code/paper-Figure-8-kernel-regression-of-all-variables.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 📄 Compute the kernel regression errors for all variables (**Figure 8**)\n",
    "\n",
    "***"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -i imports.py\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_components = 3\n",
    "sample_percentage = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Removing N2.\n",
      "\n",
      "The data set has 14550 observations.\n",
      "\n",
      "Using: T, H2, O2, OH, H2O, CO, CO2 as target variables.\n"
     ]
    }
   ],
   "source": [
    "data_tag = 'clustered-flamelet-CO-H2-10-1'\n",
    "# data_tag = 'clustered-flamelet-H2'\n",
    "# data_tag = 'clustered-flamelet-C2H4'\n",
    "%run -i load-SLFM-data.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set the kernel regression parameters here:\n",
    "train_perc = 80\n",
    "test_perc = 20\n",
    "\n",
    "if data_tag == 'clustered-flamelet-H2':\n",
    "    n_neighbors = 250\n",
    "    n_neighbors_subset = 250\n",
    "if data_tag == 'clustered-flamelet-CO-H2-10-1':\n",
    "    n_neighbors = 50\n",
    "    n_neighbors_subset = 50\n",
    "if data_tag == 'clustered-flamelet-C2H4':\n",
    "    n_neighbors = 50\n",
    "    n_neighbors_subset = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "if data_tag == 'clustered-flamelet-H2':\n",
    "\n",
    "    regressed_variables_names = ['$S_{Z,1}$',\n",
    "     '$S_{Z,2}$',\n",
    "     '$S_{Z,3}$',\n",
    "     '$T$',\n",
    "     'H',\n",
    "     'H$_2$',\n",
    "     'O',\n",
    "     'OH',\n",
    "     'H$_2$O',\n",
    "     'O$_2$',\n",
    "     'HO$_2$',\n",
    "     'H$_2$O$_2$']\n",
    "\n",
    "if data_tag == 'clustered-flamelet-CO-H2-10-1':\n",
    "    \n",
    "    regressed_variables_names = ['$S_{Z,1}$',\n",
    "     '$S_{Z,2}$',\n",
    "     '$S_{Z,3}$',\n",
    "     '$T$',\n",
    "     'H$_2$',\n",
    "     'O$_2$',\n",
    "     'O',\n",
    "     'OH',\n",
    "     'H$_2$O',\n",
    "     'H',\n",
    "     'HO$_2$',\n",
    "     'CO',\n",
    "     'CO$_2$',\n",
    "     'HCO']\n",
    "    \n",
    "if data_tag == 'clustered-flamelet-C2H4':\n",
    "    \n",
    "    regressed_variables_names = ['$S_{Z,1}$',\n",
    "     '$S_{Z,2}$',\n",
    "     '$S_{Z,3}$',\n",
    "     '$T$',\n",
    "     'H$_2$',\n",
    "     'H',\n",
    "     'O',\n",
    "     'O$_2$',\n",
    "     'OH',\n",
    "     'H$_2$O',\n",
    "     'HO$_2$',\n",
    "     'H$_2$O$_2$',\n",
    "     'CH',\n",
    "     'CH$_2$',\n",
    "     'CH$_2$*',\n",
    "     'CH$_3$',\n",
    "     'CH$_4$',\n",
    "     'CO',\n",
    "     'CO$_2$',\n",
    "     'HCO',\n",
    "     'CH$_2$O',\n",
    "     'CH$_3$O',\n",
    "     'C$_2$H$_2$',\n",
    "     'H$_2$CC',\n",
    "     'C$_2$H$_3$',\n",
    "     'C$_2$H$_4$',\n",
    "     'C$_2$H$_5$',\n",
    "     'C$_2$H$_6$',\n",
    "     'HCCO',\n",
    "     'CH$_2$CO',\n",
    "     'CH$_2$CHO',\n",
    "     'CH$_3$CHO',\n",
    "     'aC$_3$H$_5$',\n",
    "     'C$_3$H$_6$',\n",
    "     'nC$_3$H$_7$']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizing_higlight = [i for i in range(0,n_components)] + list(np.array(target_variables_indices) + n_components)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "\n",
    "## Run nonlinear regression on a manifold computed from the full state vector, $\\mathbf{X}$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "if data_tag == 'clustered-flamelet-H2':\n",
    "    scaling = 'max'\n",
    "if data_tag == 'clustered-flamelet-CO-H2-10-1':\n",
    "    scaling = '-1to1'\n",
    "if data_tag == 'clustered-flamelet-C2H4':\n",
    "    scaling = 'range'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Total time: 0.1 minutes.\n"
     ]
    }
   ],
   "source": [
    "total_tic = time.perf_counter()\n",
    "\n",
    "pca = reduction.PCA(state_space, scaling=scaling, n_components=n_components)\n",
    "PCs = pca.transform(state_space)\n",
    "PC_sources = pca.transform(state_space_sources, nocenter=True)\n",
    "\n",
    "symlog_PC_sources = preprocess.log_transform(PC_sources, method='continuous-symlog', threshold=1.e-4)\n",
    "\n",
    "OUTPUT = np.hstack((PC_sources, state_space))\n",
