notebooks/agg_results.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Analysis of results\n",
    "The following ipython notebook demonstrates how results from multiple sweeps can be analysed and summarised. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/homebrew/Caskroom/miniconda/base/envs/clinicaldg-new/lib/python3.9/site-packages/tqdm/auto.py:22: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n"
     ]
    }
   ],
   "source": [
    "import plotnine as ggp\n",
    "from scipy.stats import wilcoxon, ttest_rel\n",
    "from utils import *\n",
    "\n",
    "dbs = {\n",
    "    'aumc': 'AUMCdb',\n",
    "    'hirid': 'HiRID',\n",
    "    'eicu': 'eICU',\n",
    "    'miiv': 'MIMIC',\n",
    "    'pooled (n-1)': 'pooled (n-1)',\n",
    "    'all': 'all'\n",
    "}\n",
    "\n",
    "algos = ['ERM', 'CORAL', 'VREx', 'Fishr', 'MLDG', 'GroupDRO']"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, we need to define which task and model architecture should be summarised. These experiments can then be loaded and aggregated using the helper functions defined in utils.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1450it [00:00, 9122.25it/s]\n"
     ]
    }
   ],
   "source": [
    "task = 'icu-mortality'\n",
    "model = 'gru'\n",
    "path = Path(f'/Users/patrick/icuDG-outputs/{task}_{model}')\n",
    "\n",
    "# Load and aggregate the results\n",
    "grps = ['algorithm', 'val_env', 'test_env']\n",
    "res = load_model_performance(path)          # Load all results\n",
    "agg_res = aggregate_results(res, grps)      # Aggregate across trials\n",
    "bst_res = pick_best_result(agg_res, grps)   # Choose the best set\n",
    "summ = summ_mean_ste(bst_res, \"auroc\")      # Summarise best set\n",
    "\n",
    "# NOTE: this run summarised 5 trials of 10 hyperparameter combinations\n",
    "# - 50x ERMMerged\n",
    "# - 200x ERMID, ERM, CORAL, VREx, Fishr, MLDG, GroupDRO\n",
    "# = 1450 models "
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, we tabulate/plot the results of the different ERM runs (ERMMerged, ERMID, ERM)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>test_env</th>\n",
       "      <th>aumc</th>\n",
       "      <th>hirid</th>\n",
       "      <th>eicu</th>\n",
       "      <th>miiv</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>train_env</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>aumc</th>\n",
       "      <td>0.843±0.005</td>\n",
       "      <td>0.740±0.012</td>\n",
       "      <td>0.771±0.003</td>\n",
       "      <td>0.771±0.005</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hirid</th>\n",
       "      <td>0.735±0.007</td>\n",
       "      <td>0.842±0.009</td>\n",
       "      <td>0.735±0.003</td>\n",
       "      <td>0.757±0.004</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>eicu</th>\n",
       "      <td>0.811±0.006</td>\n",
       "      <td>0.777±0.007</td>\n",
       "      <td>0.842±0.001</td>\n",
       "      <td>0.820±0.004</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>miiv</th>\n",
       "      <td>0.816±0.005</td>\n",
       "      <td>0.800±0.006</td>\n",
       "      <td>0.825±0.002</td>\n",
       "      <td>0.861±0.002</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>pooled (n-1)</th>\n",
       "      <td>0.819±0.005</td>\n",
       "      <td>0.791±0.006</td>\n",
       "      <td>0.827±0.004</td>\n",
       "      <td>0.833±0.002</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>all</th>\n",
       "      <td>0.841±0.003</td>\n",
       "      <td>0.853±0.006</td>\n",
       "      <td>0.847±0.002</td>\n",
       "      <td>0.862±0.003</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "test_env             aumc        hirid         eicu         miiv\n",
       "train_env                                                       \n",
       "aumc          0.843±0.005  0.740±0.012  0.771±0.003  0.771±0.005\n",
       "hirid         0.735±0.007  0.842±0.009  0.735±0.003  0.757±0.004\n",
       "eicu          0.811±0.006  0.777±0.007  0.842±0.001  0.820±0.004\n",
       "miiv          0.816±0.005  0.800±0.006  0.825±0.002  0.861±0.002\n",
       "pooled (n-1)  0.819±0.005  0.791±0.006  0.827±0.004  0.833±0.002\n",
       "all           0.841±0.003  0.853±0.006  0.847±0.002  0.862±0.003"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# tabulate ERM performance\n",
