edgecolor='black') \n",
    "        \n",
    "\n",
    "ax0.set_ylabel('$\\\\mathrm{BiasAmp_{ A\\\\rightarrow{T}}}$', fontsize=20)\n",
    "ax0.set_ylim(min(xs)-0.003, max(xs)+0.002)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=20)\n",
    "ax0.grid(color='black', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$'),\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='lower right')\n",
    "\n",
    "plt.savefig(f'figures/vqa_dba_hist_coco.pdf', bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "81f37135",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "name=\"vqa_dba\"\n",
    "\n",
    "data_ = [\n",
    "         lxmert, lxmert_n,\n",
    "         albef_14m, albef_14m_coco,\n",
    "         blip, blip_coco]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"albef14m\", \"albef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                  columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 5.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 3.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(3.5+bar_width, 5.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{3M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(df)))\n",
    "xs = [x[i] for i in range(3) for x in df.values]\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    ys = [x for x in df[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(df)/2), \n",
    "            edgecolor='black') \n",
    "        \n",
    "\n",
    "ax0.set_ylabel('$\\\\mathrm{BiasAmp_{ A\\\\rightarrow{T}}}$', fontsize=20)\n",
    "ax0.set_ylim(min(xs)-0.003, max(xs)+0.002)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=20)\n",
    "ax0.grid(color='black', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$'),\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='lower right')\n",
    "\n",
    "plt.savefig(f'figures/vqa_dba_hist_cc3m.pdf', bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f4c288e",
   "metadata": {},
   "source": [
    "# GQA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "ea45c31d",
   "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>M</th>\n",
       "      <th>F</th>\n",
       "      <th>N</th>\n",
       "      <th>model</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>lxmert</th>\n",
       "      <td>-0.280802</td>\n",
       "      <td>-0.322816</td>\n",
       "      <td>-0.319380</td>\n",
       "      <td>lxmert</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>lxmert_n</th>\n",
       "      <td>-0.284339</td>\n",
       "      <td>-0.312413</td>\n",
       "      <td>-0.323148</td>\n",
       "      <td>lxmert_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concaplxmert</th>\n",
       "      <td>-0.279623</td>\n",
       "      <td>-0.316227</td>\n",
       "      <td>-0.312265</td>\n",
       "      <td>concaplxmert</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concaplxmert_n</th>\n",
       "      <td>-0.285321</td>\n",
       "      <td>-0.324549</td>\n",
       "      <td>-0.336961</td>\n",
       "      <td>concaplxmert_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef4m</th>\n",
       "      <td>-0.004323</td>\n",
       "      <td>-0.086685</td>\n",
       "      <td>-0.005023</td>\n",
       "      <td>albef4m</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef4m_n</th>\n",
       "      <td>0.037335</td>\n",
       "      <td>-0.028433</td>\n",
       "      <td>-0.046463</td>\n",
       "      <td>albef4m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concapalbef4m_n</th>\n",
       "      <td>-0.003734</td>\n",
       "      <td>-0.121012</td>\n",
       "      <td>-0.007535</td>\n",
       "      <td>concapalbef4m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef14m</th>\n",
       "      <td>0.056200</td>\n",
       "      <td>0.025659</td>\n",
       "      <td>-0.055672</td>\n",
       "      <td>albef14m</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>albef14m_n</th>\n",
       "      <td>0.051091</td>\n",
       "      <td>-0.027393</td>\n",
       "      <td>-0.060695</td>\n",
       "      <td>albef14m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>concapalbef14m_n</th>\n",
       "      <td>-0.060523</td>\n",
       "      <td>-0.100555</td>\n",
       "      <td>0.013395</td>\n",
       "      <td>concapalbef14m_n</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blip</th>\n",
       "      <td>-0.066614</td>\n",
       "      <td>-0.079057</td>\n",
       "      <td>0.011720</td>\n",
       "      <td>blip</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blip_n</th>\n",
       "      <td>0.034977</td>\n",
       "      <td>-0.013523</td>\n",
       "      <td>-0.026789</td>\n",
       "      <td>blip_n</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                         M         F         N             model\n",
       "lxmert           -0.280802 -0.322816 -0.319380            lxmert\n",
       "lxmert_n         -0.284339 -0.312413 -0.323148          lxmert_n\n",
       "concaplxmert     -0.279623 -0.316227 -0.312265      concaplxmert\n",
       "concaplxmert_n   -0.285321 -0.324549 -0.336961    concaplxmert_n\n",
       "albef4m          -0.004323 -0.086685 -0.005023           albef4m\n",
       "albef4m_n         0.037335 -0.028433 -0.046463         albef4m_n\n",
       "concapalbef4m_n  -0.003734 -0.121012 -0.007535   concapalbef4m_n\n",
       "albef14m          0.056200  0.025659 -0.055672          albef14m\n",