    "(_,n_regressed_variables) = np.shape(OUTPUT)\n",
    "(PCs_pp, centers_PCs, scales_PCs) = preprocess.center_scale(PCs, '-1to1')\n",
    "\n",
    "sample_random = preprocess.DataSampler(np.zeros((n_observations,)).astype(int), random_seed=random_seed, verbose=False)\n",
    "(idx_train, idx_test) = sample_random.random(train_perc)\n",
    "\n",
    "PCs_pp_train = PCs_pp[idx_train,:]\n",
    "OUTPUT_train = OUTPUT[idx_train,:]\n",
    "PCs_pp_test = PCs_pp[idx_test,:]\n",
    "OUTPUT_test = OUTPUT[idx_test,:]\n",
    "\n",
    "model = analysis.KReg(PCs_pp_train, OUTPUT_train)\n",
    "\n",
    "OUTPUT_predicted_test = model.predict(PCs_pp_test, bandwidth='nearest_neighbors_isotropic', n_neighbors=n_neighbors)\n",
    "OUTPUT_predicted_train = model.predict(PCs_pp_train, bandwidth='nearest_neighbors_isotropic', n_neighbors=n_neighbors)\n",
    "\n",
    "NRMSE_collected_test = []\n",
    "for i in range(0, n_regressed_variables):\n",
    "    NRMSE_test = analysis.normalized_root_mean_squared_error(OUTPUT_test[:,i], OUTPUT_predicted_test[:,i], norm='std')\n",
    "    NRMSE_collected_test.append(NRMSE_test)\n",
    "\n",
    "NRMSE_collected_train = []\n",
    "for i in range(0, n_regressed_variables):\n",
    "    NRMSE_train = analysis.normalized_root_mean_squared_error(OUTPUT_train[:,i], OUTPUT_predicted_train[:,i], norm='std')\n",
    "    NRMSE_collected_train.append(NRMSE_train)\n",
    "\n",
    "total_toc = time.perf_counter()\n",
    "print(f'\\nTotal time: {(total_toc - total_tic)/60:0.1f} minutes.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "\n",
    "## Run nonlinear regression on a manifold computed from the optimized state vector subset, $\\mathbf{X}_S$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Hydrogen"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "if data_tag == 'clustered-flamelet-H2':\n",
    "\n",
    "    scaling = 'auto'\n",
    "    selected_variables_list = [6, 3, 0, 1, 2, 4]\n",
    "    print(state_space_names[selected_variables_list])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Syngas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['CO2' 'CO' 'H' 'O' 'OH']\n"
     ]
    }
   ],
   "source": [
    "if data_tag == 'clustered-flamelet-CO-H2-10-1':\n",
    "\n",
    "    scaling = 'range'\n",
    "    selected_variables_list = [9, 8, 6, 3, 4]\n",
    "    print(state_space_names[selected_variables_list])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Ethylene"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "if data_tag == 'clustered-flamelet-C2H4':\n",
    "    \n",
    "    scaling = 'auto'\n",
    "    selected_variables_list = [15,6,4,5,16,12,19,22,0,2,26,1]\n",
    "    print(state_space_names[sorted(selected_variables_list)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Total time: 0.1 minutes.\n"
     ]
    }
   ],
   "source": [
    "total_tic = time.perf_counter()\n",
    "\n",
    "subset_pca = reduction.PCA(state_space[:,selected_variables_list], scaling=scaling, n_components=n_components)\n",
    "subset_PCs = subset_pca.transform(state_space[:,selected_variables_list])\n",
    "subset_PC_sources = subset_pca.transform(state_space_sources[:,selected_variables_list], nocenter=True)\n",
    "\n",
    "symlog_subset_PC_sources = preprocess.log_transform(subset_PC_sources, method='continuous-symlog', threshold=1.e-4)\n",
    "\n",
    "subset_OUTPUT = np.hstack((subset_PC_sources, state_space))\n",
    "(_,n_regressed_variables) = np.shape(subset_OUTPUT)\n",
    "(PCs_pp, centers_PCs, scales_PCs) = preprocess.center_scale(subset_PCs, '-1to1')\n",
    "\n",
    "sample_random = preprocess.DataSampler(np.zeros((n_observations,)).astype(int), random_seed=random_seed, verbose=False)\n",
    "(idx_train, idx_test) = sample_random.random(train_perc)\n",
    "\n",
    "PCs_pp_train = PCs_pp[idx_train,:]\n",
    "subset_OUTPUT_train = subset_OUTPUT[idx_train,:]\n",
    "PCs_pp_test = PCs_pp[idx_test,:]\n",
    "subset_OUTPUT_test = subset_OUTPUT[idx_test,:]\n",
    "\n",
    "model = analysis.KReg(PCs_pp_train, subset_OUTPUT_train)\n",
    "\n",
    "subset_OUTPUT_predicted_test = model.predict(PCs_pp_test, bandwidth='nearest_neighbors_isotropic', n_neighbors=n_neighbors_subset)\n",