    "tbl = summ.melt(id_vars=grps+['hparams_seed'])\n",
    "\n",
    "tbl['train_env'] = tbl['test_env']\n",
    "tbl['test_env'] = tbl['variable'].str.extract('^([^_]*)(?=_)')\n",
    "\n",
    "tbl = tbl[tbl['algorithm'].str.contains(\"ERM\")]\n",
    "tbl = tbl[tbl['variable'].str.contains(\"auroc\")]\n",
    "tbl = tbl[(tbl['algorithm'] != 'ERM') | (tbl.train_env == tbl.test_env)]\n",
    "\n",
    "tbl.loc[tbl['algorithm'] == 'ERM', 'train_env'] = 'pooled (n-1)'\n",
    "\n",
    "erm = tbl.pivot_table(\n",
    "    values='value', \n",
    "    index='train_env', \n",
    "    columns = 'test_env', \n",
    "    aggfunc = lambda x: x\n",
    ")\n",
    "erm = erm.loc[list(dbs.keys()), list(dbs.keys())[:4]]\n",
    "erm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x950.897 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<ggplot: (399106521)>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Plot ERM performance\n",
    "plt = tbl.copy()\n",
    "plt['value_str'] = plt['value'].str.slice(stop=5)\n",
    "plt['value'] = plt['value_str'].astype(float)\n",
    "\n",
    "g = (ggp.ggplot(plt, ggp.aes('test_env', 'train_env'))\n",
    "     + ggp.geom_tile(ggp.aes(fill='value'), colour='white', size=1)\n",
    "     + ggp.geom_text(ggp.aes(label='value_str'), colour='white') \n",
    "     + ggp.geom_hline(yintercept=2.5, size=4, colour='white')\n",
    "     + ggp.coord_equal() \n",
    "     + ggp.scale_x_discrete(\n",
    "        limits=list(dbs.keys())[:4],\n",
    "        labels=list(dbs.values())[:4]\n",
    "       )\n",
    "     + ggp.scale_y_discrete(\n",
    "        limits=list(dbs.keys())[::-1],\n",
    "        labels=list(dbs.values())[::-1]\n",
    "       )\n",
    "     + ggp.scale_fill_cmap(cmap_name='Blues', limits=[0.5, 1.0], expand=(0, 0))\n",
    "     #+ ggp.guides(fill=ggp.guide_colourbar(barheight=50))\n",
    "     + ggp.guides(fill=None)\n",
    "     + ggp.labs(\n",
    "        x='\\nEvaluation dataset',\n",
    "        y='Training dataset',\n",
    "        fill=''\n",
    "       )\n",
    "     + ggp.theme_minimal()\n",
    "     + ggp.theme(\n",
    "        axis_text_x=ggp.element_text(size=12),\n",
    "        axis_text_y=ggp.element_text(size=12, angle=45), \n",
    "        legend_text=ggp.element_text(size=12),\n",
    "        panel_grid=ggp.element_blank()\n",
    "       )\n",
    "    )\n",
    "g"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We then compare the performance of the pooled (n-1) ERM to that of all included DG algorithms."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "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>evaluated_in</th>\n",
       "      <th>aumc</th>\n",
       "      <th>hirid</th>\n",
       "      <th>eicu</th>\n",
       "      <th>miiv</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>algorithm</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>ERM</th>\n",
       "      <td>0.819±0.005</td>\n",
       "      <td>0.791±0.006</td>\n",
       "      <td>0.827±0.004</td>\n",
       "      <td>0.833±0.002</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CORAL</th>\n",
       "      <td>0.816±0.006</td>\n",
       "      <td>0.785±0.011</td>\n",
       "      <td>0.828±0.002</td>\n",
       "      <td>0.835±0.004</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VREx</th>\n",
       "      <td>0.822±0.005</td>\n",
       "      <td>0.791±0.007</td>\n",
       "      <td>0.818±0.002</td>\n",
       "      <td>0.824±0.003</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fishr</th>\n",
       "      <td>0.803±0.009</td>\n",
       "      <td>0.679±0.006</td>\n",
       "      <td>0.803±0.003</td>\n",
       "      <td>0.690±0.029</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MLDG</th>\n",
       "      <td>0.825±0.003</td>\n",
       "      <td>0.801±0.003</td>\n",
       "      <td>0.820±0.003</td>\n",
       "      <td>0.837±0.001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GroupDRO</th>\n",
       "      <td>0.823±0.007</td>\n",
       "      <td>0.801±0.008</td>\n",
       "      <td>0.826±0.002</td>\n",
       "      <td>0.848±0.002</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "evaluated_in         aumc        hirid         eicu         miiv\n",