       "albef14m_n        0.051091 -0.027393 -0.060695        albef14m_n\n",
       "concapalbef14m_n -0.060523 -0.100555  0.013395  concapalbef14m_n\n",
       "blip             -0.066614 -0.079057  0.011720              blip\n",
       "blip_n            0.034977 -0.013523 -0.026789            blip_n"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# GQA\n",
    "lxmert=[-0.28080173,-0.32281553,-0.31938049] #M,F,N\n",
    "lxmert_n=[-0.28433877,-0.31241331,-0.32314776]\n",
    "lxmert_3m=[-0.27962272,-0.31622746,-0.31226455]\n",
    "lxmert_3m_n=[-0.28532128,-0.32454924,-0.33696107]\n",
    "albef_4m=[-0.00432305,-0.08668516,-0.00502302]\n",
    "albef_4m_coco=[0.03733543,-0.02843273,-0.04646296]\n",
    "albef_4m_cc3m=[-0.00373354,-0.12101248,-0.00753453]\n",
    "albef_14m=[0.05619965,0.02565881,-0.05567183]\n",
    "albef_14m_coco=[0.05109059, -0.02739251, -0.06069485]\n",
    "albef_14m_cc3m=[-0.0605227,-0.10055479,0.01339473]\n",
    "blip=[-0.06661427,-0.07905687,0.01172039]\n",
    "blip_coco=[0.0349774,-0.01352288,-0.02678945]\n",
    "\n",
    "# Visualize data\n",
    "data_ = [lxmert, lxmert_n,\n",
    "         lxmert_3m, lxmert_3m_n,\n",
    "         albef_4m, albef_4m_coco, albef_4m_cc3m,\n",
    "         albef_14m, albef_14m_coco, albef_14m_cc3m,\n",
    "         blip, blip_coco]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"albef4m_n\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"albef14m_n\", \"concapalbef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "97e9f257",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "name=\"gqa_dba\"\n",
    "\n",
    "data_ = [\n",
    "         lxmert, lxmert_n,\n",
    "         albef_4m, albef_4m_coco,\n",
    "         albef_14m, albef_14m_coco,\n",
    "         blip, blip_coco]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"albef4m\", \"albef4m_n\", \n",
    "          \"albef14m\", \"albef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 7.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 5.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(5.5+bar_width, 7.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width, 6.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{180K}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{BLIP}_{129M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(df)))\n",
    "xs = [x[i] for i in range(3) for x in df.values]\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    ys = [x for x in df[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(df)/2), \n",
    "            edgecolor='black') \n",
    "        \n",
    "\n",
    "ax0.set_ylabel('$\\\\mathrm{BiasAmp_{ A\\\\rightarrow{T}}}$', fontsize=20)\n",
    "ax0.set_ylim(min(xs)-0.01, max(xs)+0.01)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=20)\n",
    "ax0.grid(color='black', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$'),\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='lower right')\n",
    "\n",
    "plt.savefig(f'figures/gqa_dba_hist_coco_all.pdf', bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "94af052a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "name=\"gqa_dba\"\n",
    "\n",
    "data_ = [lxmert_3m, lxmert_3m_n,\n",
    "         albef_4m, albef_4m_cc3m,\n",
    "         albef_14m, albef_14m_cc3m]\n",
    "index_ = [\"lxmert\", \"lxmert_n\",\n",
    "          \"albef4m\", \"albef4m_n\", \n",
    "          \"albef14m\", \"albef14m_n\"]\n",
    "df = pd.DataFrame(data=data_, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "df[\"model\"] = df.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 5.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 3.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(3.5+bar_width, 5.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{3M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(df)))\n",
    "xs = [x[i] for i in range(3) for x in df.values]\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    ys = [x for x in df[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(df)/2), \n",
    "            edgecolor='black') \n",
    "        \n",
    "\n",
    "ax0.set_ylabel('$\\\\mathrm{BiasAmp_{ A\\\\rightarrow{T}}}$', fontsize=20)\n",
    "ax0.set_ylim(min(xs)-0.01, max(xs)+0.01)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=20)\n",
    "ax0.grid(color='black', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$'),\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='lower right')\n",
    "\n",
    "plt.savefig(f'figures/gqa_dba_hist_cc3m.pdf', bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3ccfeb5",
   "metadata": {},
   "source": [
    "# F30K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "deeb0248",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#IR\n",
    "lxmert_original=[20.0652528548124, 11.419249592169658, 62.31647634584013]\n",
    "lxmert_original_ckpt=[19.086460032626427, 11.582381729200652, 62.31647634584013]\n",
    "lxmert_cc_ctrl=[18.92332789559543, 14.681892332789559, 57.91190864600326]\n",
    "lxmert_cc_ctrl_ckpt=[19.249592169657422, 13.050570962479608, 60.52202283849919]\n",
    "albef_4m=[18.270799347471453, 8.64600326264274, 68.35236541598695]\n",
    "albef_4m_coco_neutral=[17.94453507340946, 9.135399673735726, 67.69983686786297]\n",
    "albef_4m_cc3m_neutral=[17.94453507340946, 8.64600326264274, 69.16802610114192]\n",
    "albef_14m=[16.802610114192497, 8.482871125611746, 69.9836867862969]\n",
    "albef_14m_coco_neutral=[16.6394779771615, 8.31973898858075, 70.47308319738988]\n",