    "subset_OUTPUT_predicted_train = model.predict(PCs_pp_train, bandwidth='nearest_neighbors_isotropic', n_neighbors=n_neighbors_subset)\n",
    "\n",
    "subset_NRMSE_collected_test = []\n",
    "for i in range(0, n_regressed_variables):\n",
    "    NRMSE_test = analysis.normalized_root_mean_squared_error(subset_OUTPUT_test[:,i], subset_OUTPUT_predicted_test[:,i], norm='std')\n",
    "    subset_NRMSE_collected_test.append(NRMSE_test)\n",
    "    \n",
    "subset_NRMSE_collected_train = []\n",
    "for i in range(0, n_regressed_variables):\n",
    "    NRMSE_train = analysis.normalized_root_mean_squared_error(subset_OUTPUT_train[:,i], subset_OUTPUT_predicted_train[:,i], norm='std')\n",
    "    subset_NRMSE_collected_train.append(NRMSE_train)\n",
    "\n",
    "total_toc = time.perf_counter()\n",
    "print(f'\\nTotal time: {(total_toc - total_tic)/60:0.1f} minutes.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "\n",
    "## Compare regression results "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "regression_metrics_test = analysis.RegressionAssessment(OUTPUT_test, OUTPUT_predicted_test, variable_names=regressed_variables_names)\n",
    "subset_regression_metrics_test = analysis.RegressionAssessment(subset_OUTPUT_test, subset_OUTPUT_predicted_test, variable_names=regressed_variables_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style  type=\"text/css\" >\n",
       "</style><table id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122\" ><thead>    <tr>        <th class=\"blank level0\" ></th>        <th class=\"col_heading level0 col0\" >NRMSE</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row0\" class=\"row_heading level0 row0\" >$S_{Z,1}$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row0_col0\" class=\"data row0 col0\" >0.07720</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row1\" class=\"row_heading level0 row1\" >$S_{Z,2}$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row1_col0\" class=\"data row1 col0\" >0.06154</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row2\" class=\"row_heading level0 row2\" >$S_{Z,3}$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row2_col0\" class=\"data row2 col0\" >0.07151</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row3\" class=\"row_heading level0 row3\" >$T$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row3_col0\" class=\"data row3 col0\" >0.03349</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row4\" class=\"row_heading level0 row4\" >H$_2$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row4_col0\" class=\"data row4 col0\" >0.02462</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row5\" class=\"row_heading level0 row5\" >O$_2$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row5_col0\" class=\"data row5 col0\" >0.01544</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row6\" class=\"row_heading level0 row6\" >O</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row6_col0\" class=\"data row6 col0\" >0.02378</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row7\" class=\"row_heading level0 row7\" >OH</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row7_col0\" class=\"data row7 col0\" >0.02771</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row8\" class=\"row_heading level0 row8\" >H$_2$O</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row8_col0\" class=\"data row8 col0\" >0.04815</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row9\" class=\"row_heading level0 row9\" >H</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row9_col0\" class=\"data row9 col0\" >0.04144</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row10\" class=\"row_heading level0 row10\" >HO$_2$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row10_col0\" class=\"data row10 col0\" >0.04887</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row11\" class=\"row_heading level0 row11\" >CO</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row11_col0\" class=\"data row11 col0\" >0.02456</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row12\" class=\"row_heading level0 row12\" >CO$_2$</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row12_col0\" class=\"data row12 col0\" >0.03531</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122level0_row13\" class=\"row_heading level0 row13\" >HCO</th>\n",