       "algorithm                                                       \n",
       "ERM           0.819±0.005  0.791±0.006  0.827±0.004  0.833±0.002\n",
       "CORAL         0.816±0.006  0.785±0.011  0.828±0.002  0.835±0.004\n",
       "VREx          0.822±0.005  0.791±0.007  0.818±0.002  0.824±0.003\n",
       "Fishr         0.803±0.009  0.679±0.006  0.803±0.003  0.690±0.029\n",
       "MLDG          0.825±0.003  0.801±0.003  0.820±0.003  0.837±0.001\n",
       "GroupDRO      0.823±0.007  0.801±0.008  0.826±0.002  0.848±0.002"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# tabulate DG performance\n",
    "tbl = summ.melt(id_vars=grps+['hparams_seed'])\n",
    "\n",
    "tbl['evaluated_in'] = tbl['variable'].str.extract('^([^_]*)(?=_)')\n",
    "\n",
    "tbl = tbl[~tbl['algorithm'].isin([\"ERMMerged\", \"ERMID\"])]\n",
    "tbl = tbl[tbl['test_env'] == tbl['evaluated_in']]\n",
    "\n",
    "dg = tbl.pivot_table(\n",
    "    values='value', \n",
    "    index='algorithm', \n",
    "    columns = 'evaluated_in', \n",
    "    aggfunc = lambda x: x\n",
    ")\n",
    "dg = dg.loc[algos, list(dbs.keys())[:4]]\n",
    "dg"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we use a Wilcoxon rank-sum test to test for differences between "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/f4/k7z3btlx0tdgf95znm5mkj_w0000gn/T/ipykernel_29848/1401068676.py:25: FutureWarning: The behavior of indexing on a MultiIndex with a nested sequence of labels is deprecated and will change in a future version. `series.loc[label, sequence]` will raise if any members of 'sequence' or not present in the index's second level. To retain the old behavior, use `series.index.isin(sequence, level=1)`\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "algorithm  test_env\n",
       "CORAL      aumc        1.0000\n",
       "           eicu        1.0000\n",
       "           hirid       1.0000\n",
       "           miiv        0.8125\n",
       "VREx       aumc        0.6250\n",
       "           eicu        0.3125\n",
       "           hirid       1.0000\n",
       "           miiv        0.0625\n",
       "Fishr      aumc        0.3125\n",
       "           eicu        0.0625\n",
       "           hirid       0.0625\n",
       "           miiv        0.0625\n",
       "MLDG       aumc        0.1875\n",
       "           eicu        0.3125\n",
       "           hirid       0.4375\n",
       "           miiv        0.1250\n",
       "GroupDRO   aumc        0.8125\n",
       "           eicu        1.0000\n",
       "           hirid       0.4375\n",
       "           miiv        0.0625\n",
       "dtype: float64"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# test for differences to ERM\n",
    "# NOTE: because of the small size of resamples, the smallest possible p-value is 0.0625\n",
    "\n",
    "exp = ['algorithm', 'test_env', 'hparams_seed']\n",
    "setting = tbl[exp]\n",
    "runs = setting.merge(res, on=exp)\n",
    "\n",
    "drop_cols = [c for c in runs.columns if \"es_\" in c or \"nll\" in c]\n",
    "perf = runs.drop(columns=drop_cols+['folder']).melt(id_vars=grps+['hparams_seed', 'trial'])\n",
    "perf = perf[perf['variable'].str.extract(\"^([a-z]+)\").values.squeeze() == perf['test_env'].values]\n",
    "\n",
    "paired = perf[perf.algorithm == \"ERM\"].\\\n",
    "    drop(columns=['algorithm', 'val_env', 'hparams_seed', 'variable']).\\\n",
    "    merge(\n",
    "        perf[perf.algorithm != \"ERM\"].drop(columns=['val_env', 'hparams_seed', 'variable']), \n",
    "        on=['test_env', 'trial'],\n",
    "        suffixes=['_erm', '_dg']\n",
    "    )\n",
    "\n",
    "def compare_to_erm(df, test=ttest_rel):\n",
    "    # Perform a one-sided test for hypothesis DG != ERM\n",
    "    res = test(df.value_dg.to_numpy(), df.value_erm.to_numpy(), alternative=\"two-sided\")\n",
    "    return res.pvalue\n",
    "\n",
    "paired.groupby(['algorithm', 'test_env']).apply(compare_to_erm, test=wilcoxon).loc[algos, :]\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "clinicaldg-new",
   "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": "3.9.13"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "b38954ae0c531d88f125c33d0f04941ad01e34d8f09fddf489edf91850c80e8c"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}