    "albef_14m_cc3m_neutral=[16.802610114192497, 9.624796084828711, 68.67862969004894]\n",
    "blip=[15.660685154975528, 9.624796084828711, 69.9836867862969]\n",
    "blip_coco_neutral=[15.98694942903752, 8.64600326264274, 69.65742251223492]\n",
    "\n",
    "data_ir = [lxmert_original, lxmert_original_ckpt,\n",
    "         albef_4m, albef_4m_coco_neutral,\n",
    "         albef_14m, albef_14m_coco_neutral,\n",
    "         blip, blip_coco_neutral]\n",
    "index_ = [\"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"concapalbef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "\n",
    "dfir = pd.DataFrame(data=data_ir, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "dfir[\"model\"] = dfir.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 7.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 5.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(5.5+bar_width, 7.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width, 6.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{180K}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{BLIP}_{129M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(dfir)))\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    ys = [x for x in dfir[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(dfir)/2), \n",
    "            edgecolor='black')  \n",
    "\n",
    "ax0.set_ylabel('\\%', fontsize=15)\n",
    "ax0.set_ylim(0, 80)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=18)\n",
    "ax0.grid(color='gray', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$')\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='upper right', bbox_to_anchor=(1.19,1.0))\n",
    "plt.savefig(f'figures/f30kIR_not_dba_hist_coco.pdf', bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "264f207e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAjgAAAGoCAYAAABL+58oAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAAsTAAALEwEAmpwYAAA/nUlEQVR4nO3df5DU1Z3v/9dhCCjIMDM01ygYZVBDgCw6qFVyJ7XxznATNlpFshAXLCNY1+Ga/SarkcAmsLlZV292ckmumu9Xv8CtDLuW7pUhZKpC+eMLVEgyWRMUEkszubs6IyokRpqZYdARFTjfP/rzGZsP3TPdn9P96f50Px9VXUN/+vSnz3x8e857zvmc08ZaKwAAgEoyrtQVAAAAKDQSHAAAUHFIcAAAQMUhwQEAABWHBAcAAFQcEhwAAFBxxkf5YcaYdZKuldQnqVHSE9baHYEyyyTd7JWpk7TZWnswynoCAIB4iyzB8ZMba+3ytGMDxpg+P4HxkputkmZZaweNMY2SDhhjWkhyAABArqKcoloj6bnAsecltaY9b5e03Vo7KEnW2j5Je7zjAAAAOYkywRlUKslJ1yrJH71pUmraanegzG5JrcaYuiLXDwAAVIgoE5z1khqNMQeMMU3GmN2S1ltr93ivN3o/+wLv6/d+XhNFJQEAQPxFdg+OtXaPMWaxUiMyByR911r73bQi13o/BwNv9Z83ph80xrRJapOkK668cuHTP90joJguODlU6iogZk5OaCh1FRBDH5t5kSl1HSpBpKuolFoVtUNSk6R1xpg6a60/bXUsnxNZa7dI2iJJVy9ssg2JaYWsJ3COqcM1pa4CYmZ4YqLUVQCqVpSrqNokrbHWLvSed0pqM8b0eiM5/tRUXeCt/vPno6gnAACIvyjvwWmX9IT/xFsuflAf3njsJzjBMd2GwOsAAACjinon48HA85EbZ7x9bgYlLQ6UWSzpoL90HAAAYCxRJjjbde4y8SZJm9Oe36HUtFWdJHk/W73jAAAAOYlyFdUaY0y7d+/Nc5JmS9qdvpLKWrvDGCNJ7caYXqVWVi1nF2MAAJCPSFdRWWvX51Bmh1IrrQAAiLV33j6hwYFjOnXqA9kc33Pqgw8OFbNOFeKMMebNU6dO/X1TU9MzmQpEvUwcAICq8M7bJzTQn9SMmTN03nnnyZuhGFPNuHHJIlct9s6cOWPefffdqYcOHfq/Dx48+H9lSnKivskYAICqMDhwTDNmztD555+fc3KD3IwbN85Onjz53csuu+z98ePH/7eMZaKuFAAA1eDUqQ903nnnlboaFe38888/aa39aKbXSHAAACgCKzFyU2Tjxo2zypLLkOAAAICKw03GAABE6NJZl+nI62+MVmRhIT/vwosvOvXmkT+8UMhzxgEJDgAAETry+hv63ovPRvZ593zy+sj7+p6engn9/f3jm5ubh6P+bB9TVAAAoGC6u7snfeYzn7nyscceqy9lPUhwAABAwTQ3Nw/Pnz+/ZCM3PhIcAAAgSerq6poyb968T2zatClx5513zrjkkkvmb9q0KdHV1TVlyZIljbW1tVd1d3dP8stv3Ljxwo6OjvqVK1deunHjxgtHO+/GjRsvXLRo0RUrV668NIrfhQQHAABIkpYuXXpiaGiopre3d+Ijjzxy5Ac/+MFrX//61y9NJBKnn3rqqb5bbrnl6MMPPzxdSt1n89BDD120evXqgYceeujw/fffPzPTOXt6eiY888wztffdd9+f/vVf//XlXbt21Xd0dBR9+oqbjAEAwIja2trT11133bCUmm5K/zl79uz3du7cOUmS5s6d+/6vfvWrHil1340kJZPJmkQicTr9fI8//nj9oUOHJvojPF/96lf/eMUVV7xX7N+DBAcAAITS398/fsmSJTNbWlqGspXp6+s7r6WlZWjt2rWRfscWU1QAACBvPT09E1asWNH41FNP9Y2WvEydOvXUzp07z5qSSr+Pp1hIcAAAwIihoaGa4LFkMjly7Pjx4+Ml6cknn6z1j/X09EzIdr5bbrll4Nlnn63dtGlTQpI6Ojrq089XLExRAQAQoRkfu0T3fPL6yD7vwosvOpVr2a6urimHDx+euH379vqbbrpp6IEHHkhI0gMPPJC46667kjt37qx/4403JnZ3d09atWrVwIMPPvjRefPmfWLt2rVvzp07d/jWW2+9dMOGDW/+8pe/rH399dff6+npOdrc3Dy8YcOGw/fee+/Me++9d+a3vvWtw6tXrx4o3m+cYqy1xf6Mort6YZP96bPdpa4GKtzU4cFSVwExMzxxeqmrgBI69OrLmjNnTt7vqxk37kARqlOxXnjhhcSCBQsuCx5nigoAAFQcEhwAAFBxSHAAAEDFIcEBAAAVhwQHAABUHBIcAABQcUhwAABAxSHBAQAAFYedjAEAiNDsj31Mrx05MlqRhYX8vBkXXnjq8JtvvlDIc8YBCQ4AABF67cgRffDtr0X2eR/59vfz7uv974168MEHP/qb3/zm94lE4nTha5bS09Mzob+/f3xzc/NwIc/LFBUAABjR0dFRf/Dgwclr165N/su//EtfMZOb7u7uSZ/5zGeufOyxx+rHLp0fEhwAADBi//79k6ZOnXpKkgo9qhLU3Nw8PH/+/KJ8BgkOAACQlBq9+fnPf17785//vHbjxo0XJpPJGin1LeMbN268cNGiRVesXLnyUv/YvHnzPrFp06bEnXfeOeOSSy6Zv2nTpkRXV9eUJUuWNNbW1l7V3d09yT/3xo0bL+zo6KhfuXLlpRs3brwwWx0yfVYY3IMDAAAkSatXrx7YvXt3rSTdd999f5JS98g888wztY888sgRSX+qra29qqOjY2j16tUDX/nKV2p6e3snPvLII0c+85nPDH3+85+/8he/+MXvn3rqqb4777xzxsMPPzy9ubn5tZ6engkPPfTQRUNDQ7+96aabhqZPn36Vf/50o31Wvr8LCQ4AAMjq8ccfrz906NBEf9Tlq1/96h+vuOKK9ySptrb29HXXXTcsfTid5f+cPXv2ezt37pwkSXPnzn3/V7/6VY+Uuu9GkpLJZE3w/p7RPitfJDgAACCrvr6+81paWobWrl2bdDlPf3//+CVLlsxsaWkZKvZnSdyDAwAARjF16tRTO3fuPGuVU/q9Nbno6emZsGLFisannnqqb7TkpRCf5SPBAQAAWd1yyy0Dzz77bK2/N05HR0e9f/Px0NBQTbC8/5okHT9+fLwkPfnkk7X+sZ6englhPitfkU1RGWMOSGrK8NKgtbY+rdwySTdL6pNUJ2mztfZgJJUEAKDILp0xQx/59vcj+7wZF154Ktey3d3dk37xi19M8f/d3Nw83NzcPLxhw4bD995778x777135re+9a3Dq1evHujq6ppy+PDhidu3b6+/6aabhh544IGEJD3wwAOJu+66K7lz5876N954Y2J3d/ekVatWDTz44IMfnTdv3ifWrl375ty5c4dvvfXWSzds2PDmL3/5y9rXX3/9vZ6enqPZPivM722stWHel9+HGNMkqV1Sp6T+tJdultRvrV3jlVsmaaukWdbaQWNMo6QDklpGS3KuXthkf/psd9HqD0jS1OHBUlcBMTM8cXqpq4ASOvTqy5ozZ07e76sZN+5AEapTsV544YXEggULLgsej2oEp9Fauzh40BjzDUmb0w