       "                        <td id=\"T_79fa05c6_f194_11ec_b4ff_acde48001122row13_col0\" class=\"data row13 col0\" >0.07009</td>\n",
       "            </tr>\n",
       "    </tbody></table>"
      ],
      "text/plain": [
       "<pandas.io.formats.style.Styler at 0x1065ccc50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "regression_metrics_test.print_metrics(table_format=['pandas'],\n",
    "                                   float_format='.5f',\n",
    "                                   metrics=['NRMSE'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style  type=\"text/css\" >\n",
       "    #T_79fd5276_f194_11ec_b4ff_acde48001122row0_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row1_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row2_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row3_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row4_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row5_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row6_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row7_col0 {\n",
       "            : ;\n",
       "            background-color:  salmon;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row8_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row9_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row10_col0 {\n",
       "            : ;\n",
       "            background-color:  salmon;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row11_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row12_col0 {\n",
       "            background-color:  lightgreen;\n",
       "            : ;\n",
       "        }    #T_79fd5276_f194_11ec_b4ff_acde48001122row13_col0 {\n",
       "            : ;\n",
       "            background-color:  salmon;\n",
       "        }</style><table id=\"T_79fd5276_f194_11ec_b4ff_acde48001122\" ><thead>    <tr>        <th class=\"blank level0\" ></th>        <th class=\"col_heading level0 col0\" >NRMSE</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row0\" class=\"row_heading level0 row0\" >$S_{Z,1}$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row0_col0\" class=\"data row0 col0\" >0.03200</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row1\" class=\"row_heading level0 row1\" >$S_{Z,2}$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row1_col0\" class=\"data row1 col0\" >0.05717</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row2\" class=\"row_heading level0 row2\" >$S_{Z,3}$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row2_col0\" class=\"data row2 col0\" >0.04615</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row3\" class=\"row_heading level0 row3\" >$T$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row3_col0\" class=\"data row3 col0\" >0.02313</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row4\" class=\"row_heading level0 row4\" >H$_2$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row4_col0\" class=\"data row4 col0\" >0.01726</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row5\" class=\"row_heading level0 row5\" >O$_2$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row5_col0\" class=\"data row5 col0\" >0.01521</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row6\" class=\"row_heading level0 row6\" >O</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row6_col0\" class=\"data row6 col0\" >0.01723</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row7\" class=\"row_heading level0 row7\" >OH</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row7_col0\" class=\"data row7 col0\" >0.03447</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row8\" class=\"row_heading level0 row8\" >H$_2$O</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row8_col0\" class=\"data row8 col0\" >0.03658</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row9\" class=\"row_heading level0 row9\" >H</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row9_col0\" class=\"data row9 col0\" >0.03002</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row10\" class=\"row_heading level0 row10\" >HO$_2$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row10_col0\" class=\"data row10 col0\" >0.10907</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row11\" class=\"row_heading level0 row11\" >CO</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row11_col0\" class=\"data row11 col0\" >0.01792</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row12\" class=\"row_heading level0 row12\" >CO$_2$</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row12_col0\" class=\"data row12 col0\" >0.02388</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                        <th id=\"T_79fd5276_f194_11ec_b4ff_acde48001122level0_row13\" class=\"row_heading level0 row13\" >HCO</th>\n",
       "                        <td id=\"T_79fd5276_f194_11ec_b4ff_acde48001122row13_col0\" class=\"data row13 col0\" >0.36751</td>\n",
       "            </tr>\n",
       "    </tbody></table>"
      ],
      "text/plain": [
       "<pandas.io.formats.style.Styler at 0x1a1b52ff50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "subset_regression_metrics_test.print_metrics(table_format=['pandas'],\n",
    "                                   float_format='.5f',\n",
    "                                   metrics=['NRMSE'],\n",
    "                                   comparison=regression_metrics_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "\n",
    "## ✏️ Plot NRMSE comparison on test data only"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x277.2 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(6, 0.35*len(state_space_names)))\n",
    "fig.tight_layout(pad=-30)\n",
    "spec = fig.add_gridspec(ncols=1, nrows=1)\n",
    "widths = [1]\n",
    "heights = [1]\n",
    "spec = fig.add_gridspec(ncols=1, nrows=1, width_ratios=widths, height_ratios=heights)\n",
    "\n",
    "x_range = np.array([i for i in range(0,n_regressed_variables)])\n",
    "\n",
    "ax = fig.add_subplot(spec[0,0])\n",
    "\n",
    "plt.scatter(subset_NRMSE_collected_test, x_range, c='r', marker='<', s=120, edgecolor='none', zorder=5, label='Optimized subset')\n",
    "plt.scatter(NRMSE_collected_test, x_range, c='k', marker='o', s=60, edgecolor='none',  zorder=5, label='Full state vector')\n",
    "\n",
    "plt.yticks(x_range, regressed_variables_names, fontsize=12)\n",
    "plt.grid(alpha=0.3, zorder=0)\n",
    "plt.ylim([-0.5,len(x_range)-0.5])\n",
    "plt.xlabel('NRMSE [$-$]', fontsize=16)\n",
    "\n",
    "ax.set_xscale('log')\n",
    "\n",
    "for i in optimizing_higlight:\n",
    "    ax.get_yticklabels()[i].set_color('r')\n",
    "    ax.get_yticklabels()[i].set_fontweight(1000)\n",
    "\n",
    "if data_tag == 'clustered-flamelet-H2':\n",
    "    ax.set_xticks([0.07,0.08,0.09,0.1,0.2,0.3,0.4,0.5,0.6])\n",
    "    ax.get_xaxis().set_major_formatter(matplotlib.ticker.ScalarFormatter())\n",
    "    plt.xlim([0.065,0.6])\n",
    "    lgnd = plt.legend(ncol=1, fontsize=12, markerscale=marker_scale_legend, loc=\"upper left\")\n",
    "    \n",
    "if data_tag == 'clustered-flamelet-CO-H2-10-1':\n",
    "    ax.set_xticks([0.01,0.02,0.03,0.04,0.05,0.1,0.2,0.3,0.4,0.5])\n",
    "    ax.get_xaxis().set_major_formatter(matplotlib.ticker.ScalarFormatter())\n",
    "    plt.xlim([0.01,0.5])\n",
    "    lgnd = plt.legend(ncol=1, fontsize=12, markerscale=marker_scale_legend, loc=\"center right\")\n",
    "\n",
    "if data_tag == 'clustered-flamelet-C2H4':\n",
    "    ax.set_xticks([0.02,0.03,0.04,0.05,0.1,0.2,0.3,0.4,0.5,0.6])\n",
    "    ax.get_xaxis().set_major_formatter(matplotlib.ticker.ScalarFormatter())\n",
    "    plt.xlim([0.025,0.6])\n",
    "    lgnd = plt.legend(ncol=1, fontsize=12, markerscale=marker_scale_legend, loc=\"upper left\")\n",
    "    \n",
    "plt.xticks(fontsize=12, rotation=90)\n",
    "plt.gca().invert_yaxis()\n",
    "plt.grid(b=True, which='minor', alpha=0.3, zorder=0)\n",
    "plt.gca().spines[\"top\"].set_visible(False)\n",
    "plt.gca().spines[\"right\"].set_visible(False)\n",
    "\n",
    "for i in range(0, n_regressed_variables):\n",
    "\n",
    "    x_values = [x_range[i], x_range[i]]\n",
    "    y_values = [NRMSE_collected_test[i], subset_NRMSE_collected_test[i]]\n",
    "    plt.plot(y_values, x_values, color='k', ls='-', lw=1, zorder=3)\n",
    "    \n",
    "plt.savefig('../figures/Figure-8.png', dpi=200, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***"
   ]
  }
 ],
 "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}