61S9purR2UJGttnzFmj3f8nPcDAABkEsk9ONbaHcFj3uhMo7V2j/e8SVKjpN2BorsltRpj6opdTwAAUBlKeZPxMkl70p43ej/7AuX8Ka1ril4jAABQEUq5D87Nkr6T9vxa7+dgoJz/vDH9oDGmTVKbJF08Y4b6k8cKX0Mgzfsns27dAGR0coIpdRXGdONf/Gf98Y9v5lz+oos+ql1P/n9FrFHlOHP6jE6dyvn+3hE1E7IuMkIeSpLgeNNNTYGpq7wyFGvtFklbpNRNxg2JaYWrIJDB1OFQKxVRxYYnJkpdhTG9+OJLsvYnOZc35iZNm1b+v1c5ODE0oPHj2U+3VEo1RdUmKXhfjj81VRc47j9/voj1AQAAFaRUCc7Nkp4IHPMTnIbA8YbA6wAAAKOKfOwsy/SUrLUHjTGDSi0HT7/5eLGkg/7ScQAIY9aCa/Xa63/IufylH7tYr77wXBFrBKCYSjE52KazE5h0d0jaaoz5jrfRX52kVkktUVUOQGV67fU/yNqfaN++F7V8+T+qs/Nv9elPfzJreWNuirB2qCazGy/Ta6+9MVqRhYX8vBkzLjp1+PAfXijkOaPS09Mzob+/f7z/DeX5KEWCc7PO3txvhLV2hzFGktqNMb1Kraxazlc1ACiEXJMbQJLmzmnU668fzrn8xz42Uz3/Z+y7KV577Y28bux2ZcxNJb3Tubu7e9KcOXPeSyQSp/N934oVKxpvvPHGgVgkONbaUTNTb+rqnI0BAcAVyQ3y8frrh0cSkVySY0b9zpVMJmtWrFjR+Jvf/Ob3+b63ubl5eP78+XknNj6+TRyoILMWXKtx9TNyfsxacO3YJ60guSY3+/a9GEFtEBfVNPLX1dU1Zd68eZ/o6OioX7JkSWNtbe1VXV1dU9Jf37hx44WLFi26YuXKlZdK0qZNmxLGmIU9PT0Tenp6JixatOiKRYsWXSFJP/nJT2oPHz488e/+7u8+2t3dPamjo6N+0aJFV3R0dNRfcskl8zdu3HihJG3cuPHCjo6O+pUrV17qH3PFAn2ggvj3meSq2v7izDW5Wb78HyOoDeKgmpIbSVq6dOmJz3/+85N6e3snPPXUU3133nnnjA0bNsxcunTp73t6eiY888wztY888sgRSX+qra29qqOjY2jt2rXJr3/965dK0ty5c9//whe+MLBz5856SVq9evXA7bffrn/4h394M5FInJ4zZ857t99+e+OCBQuGn3nmmX/v7+8f39PTM+Ghhx66aGho6Lc33XTT0PTp06+67777/uT6u5DgAIAnvTO74YZvlro6KAPVlNz4pkyZcvqaa64ZlqTrrrtueNeuXfWS9Pjjj9cfOnRooj/C8tWvfvWPV1xxxXv5nNu/D+eWW24ZmDt37vuS3pekX/3qVz1S6r4bKTW1le89O0EkOACg6vtLHbmp9mnN+vr6ke+a6OvrO6+lpWVo7dq1yUJ/Tn9///glS5bMbGlpKdh34nAPDoCqR3KDbJjW/NDUqVNP+VNPPn/EJej48eNnDaC89dZbWb/rpqenZ8KKFSsan3rqqb5CJk8kOACqGskNXKTHTyUaGBgYSVRuueWWgWeffbZ206ZNCUnq6OioTyaTNVJqWuvXv/71ZEnau3dvbU9Pz6SOjo6RZKi/vz/rjNGTTz5Z6/+7p6enYN80SoIDoGqR3MBFJcZPV1fXlBMnTtRs3769IZlM1mzfvr3+8OHDE7u7uyc1NzcPb9iw4fC99947s7a29qpjx47VLF269IQkfetb3zr8N3/zN5cuWbKksaWlZej6668f8qe3VqxYkVyzZs2lXV1dU/yk5+GHH57uJ0erVq0akKR58+Z94te//vXkuXPnDt96662Xdnd3T/rlL39Z+/Of/7w2TOJjrLWFuzIlcvXCJvvTZ7tLXQ1UuKnDg6WuwpjG1c/IexXVmYEjRaxR+Qhem7E6p0Jcm+GJ053eH4ULJk3IO2beHn6/iDUqH6Ndm0zxE7w2h159WXPmzDnnvTnsZFxQcd7JOBcvvPBCYsGCBZcFj3OTMYCqU4l/eSM6rvHT23do1Ndrxo07ELJqSMMUFYCqQnIDF8RPfJDgAKgadE5wQfzECwkOgKpB54SwSG7ihwQHQNWo9k3bEA7JTTyR4ACoGmzahnyR3MQXCQ4AeCp90zbkj+QmvkhwAED8pY7MmNaMLxIcAFWP5AbZMK0ZX2z0B6CqkdzARXr83HDDN3N6T+8rr+j06dOjFVlYkMp5ampqTl155ZU57WTc1dU15Stf+cqlhw8fnvi73/3uxblz5561bXUymayZPn36VVOmTDn94IMPvrZ69eqBQta1kBjBAVC1SG7gImz8jJHcFNzp06dzHsxYunTpiRtvvHFAkn7wgx+c810j27Ztq58yZcrp+fPnv1POyY1EggOgSpHcwEUlx8+0adNOffaznx147LHHzklw9u7dWzt//vx3SlGvfJHgAKg6ldw5ofiqIX7WrFlz9MSJEzWbNm1K+Me6u7sntbS0DE2dOjXaIaiQSHAAVJVq6JxQPNUSP83NzcPXX3/9UEdHx8gozsMPPzx97dq1yVLWKx8kOACqRrV0TiiOaoufdevWvdnT0zOpu7t7UjKZrCl1ffLFKioAVaOaOicUVrUlN1LqhuOZM2e+d//993+0paVl6Mtf/vLRUtcpH4zgAKgabNqGMKotuTl27NjI4Mdtt9129Omnn67fu3dvbXNz83Ap65UvEhwAVYNN25CvaktufG+99VaNJN11111JSWppaRnyXzt+/HgspqtIcADAw3dRIagYyU1NTbT5QU1Nzalcy3Z1dU3ZtWtX/T333DMzmUzWJBKJ0ytWrEj6Nxd3dHTUv/TSS5NfeumlyR0dHfXFq7U77sEBAFXvX+oYXTGmNWdffvmor9eMG3cg55MV2NKlS08sXbr0pfRjjz/++Gv+v1evXj1Q7hv8+RjBAVD1SG6QDdOa8UWCA6CqkdzABdOa5YsEB0DVIrmBC+KnvJHgAKhKdE5wQfyUPxIcAFWHzgkuiJ94IMEBUFXonOCC+ImPkiY4xphGY0xrKesAoHrQOcEF8RMvJdkHxxjTJmmNpM2StgdeWybpZkl9kuokbbbWHoy6jgAqD50TwipUcmOt1dtvvz16mTNnCjr4MHXq1DPjxhVnPKOjo6P+9ttvb/zxj3/870uXLj0hSXfeeeeM48ePj0/fP6cUIk1wjDF1kg5IOmitXZjh9WWStkqaZa0dNMY0SjpgjGkhyQHgiu+iQhiFHLl5++23NfzOO5oyZcrIsV/84he69dZb9eijj+pTn/qUNG7c1aOd45zyozhx4oQk/aa+vv6MU8WzWL169cCmTZuGv/SlL83u6+t7MZFInL7lllsGyuHbx6OeojogadBauzzL6+2StltrByXJWtsnaY93HJAkLbjy46qfODnnx4IrP17qKqNMsGkb8lWMaakpU6aMPA4cOKDbbrtNP/rRj/QXf/EXZ72W6ZFv+d/+9rd51a2rq2vKJZdcMt8Ys7Cnp2dC8PVkMlljjFlYW1t7lf9VDZ///Of7b7zxxoE///M/v1KSGhoacv5qiGKKLMExxmyW1CjpjiyvN3mv7w68tFtSqzf6A+j11w7L2p9kfPz0p/9diUStfvrT/z5y7PXXDpe6yogJNm1DUDGnNfft26fly5ers7NTn/70p4tS/tZbb82rTkuXLj1x4403DkjSD37wg+nB17dt21Y/ZcqU0/Pnz38n/Ssb/OmolStXXprXBxZRlCM4bZIGJV1jjNntPdrSXm/0fvYF3tfv/bymyPVDzHEDIFwQP8ikWNOaUSQ3y5cv16OPPppXvSRp2rRppz772c8OPPbYY+ckOHv37q2dP3/+O5ne97Of/ezfd+3aVX/fffddlPeHFkEk9+CkrZTqk/S8UjcWf1HSZmPMQmvtGknXemUGA2/3nzemH/SSozZJunjGDPUnjxW+4ihLiURCR4++e9ax/ftf1t13b9W2bes1b97lZ72eSCQKEh/vnxxyPkexZbo2Y5bvj8X35jkb7dpkip9CXJuTE4zT+6MQJmaOHUsWsUblI5FInNOeZOLHT/DanDl9RqdOnT1bc+rUKf3rr3+t2267rejJTWdnpxYuXHhOHXKxZs2ao08//XT9pk2bEv43iXd3d09qaWkZ2rt3b+3x48fPuccmkUicfvLJJ//9U5/61Ce++MUv9p971mhFdZNxnffzO2k3C28xxiyW1GaMWS8prx7IWrtF0hZJunphk21ITCtUXVHmksmkpk8/f+T5vn0vatWq9qx/aSWTSRUiPqYOl/yeuTEFr01O5Rvqi1ij8pHt2mSLn0Jcm+GJCaf3RyFMzEybVv6/VyHkcm3S4+eGG7551rU5MTSg8ePP7mbHjx+vW2+9VT/60Y+Kntx8+tOf9m8yzltzc/Pw9ddfP9TR0THdT3Aefvjh6Y8//vhre/furfXLdXR01P/4xz9uWLly5cDcuXPfb25uHv7hD38YnIkpiaimqAYDP33+/TbX6MOpqbpAGf/58wWuEyoA0wpwQfzARdj4efTRRyNJblytW7fuzZ6enknd3d2Tsq2KWr169cDvfve738+dO/f99GP+kvFSimoEx09eGrMc79eH99o0BMo0BMoCkuic4Ib4gQuX+BlraXfq/KVNbqTUDcczZ8587/777/9oS0vL0Je//OWjBTlxRCIZwfGWe/dJCu590+i9ftCbuhqUtDhQZrFS++YMFrmaiBE6J7ggfuCi2PFT6uTm2LFjI4Mft91229Gnn366fu/evbXNzc3DziePUJSrqNYodb9N+ijOcknr057f4ZWpk0Y2BmxVlqXlqF50TgiL5AYuChE/J06cyPp48skn9Zd/+Zf6p3/6Jy1cuHDUsrmWD+Ott96qkaS77rorKUktLS0jqywy3WBcjiLbydhau8e7qXizMca/0bjTu1nYL7PDGCNJ7caYXqVWVi1nF2MEsSMtwiC5gYtCxM8FF1wgSTp9JvPGwov+43/UK6+88htJOa1+uv766/XKK68oW/nzzz9fU6dOzXkX466urim7du2qP3To0MRHH330tUQicXrFihVJ/0bjjo6O+pdeemmy/+/0vXDKjbHWlroOzq5e2GR/+mx3qauBiNRPnCxrfzJmOb8xSiaHNPBexm0b8jJ1eND5HMU2rn5GTtfGZ8xNOjNwpIg1Kh/j6mcokajNuXMqxLUZnnjONiJl54JJE/KOmbeH3x+7YAVIvza5JDfBa3Po1Zc1Z86cvD+3Zty4A+FqXJ1eeOGFxIIFCy4LHi/pt4kDxcKOtMiEkT+EwchfPJHgoOLQGCEbvosK+aI9iS8SnDKWz5dK8oWSKTRGcMHIH4JoT+IrspuMkT//SyVzYcxNRa5N+SO5gQviB5kwrRlfjOCgItA5wQXxg2xcpjWNpEpYyFPOzpw5YyRlXCVGgoPYo3OCC+IHLkab1hw//iM6efJkCWpVPd59993zjDFvZnqNBAexRucEF8QPXIwVP3X103Tk8BG9++67jOQU2JkzZ8w777xz/qFDhyacOnXq7zOV4R4cxBadE1wQP3CRS/xMvmCKJOkPR/6gU6c+UK4pzqkPPqiOr2t3c8YY8+apU6f+vqmp6ZlMBUhwEEt0TnBB/MBFPvEz+YIpI4lOriaf/5HLHKoHD1NUiCU6J4RFcgMXxE98kOAglli6iTDonOCC+IkXEhzEEjvSIgw6J4RFchM/JDioSOxIi0wY+UMYJDfxRIKDikNjhGwY+UO+aE/iiwQHFYXGCC4Y+UMQ7Ul8keCgYpDcwAXxg0yY1owvEhxUBDonuCB+kA3TmvFFgoPYo3OCC+IHLpjWLF8kOIg1Oie4IH7ggvgpbyQ4iC0aF7ggfuCC+Cl/JDiIJRoXuCB+4IL4iQcSHMQSjQvConOCC+InPkhwEEss3UQYdE5wQfzECwkOYomlmwiDzglhkdzEDwkOKhJLN5EJI38Ig+QmnkhwUHFojJANI3/IF+1JfJHgoKLQGMEFI38Ioj2JLxIcVAySG7ggfpAJ05rxRYKDikDnBBfED7JhWjO+SHAQe3ROcEH8wAXTmuWLBAexRucEF8QPXBA/5Y0EB7FF4wIXxA9cED/ljwQHsUTjAhfED1wQP/FQ8gTHGNNY6jogfmhcEBadE1wQP/ERaYJjjGk0xtj0h6RlgTLLjDGdxph2Y8xmY0xTlHVEPLB0E2HQOcEF8RMv4yP+vPXeY9A/YK3d4v/bGLNM0lZJs6y1g97ozgFjTIu19mDEdUUZY+kmwqBzQlgkN/ETWYJjjKmT1GCtXTNKsXZJ2621g5Jkre0zxuzxji8ueiVRMdIboxtu+Gapq4MywcgfwiC5iacop6i+IWmZMabXm3qqS3/Rm4pqlLQ78L7dklqD5YFsaIyQDSN/yBftSXxFmeD0StohqUFSm6QBb0rK599s3Bd4X7/385riVg+VgMYILti0DUG0J/EV2RSVd6/NFumse206jTGzrbV9kq71ig4G3uo/P2u1lTGmTalESRfPmKH+5LHiVLyEEomEjh59N+eylXgNMsl2Xfbvf1l3371V27at17x5l4+UKdS1ef/kkPM5ii2fmBkp3z9QxBqVj7GuTTB+CnFtTk4wTu+PQpiYOXYsWcQalY9EIqFt2+4+qz3JZv/+lwt2bSbPvMj5HIj+JmNJkrV2hzGmT9IBpVZRfVdSXj1QesJ09cIm25CYVvB6lloymdT06efnXLYSr0Emma7Lvn0vatWq9ox/aRXq2kwdrnE+R7HlEzMj5Rvqi1ij8jHatckUP4W4NsMTE07vj0KYmJk2rfx/r0JIJpP63Of+bMxyfvwkk0NVc23ioGT74Hirog5Kmu0d8qem6gJF/efPF79WiCOmpeCC+IELpjXLV6k3+vNHcfx/S6l7dNI1BF4HRtA5wQXxAxfET3krdYLTJGmPNDKiM6hzl4MvlnTQXzoO+Ghc4IL4gQvip/xFkuAYY5q83Ylb046tk9Tu3WDsu0NSm78k3PvZ6h0HRtC4wAXxAxfETzxEdZPxoLw9bowxO5SabnoiuDuxd/OxJLUbY3qVWlm1nF2MEUTjgrDonOCC+ImPSBIcb5RmYY5ldyi1Xw6QFTvSIgw6J7ggfuKl1PfgAKGwIy3CoHNCWCQ38UOCg4rE0k1kwsgfwiC5iScSHFQcGiNkw8gf8kV7El8kOKgoNEZwwcgfgmhP4osEBxWD5AYuiB9kwrRmfJHgoCLQOcEF8YNsmNaMLxIcxB6dE1wQP3DBtGb5IsFBrNE5wQXxAxfET3kjwUFs0bjABfEDF8RP+SPBQSzRuMAF8QMXxE88kOAglmhcEBadE1wQP/FBgoNYYukmwqBzggviJ15IcBBLLN1EGHROCIvkJn5IcFCRWLqJTBj5QxgkN/FEgoOKQ2OEbBj5Q75oT+KLBAcVhcYILhj5QxDtSXyR4KBikNzABfGDTJjWjC8SHFQEOie4IH6QDdOa8UWCg9ijc4IL4gcumNYsXyQ4iDU6J7ggfuCC+ClvJDiILRoXuCB+4IL4KX8kOIglGhe4IH7ggviJBxIcxBKNC8Kic4IL4ic+SHAQSyzdRBh0TnBB/MQLCQ5iiaWbCIPOCWGR3MQPCQ4qEks3kQkjfwiD5CaeSHBQcWiMkA0jf8gX7Ul8keCgotAYwQUjfwiiPYkvEhxUDJIbuCB+kAnTmvFFgoOKQOcEF8QPsmFaM75IcBB7dE5wQfzABdOa5YsEB7FG5wQXxA9cED/ljQQHsUXjAhfED1wQP+WPBAexROMCF8QPXBA/8UCCg1iicUFYdE5wQfzEx/h832CMaZE0S9Lz1trfhv1gY0y7pDpr7ZrA8WWSbpbUJ6lO0mZr7cGwn4PKxNJNhEHnBBfET7zkleAYY/6TpGPW2r3GmFnGmP8iabu1dijP87RKWidpS+D4MklbJc2y1g4aYxolHTDGtJDkIB1LNxEGnRPCIrmJn1GnqIwxVwUO1UtqkCRr7avW2v8lqS2fDzTG1ElaLilTwtKuVMI06H1Gn6Q93nEgZyzdRCaM/CEMkpt4GusenO8aYx4xxtR6z/dI2mKMuccYM8U7Ni3Pz2yXtD540BjTJKlR0u7AS7sltXqJETAmGiNkw8gf8kV7El9jJThrlEpq/pcxZq219rik/yzpTkmDxphjSiUlOTHGtEnq9EdoAvzz9AWO93s/r8n1c1C9aIzggpE/BNGexNdY9+D0S7LW2i8aY1qMMdsl/b/W2suNMbOUevHVXD7Iu59mtrV2S5Yi13o/BwPH/ednJVJestQmSRfPmKH+5LFcqhEriURCR4++m3PZSrwGmWS7Lvv3v6y7796qbdvWa968y0fKFOravH8yr1vNSiKfmBkp3z9QxBqVj7GuTTB+CnFtTk4wTu+PQpiYOXYsWcQalY9EIqFt2+4+qz3JZv/+lwt2bSbPvMj5HBg7wWmV1GiM+YKkQS/RucMYs0bSemvtoTw+q91au3yU1/PqgbxEaYskXb2wyTYk8p0pK3/JZFLTp5+fc9lKvAaZZLou+/a9qFWr2jP+pVWoazN1uMb5HMWWT8yMlG+oL2KNysdo1yZT/BTi2gxPTDi9PwphYmbatPL/vQohmUzqc5/7szHL+fGTTA5VzbWJg7GmqK6x1v4Pa+1OSa8aY/6LtXarUiMn/9UY8520+3Oy8paEP2GMafQfSi0Bb/D+LX04NVUXeLv//Pmxfx1UI6al4IL4gQumNcvXWAnObP9mYm8q6oAx5j9Za49ba/9WqRGUrTl8TqukTkm9aY9GScsk9XrTTX6C0xB4r/88eG8OQOcEJ8QPXBA/5W2sBGeLpEPeqqnLJFlJC/0XvaXiN4/1IdbahdZak/5Qapn4Fu/5Fm+fm0FJiwNvXyzpYJYbk1HFaFzggviBC+Kn/I2a4Fhr9yi1q/CdSo2gdEraXMT63CGpzV8S7v1s9Y4DI2hc4IL4gQviJx7G3MnYS3Iuj6AustbuMMZIUrsxpleplVXL2cUYQTQuCIvOCS6In/jI+7uoCsVauzDL8R2SdkRcHcQMO9IiDDonuCB+4oVvE0cssSMtwqBzQlgkN/FDgoOKxNJNZMLIH8IguYknEhxUHBojZMPIH/JFexJfJDioKDRGcMHIH4JoT+KLBAcVg+QGLogfZMK0ZnyR4KAi0DnBBfGDbJjWjC8SHMQenRNcED9wwbRm+SLBQazROcEF8QMXxE95I8FBbNG4wAXxAxfET/kjwUEs0bjABfEDF8RPPJDgIJZoXBAWnRNcED/xQYKDWGLpJsKgc4IL4ideSHAQSyzdRBh0TgiL5CZ+SHBQkVi6iUwY+UMYJDfxRIKDikNjhGwY+UO+aE/iiwQHFYXGCC4Y+UMQ7Ul8keCgYpDcwAXxg0yY1owvEhxUBDonuCB+kA3TmvFFgoPYo3OCC+IHLpjWLF8kOIg1Oie4IH7ggvgpbyQ4iC0aF7ggfuCC+Cl/JDiIJRoXuCB+4IL4iQcSHMQSjQvConOCC+InPkhwEEss3UQYdE5wQfzECwkOYomlmwiDzglhkdzEDwkOKhJLN5EJI38Ig+QmnkhwUHFojJANI3/IF+1JfJHgoKLQGMEFI38Ioj2JLxIcVAySG7ggfpAJ05rxRYKDikDnBBfED7JhWjO+SHAQe3ROcEH8wAXTmuWLBAexRucEF8QPXBA/5Y0EB7FF4wIXxA9cED/ljwQHsUTjAhfED1wQP/FAgoNYonFBWHROcEH8xEekCY4xps0Y02uMscaY3caYxgxllhljOo0x7caYzcaYpijriHhg6SbCoHOCC+InXsZH9UHGmHXePxdLapTUKWm3pNlpZZZJ2ipplrV20EuADhhjWqy1B6OqK8ofSzcRBp0TwiK5iZ8oR3D6rLXftdb2WWv3SFqvVKKTrl3SdmvtoCRZa/sk7fGOAzlj6SYyYeQPYZDcxFNkCY61dkfgUJ1SyYskyZuKalRqVCfdbkmtxpi6YtYPlYPGCNkw8od80Z7EV0luMvaSlWslLU877I/m9AWK93s/rylytVABaIzggpE/BNGexFdk9+BII4nNNyS1KTWC0y9pjffytd7PwcDb/OdnTWcZY9q88+jiGTPUnzxW6OqWXCKR0NGj7+ZcthKvQSbZrsv+/S/r7ru3atu29Zo37/KRMoW6Nu+fHHI+R7HlEzMj5fsHilij8jHWtQnGTyGuzckJxun9UQgTM8eOJYtYo/KRSCS0bdvdZ7Un2ezf/3LBrs3kmRc5nwMRJzjevTXrjTFPKHWTcZsxZrc3fZVXD2St3SJpiyRdvbDJNiSmFbq6JZdMJjV9+vk5l63Ea5BJpuuyb9+LWrWqPeNfWoW6NlOHa5zPUWz5xMxI+Yb6ItaofIx2bTLFTyGuzfDEhNP7oxAmZqZNK//fqxCSyaQ+97k/G7OcHz/J5FDVXJs4KMkUlbX2oLXWXz212PvpT03VBYr7z58vcrUQU0xLwQXxAxdMa5avUm/0tyft336C0xAo0xB4HRhB5wQXxA9cED/lrdQJjuStmvL2uRnUhyM6vsWSDvpLxwEfjQtcED9wQfyUv0gSHGNMnbczcVPasVal9sZJXz5+h1L35dT575PU6h0HRtC4wAXxAxfETzxEeZNxk6R1xpgdkp5TalRmTXoBa+0OY4wktRtjeuUtJWcXYwTRuCAsOie4IH7iI5IEx5teCk49ZSu7Q1JwU0DgLOxIizDonOCC+ImXcrgHB8gbO9IiDDonhEVyEz8kOKhILN1EJoz8IQySm3giwUHFoTFCNoz8IV+0J/FFgoOKQmMEF4z8IYj2JL5IcFAxSG7ggvhBJkxrxhcJDioCnRNcED/IhmnN+CLBQezROcEF8QMXTGuWLxIcxBqdE1wQP3BB/JQ3EhzEFo0LXBA/cEH8lD8SHMQSjQtcED9wQfzEAwkOYonGBWHROcEF8RMfJDiIJZZuIgw6J7ggfuKFBAexxNJNhEHnhLBIbuKHBAcViaWbyISRP4RBchNPJDioODRGyIaRP+SL9iS+SHBQUWiM4IKRPwTRnsQXCQ4qBskNXBA/yIRpzfgiwUFFoHOCC+IH2TCtGV8kOIg9Oie4IH7ggmnN8kWCg1ijc4IL4gcuiJ/yRoKD2KJxgQviBy6In/JHgoNYonGBC+IHLoifeCDBQSzRuCAsOie4IH7igwQHscTSTYRB5wQXxE+8kOAglli6iTDonBAWyU38kOCgIrF0E5kw8ocwSG7iiQQHFYfGCNkw8od80Z7EFwkOKgqNEVww8ocg2pP4IsFBxSC5gQviB5kwrRlfJDioCHROcEH8IBumNeOLBAexR+cEF8QPXDCtWb5IcBBrdE5wQfzABfFT3khwEFs0LnBB/MAF8VP+SHAQSzQucEH8wAXxEw+RJjjGmHXGmF5jjDXG7DbGNGYos8wY02mMaTfGbDbGNEVZR8QDjQvConOCC+InPsZH9UHGmM3eP9dLulbSOkkHjDGzrLWDXpllkrZKmmWtHfQSoAPGmBZr7cGo6oryx9JNhEHnBBfET7xEMoJjjKmTNGitXWOt3WGtXS9pjaQ6SV9MK9ouabuf8Fhr+yTt8Y4DI1i6iTDonBAWyU38RDVF1eglNSOstVu8f9ZJkjcV1Shpd+C9uyW1ekkSkBOWbiITRv4QBslNPEWS4GSaXkpLWPzX/Ptx+gJF+72f1xS+ZqhENEbIhpE/5Iv2JL4iuwcng1ZJfdbaPd7za72fg4Fy/vOzbkg2xrRJapOki2fMUH/yWHFqWUKJREJHj76bc9lKvAaZjHZd9u9/WXffvVXbtq3XvHmX6+jRdwt2bd4/OeR8jmLLJ2ZGyvcPFLFG5SOXa5MeP6tW/U/na3NygnF6fxTCxMyxY8ki1qh8JBIJtbVtOas9Gat8Ia7N5JkXOZ8DpU1wviFpedrzvHogb4priyRdvbDJNiSmFbBq5SGZTGr69PNzLluJ1yCTbNdl374XtWpV+zl/aRXq2kwdrnE+R7HlEzMj5Rvqi1ij8jHWtQnGTyGuzfDEhNP7oxAmZqZNK//fqxCSyaQ6O7+W88hfNV2bOCjJPjjGmHZJ6wNTV/7UVF2guP/8+SJXCzHGMDJcED/IhmnN+Io8wfGWgj+XNjXl8xOchsDxhsDrwFnonOCC+IELFjSUr0inqIwxrZJkrd0RON5orT1ojBmUtFippeG+xZIO+kvHEQ/zLp+jP7zxRk5lL77kEv3ulf8T6nPonOCC+IEL4qe8RbnRX6tSm/xt9kZxfDdLusP79x2SthpjvuNt9Fen1M3ILVHVE4Xxhzfe0J0//H/0z/ds0Je+d78uvzb7htT3fPL6UJ9B4wIXxA9cED/lL5IEx9vjxt/fpjXw8o60jf12GGMkqd0Y06vUyqrl5bqLcT6jFJLbSEUc5ZLchEXjAhfED1wQP/EQSYLjJSg5rZf0pq92jFmwDPzhjTf0vRefzbl82JGKuCpWciOxIy3Co3OCC+InPvg2cRRNLsnNK8+FG5xjR1qEQecEF8RPvJRyHxxUuVeeO6h/vmdDqPfGdenmxz55nQ4fPpJz+ZkzZ+j1F/cXsUbVhc4JYZHcxA8JDkrCT26+9L379cjtf13w86c3Rjfc8M2Cnz+sw4ePMK2ZxYUfv0pH3zqac/np/2G6/vRvv83rMxj5QxgkN/FEgoPIpSc33IAM39G3juaU/Pnxk08y5IvryB9Kh/YkvrgHB5EiuYGL9PgpBjZtQxDtSXyR4CAyJDdwQfygFJjWjC8SHESCzgkuiB+UCtOa8UWCg6Kjc4IL4gfljGnN8sVNxigqOie4IH5SPv7x2TqSx67pMy65RP/2b71FrFH5yOfaFPq6xCV+qhUJDoqGzgkuiJ8PHWHX9KyyXZtM8VPI6xKn+KlWTFGhaOicEBbJDVwQP5BIcFBEfBcVwqBzggviBz4SHBQN30WFfNE5wQXxg3QkOCiZavwuKoyumJ2TxMhfJSO5QRAJDkqCHWmRSa6dEyN/SEdyg0xIcBA5GiNkk2tyw8gf0tGeIBMSHESK5AYuGPlDJkxrIhMSHESG5AYuiB9kw7QmMmGjvwhNrKlR/cTJpa5GSdA5wQXxAxdMa1YnEpwIvXf6tD749tdyLv+Rb3+/iLWJDp0TXBA/cJEeP4/c/tcFP396/NxwwzcLfn6ExxQViorOCS6IH7ggfqobCQ6KhsYFLogfuCB+QIKDoqFxQVh0TnBB/EAiwUERsXQTYdA5wQXxAx8JDoqGpZvIF50TXBA/SEeCg5Jh6SaC+C4qhEVygyASHJREemPk7w+U6yMX7EgbT3wXFcIguUEm7IODyAUbo7H2B/rZq2/orzp36X8vv1F/PuuSMfcHojGKL76LCmGQ3CATRnAQqXz/0gomN2OhMapsfBcVMmFaE5mQ4CAyJDdwwTQEsmFaE5mQ4CASJDdwQXIDF0xrVicSHBQdyQ1ckNzABdOa1YsEB0VFcgMXJDdwQfxUNxIcFA3JDVzQOcEF8QMSHBQNyQ3ConOCC+IHUsQJjjGmzhjTaYxZluX1Zd7r7caYzcaY4qz7QyTyaVzySW4klm5WMjonuCB+4Itsoz8vqVkjqVXSE1le3ypplrV20BjTKOmAMabFWhtufR9KKp+lm7kmNz979Q1JLN2sVHROcEH8IF1kIzjW2h2S1o9SpF3SdmvtoFe+T9Ie7zgqUPrSzVyTm7/q3CWJpZuViu+iQlgkNwgqi3twvKmoRkm7Ay/tltRqjKmLvFIoqnyXbqbfo5MLlm7GE99FhTBIbpBJWSQ4SiU3ktQXON7v/bwmwrqgyFhdhWz4LiqEQXKDTMrlyzav9X4OBo77zxsDx2WMaZPUJkkXz5ih/uSxYtUtq0QiodNDb+dVPplHTplIJHT06Ls5ly3FNcgm27V5/aUe7Wp/QKv/8e/1sY9fqdNDb496XZ47/Efds/dZ/XDlFzRv5kVKavTrsn//y7r77q3atm295s27XEePvluwa/P+ySHnc4SJmaP9A/mVzzFmwpy/mHK5Nunx8783/EPBr016/Kxa9T+dr83JCcbp/b4wcXPsWDLnsvnGTK7njkIikdBf3f93I+3JWGXzqXsikVBb25az2pNCnj+byTMvcj4HyifBybv3sdZukbRFkq5e2GQbEtMKXqmxJJNJ1dRekFf5hM7kVX769PNzLluKa5BNpmvzynMH9c9/+9/0pe/dr1lpf2lluy4/e/UN3e6P3My8UPLKZLsu+/a9qFWr2s/5S6tQ12bqcI3zOcLEzPSG+vzK5xgzYc5fTGNdm2D8FPraBOOnENdmeGLC6f2+MHEzbVpunx0mZnI9dxSSyaRmLbpuzHKvPHcw77onk0l1dn4t55G/crs21a5cEhx/aqoucNx//nxkNUFRMC0FF9xjARd+/EysqdEFkybk9V6mNeOr3BKchsDxhsDriCGSG7gguYGL9Ph55Pa/1gff/tqo5dPbn9Z/6hzz/Onxc8MN3yxUtVEAZXGTsbfPzaCkxYGXFks66C8dR/yQ3MAFyQ1c0P5Ut7JIcDx3SGrzl4R7P1u944ghGhe4ILmBC9ofRLmTcas+3OhvjTFm0Fq7x3/dWrvDGCNJ7caYXqVWVi1nF+P4onFBWCQ3cEFyAynCBMdLZvaMUWaHpB3R1AjFxndRuZtYU6Nx9TNKXY1I5do5hb021dA5hbmZtlKQ3MBXLjcZowLxXVTu3jt9esybItN95NvfL2Jtii+fzum906e157blOXdOH/n296umc8onbuIeM+lIbpCunO7BQZXhu6gQlO+0FCN/8JHcIIgEByXBd1Ehk3y/i4qRP0gkN8iMBAeRozFCNvl+FxUjf5BY0IDMSHAQKZIbuGDkD5kwrYlMSHAQGZIbuCB+kA3TmsiEBAeRoHOCC+IHLpjWrE4kOCg6Oie4IH7ggmnN6kWCg6Kic4IL4gcuiJ/qRoKDoqFxgQviBy6IH5DgoGhoXBAWnRNcED+QSHBQRCzdRBh0TnBB/MBHgoOiYekm8kXnBBfED9KR4KBkWLqJIL6LCmGR3CCIBAclwdJNZMJ3USEMkhtkQoKDyNEYIRu+iwph0J4gExIcRIrkBi4Y+UMmTGsiExIcRIbkBi6IH2TDtCYyIcFBJOic4IL4gQumNasTCQ6Kjs4JLogfuGBas3qR4KCo6JzggviBC+KnupHgoGhoXOCC+IEL4gckOCgaGheERecEF8QPJBIcFBFLNxEGnRNcED/wkeCgaFi6iXzROcEF8YN0JDgoGZZuIojvokJYJDcIIsFBSbB0E5nwXVQIg+QGmZDgIHI0RsiG76JCGLQnyIQEB5EiuYELRv6QCdOayIQEB5EhuYEL4gfZMK2JTEhwEAk6J7ggfuCCac3qRIKDoqNzggviBy6Y1qxeJDgoKjonuCB+4IL4qW4kOCgaGhe4IH7ggvgBCQ6KhsYFYdE5wQXxA0kaX+oKBBljlkm6WVKfpDpJm621B0taKYTC0k2EQecEF8QPfGWV4HjJzVZJs6y1g8aYRkkHjDEtJDnxw9JN5IvOCS6IH6QrtymqdknbrbWDkmSt7ZO0xzuOCsPSTQTxXVQIi+QGQWWT4BhjmiQ1StodeGm3pFZjTF3klULRsHQTmfBdVAiD5AaZlE2Co1RyI6XuvUnX7/28JsK6oIhojJAN30WFMGhPkImx1pa6DpIkY0y7pHWSZntTU/7xVqVGcdZYa7ekHW+T1OY9/bikf4uwupUsISlZ6kogVogZhEHcZJe01n621JWIu3K6yfhYPoW9ZGfLmAWRF2PM89ZaRsuQM2IGYRA3KLZymqLyR23qAsf9589HVhMAABBr5ZjgNASONwReBwAAGFXZJDjePjeDkhYHXlos6aC/dBxFx7Qf8kXMIAziBkVVNjcZSxk3+quT9KokNvoDAAA5K6sERxpJchZL6pV0rVJf1bCntLUCAABxUjZTVD5r7Q5r7Rpr7XettctJboDMjDF1xpjgxpjBMsuMMQeMMdZ7rBtt00xjTKsxpjOtfKcxZrd3jt3etg2jfcaAV85/327/PGHLww1xgqplreVRwIekZZIOSLLeY52kulHKNym1z49ffrek1sD5er3XNktq9I63pr9vjDq1pZ2/Pe0c6XUd8M7X6f30z92Zdp7WwO/mlzvgva8p8LmbvXK9XpneLM/Xlfq/WxwfXmzZ4HXPUK7Vj588zm0l9QaOtXvH2/L5DKU28ex0Kc+DOCFOeOT7KHkFKvERsqHoDCYUaa8dSE960o7XpSUbWZOEtKSkN5+6ZmlwmjIkPnVpn9EU+J2WpT33G9rgsfZS/zeL4yPtmo/ayKf9N3PquMY4PupnBOM33/I8iBPihEe+j7KboqoQ/WMXOZu1drlSS+GXed/LJUkyxmxWlvuQbGplmb98fk2m83rnGm0Poax1takdpTcHDg8G3+fV4zsZ6tFvrd0xymfLWvvd0V5HZt69ak9I2qFUzNRF9NGD+b7BGNOUKX4LVR7ZESeoZiQ45WW597NTGmmcZNO+oiKDPqWWWzb65QO+oZDfxp5nAzLo/RzZx8hamzHpCrLWrs+zakglklv0YWL5jWJ/oBdfdUp1lvnIKQ68z2iUdHOe50d2xAmqVjl9VUPVs9YeNMZ8V9I67+a5Rmvtwhze2q7UfTbfUFqj4v21Vmet7TPGhKnSGuXe6PjlngjzQcidNyrX542cHTTG9Cn137/QiWKDN4LYoNR0ZaMC3wmXwRe9zsd3jaTRkuRg+VZJjOoVAHGCascITpnxRjP6lLoB+DtjFPff06dU49CUPr2l/EZvvpi2OmG3MWZA5+4qna7RW+WwzhhzQKkGZ/1YU1IoiDU6+79ru6Q67wtoC6lfqc5wvVKx2Cdp8xifs91au9h/SGoZ4zOC5bNNtdYZY3rTnrcaY3q9lTR1Gcq3pa3uaQy+XiWqKk68/+YHsp2AGKo+JDjladD7mc/Ukl82fQi6NY8ppnwbnAal9ilqV2qUqN71fhqvwdntLQndHDhOQ6SRUbk2pTqQ3d7yX39qs+BTfdbaQWttn01t37BQH3ZeTWO913v/QaVW2uX6eVskPZfhpXadPf25Rx/+xZ+pI/WvyWbvD4CqUqVx8rxSNyVnQwxVGRKcMuN17E8oNfzamN7Rj8b7n9W/SbnR++spp/dmOd9YDc7z3miTX89Q9/n4vMSkzkuuFkpq8xtHGqKztElanp6MetfMvw/rnP1HCsz/75Dz54wxVZGp/FmjgN49HZn2OxlU6vf+RqC8v4VCNau6ONEoNzYTQ9WJBKeMeElJg01tcuhPVbXl0Rj5ScZ6jT0HPqZc3u/V86BS9w1lusk518/q8xssL0nxH75B0RBJqf+umaYB/WQ277/OvYR4XY7F/b+AB/P8jKbAZ4w2/XlW3bx/Zktc/WmX9NhbLr7nqKriZIxzEkNVigSnBDI1FN5oxRpvubhvZFVVlqmZs455CcmgUn+9FWT5ZIYG55x6KDWdNejV03mKyPu8dnvuF6xWdUPkJcAZv5PNG3Hrk9SaYVogayfh/ffarRzixUsmlyl142rwmo/2GXVK/fUcJiazddSSRpLhHfISX+/36c1WvhpUaZyMhhiqVqXagKeSHxp787xenb0hXp1SOwmfs9OoPtwNeHe2cwWO+Zvp1QWOh9nory5DXc/Z6C9wnt7gZ6eVybrDaVqZ9N2V0z+33fvZKelA2u+/Th9ueFixm35513fAu77LMrze5r3u/zdoTXtfcPdpf7dqfyfpgcDndAbK+7tV93rxGIyt0T5j5HiW8gPKstu39zvVpf23Hgi87seEH5NNXozVVUNMECfn/G6N6e8nhniUvAKV9gjRULSlHfdv2E0/V/BrHJoyfE761y/UKS1Z8f6HTW+I0suGaXDS63NWg6MPE5iRhjPwe6Y3rFmTHK98p87eLZmGqMoeaf/P9AZix/9/oD2t7AGvvB8nxESVPbIkOMRQFT/K7tvEAenDb5W33maBxph2620I6C0F7Zd00Fq73hvaHvDKs7NpBfL+G79qra1PO7Y5LT78m0hn29S+T3UiJqqKN+V2wFqbcdMvYqj6cA8OykbgnqJrdfYqsPTXvqPUaFLoVWKIN+8+jy8aY9qNMXU2dY/FlrSOyb8ZfU2m+9dQWbz/xjd7/85psQMxVPkYwUFZ8P766lTqZr/nlLpB8aD3mj/fv0XSd6y1g/5fXmkN0TrvvXfYc29OBgBUGRIcAABQcZiiAgAAFYcEBwAAVBwSHAAAUHFIcAAAQMUhwQEAABWHBAcAAFSc/x/waKMMgymHVgAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#IR\n",
    "lxmert_original=[20.0652528548124, 11.419249592169658, 62.31647634584013]\n",
    "lxmert_original_ckpt=[19.086460032626427, 11.582381729200652, 62.31647634584013]\n",
    "lxmert_cc_ctrl=[18.92332789559543, 14.681892332789559, 57.91190864600326]\n",
    "lxmert_cc_ctrl_ckpt=[19.249592169657422, 13.050570962479608, 60.52202283849919]\n",
    "albef_4m=[18.270799347471453, 8.64600326264274, 68.35236541598695]\n",
    "albef_4m_coco_neutral=[17.94453507340946, 9.135399673735726, 67.69983686786297]\n",
    "albef_4m_cc3m_neutral=[17.94453507340946, 8.64600326264274, 69.16802610114192]\n",
    "albef_14m=[16.802610114192497, 8.482871125611746, 69.9836867862969]\n",
    "albef_14m_coco_neutral=[16.6394779771615, 8.31973898858075, 70.47308319738988]\n",
    "albef_14m_cc3m_neutral=[16.802610114192497, 9.624796084828711, 68.67862969004894]\n",
    "blip=[15.660685154975528, 9.624796084828711, 69.9836867862969]\n",
    "blip_coco_neutral=[15.98694942903752, 8.64600326264274, 69.65742251223492]\n",
    "\n",
    "data_ir = [\n",
    "         lxmert_cc_ctrl, lxmert_cc_ctrl_ckpt,\n",
    "         albef_4m, albef_4m_cc3m_neutral,\n",
    "         albef_14m, albef_14m_cc3m_neutral]\n",
    "index_ = [\n",
    "          \"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"concapalbef14m_n\"]\n",
    "\n",
    "dfir = pd.DataFrame(data=data_ir, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "dfir[\"model\"] = dfir.index\n",
    "\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 5.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 3.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(3.5+bar_width, 5.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{3M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(dfir)))\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    #ir\n",
    "    ys = [x for x in dfir[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(dfir)/2), \n",
    "            edgecolor='black')    \n",
    "\n",
    "ax0.set_ylabel('\\%', fontsize=15)\n",
    "ax0.set_ylim(0, 80)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=18)\n",
    "ax0.grid(color='gray', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$')\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='upper right', bbox_to_anchor=(1.19,1.0))\n",
    "plt.savefig(f'figures/f30kIR_not_dba_hist_cc3m.pdf', bbox_inches='tight')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "9ed9739d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#TR\n",
    "lxmert_original=[18.75, 8.630952380952381, 25.595238095238095]\n",
    "lxmert_original_ckpt=[19.345238095238095, 7.142857142857142, 24.107142857142858]\n",
    "lxmert_cc_ctrl=[24.702380952380953, 8.630952380952381, 21.726190476190478]\n",
    "lxmert_cc_ctrl_ckpt=[20.833333333333336, 9.226190476190476, 22.916666666666664]\n",
    "albef_4m=[15.476190476190476, 6.547619047619048, 24.702380952380953]\n",
    "albef_4m_coco_neutral=[14.880952380952381, 6.547619047619048, 24.404761904761905]\n",
    "albef_4m_cc3m_neutral=[14.880952380952381, 6.845238095238096, 23.809523809523807]\n",
    "albef_14m=[14.583333333333334, 6.547619047619048, 24.107142857142858]\n",
    "albef_14m_coco_neutral=[14.880952380952381, 6.547619047619048, 24.107142857142858]\n",
    "albef_14m_cc3m_neutral=[14.583333333333334, 7.142857142857142, 24.404761904761905]\n",
    "blip=[13.988095238095239, 8.035714285714286, 25.595238095238095]\n",
    "blip_coco_neutral=[16.071428571428573, 7.738095238095238, 25.297619047619047]\n",
    "data_tr = [lxmert_original, lxmert_original_ckpt,\n",
    "         albef_4m, albef_4m_coco_neutral,\n",
    "         albef_14m, albef_14m_coco_neutral,\n",
    "         blip, blip_coco_neutral]\n",
    "index_ = [\"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"concapalbef14m_n\", \n",
    "          \"blip\", \"blip_n\"]\n",
    "dftr = pd.DataFrame(data=data_tr, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "dftr[\"model\"] = dftr.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 7.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 5.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(5.5+bar_width, 7.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width, 6.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{180K}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{BLIP}_{129M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(dftr)))\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    #tr\n",
    "    ys = [x for x in dftr[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(dftr)/2), \n",
    "            edgecolor='black')    \n",
    "\n",
    "ax0.set_ylabel('\\%', fontsize=15)\n",
    "ax0.set_ylim(0, 35)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=18)\n",
    "ax0.grid(color='gray', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$')\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='upper right', bbox_to_anchor=(1.19,1.0))\n",
    "plt.savefig(f'figures/f30kTR_not_dba_hist_coco.pdf', bbox_inches='tight')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e044679e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#TR\n",
    "lxmert_original=[18.75, 8.630952380952381, 25.595238095238095]\n",
    "lxmert_original_ckpt=[19.345238095238095, 7.142857142857142, 24.107142857142858]\n",
    "lxmert_cc_ctrl=[24.702380952380953, 8.630952380952381, 21.726190476190478]\n",
    "lxmert_cc_ctrl_ckpt=[20.833333333333336, 9.226190476190476, 22.916666666666664]\n",
    "albef_4m=[15.476190476190476, 6.547619047619048, 24.702380952380953]\n",
    "albef_4m_coco_neutral=[14.880952380952381, 6.547619047619048, 24.404761904761905]\n",
    "albef_4m_cc3m_neutral=[14.880952380952381, 6.845238095238096, 23.809523809523807]\n",
    "albef_14m=[14.583333333333334, 6.547619047619048, 24.107142857142858]\n",
    "albef_14m_coco_neutral=[14.880952380952381, 6.547619047619048, 24.107142857142858]\n",
    "albef_14m_cc3m_neutral=[14.583333333333334, 7.142857142857142, 24.404761904761905]\n",
    "blip=[13.988095238095239, 8.035714285714286, 25.595238095238095]\n",
    "blip_coco_neutral=[16.071428571428573, 7.738095238095238, 25.297619047619047]\n",
    "data_tr = [lxmert_cc_ctrl, lxmert_cc_ctrl_ckpt,\n",
    "         albef_4m, albef_4m_cc3m_neutral,\n",
    "         albef_14m, albef_14m_cc3m_neutral]\n",
    "index_ = [\"lxmert_3m\", \"lxmert_3m_n\",\n",
    "          \"albef4m\", \"concapalbef4m_n\", \n",
    "          \"albef14m\", \"concapalbef14m_n\"]\n",
    "dftr = pd.DataFrame(data=data_tr, \n",
    "                  index=index_,\n",
    "                 columns=[\"M\",\"F\",\"N\"])\n",
    "dftr[\"model\"] = dftr.index\n",
    "\n",
    "f, ax0 = plt.subplots(figsize=(8,6), tight_layout=True)\n",
    "\n",
    "bar_width = 2./len(models)\n",
    "\n",
    "cm = plt.get_cmap('Set3') \n",
    "colors = cm.colors\n",
    "bcm = plt.get_cmap('Pastel2')\n",
    "bcolors = bcm.colors\n",
    "ax0.set_xlim(-0.5, 5.5+bar_width) \n",
    "ax0.axvspan(-0.5, 1.5+bar_width, facecolor=bcolors[0], alpha=0.2)\n",
    "ax0.axvspan(1.5+bar_width, 3.5+bar_width, facecolor=bcolors[1], alpha=0.3)\n",
    "ax0.axvspan(3.5+bar_width, 5.5+bar_width, facecolor=bcolors[2], alpha=0.2)\n",
    "\n",
    "ax0.set_xticks([0.5+bar_width, 2.5+bar_width, 4.5+bar_width])\n",
    "ax0.set_xticklabels(['$\\\\mathrm{\\\\textsc{LXMERT}_{3M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{4M}}$', \n",
    "                     '$\\\\mathrm{\\\\textsc{ALBEF}_{14M}}$'], fontsize=18)\n",
    "\n",
    "plot_dict = {\n",
    "    'M': colors[0],\n",
    "    'F': colors[3],\n",
    "    'N': colors[1]}\n",
    "row_names = list(plot_dict.keys())\n",
    "\n",
    "X = np.array(range(len(dftr)))\n",
    "\n",
    "for i, name in list(enumerate([y for x in row_names for y in x])):\n",
    "    #tr\n",
    "    ys = [x for x in dftr[name].values]    \n",
    "    label=None\n",
    "    ax0.bar(X + i*bar_width, ys, \n",
    "            width=bar_width, label=label, color=plot_dict[name], \n",
    "            hatch=[None, \"//\"]*int(len(dftr)/2), \n",
    "            edgecolor='black')    \n",
    "\n",
    "ax0.set_ylabel('\\%', fontsize=15)\n",
    "ax0.set_ylim(0, 35)\n",
    "\n",
    "ax0.tick_params(axis=\"y\", labelsize=18)\n",
    "ax0.grid(color='gray', alpha=0.2, axis='y')\n",
    "\n",
    "ax0.spines['right'].set_visible(False)\n",
    "ax0.spines['top'].set_visible(False)\n",
    "\n",
    "h = [Patch(facecolor=colors[0], edgecolor='k', label='male'),\n",
    "     Patch(facecolor=colors[3], edgecolor='k', label='female'),\n",
    "     Patch(facecolor=colors[1], edgecolor='k', label='neutral'),\n",
    "     Patch(facecolor='#d6d6d6', edgecolor='#d6d6d6', label='$\\\\mathrm{M}$'),\n",
    "     Patch(facecolor='white', edgecolor='black', hatch='//',linewidth=0.1, label='$\\\\mathrm{M^{N}}$')\n",
    "    ]\n",
    "ax0.legend(handles=h, ncol=1, scatterpoints=1, fontsize=14, loc='upper right', bbox_to_anchor=(1.19,1.0))\n",
    "plt.savefig(f'figures/f30kTR_not_dba_hist_cc3m.pdf', bbox_inches='tight')\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}