Оптимизация гиперпараметров.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "#импорт библиотек\n",
    "import numpy as np #для матричных вычислений\n",
    "import pandas as pd #для анализа и предобработки данных\n",
    "import matplotlib.pyplot as plt #для визуализации\n",
    "import seaborn as sns #для визуализации\n",
    "\n",
    "from sklearn import linear_model #линейные моделиё\n",
    "from sklearn import tree #деревья решений\n",
    "from sklearn import ensemble #ансамбли\n",
    "from sklearn import metrics #метрики\n",
    "from sklearn import preprocessing #предобработка\n",
    "from sklearn import model_selection #методы разделения и валидации\n",
    "from sklearn.model_selection import train_test_split #сплитование выборки\n",
    "from sklearn.metrics import f1_score\n",
    "%matplotlib inline\n",
    "plt.style.use('seaborn')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.read_csv('_train_sem09__1_.zip')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "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>Activity</th>\n",
       "      <th>D1</th>\n",
       "      <th>D2</th>\n",
       "      <th>D3</th>\n",
       "      <th>D4</th>\n",
       "      <th>D5</th>\n",
       "      <th>D6</th>\n",
       "      <th>D7</th>\n",
       "      <th>D8</th>\n",
       "      <th>D9</th>\n",
       "      <th>...</th>\n",
       "      <th>D1767</th>\n",
       "      <th>D1768</th>\n",
       "      <th>D1769</th>\n",
       "      <th>D1770</th>\n",
       "      <th>D1771</th>\n",
       "      <th>D1772</th>\n",
       "      <th>D1773</th>\n",
       "      <th>D1774</th>\n",
       "      <th>D1775</th>\n",
       "      <th>D1776</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.497009</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.132956</td>\n",
       "      <td>0.678031</td>\n",
       "      <td>0.273166</td>\n",
       "      <td>0.585445</td>\n",
       "      <td>0.743663</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>0.366667</td>\n",
       "      <td>0.606291</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.111209</td>\n",
       "      <td>0.803455</td>\n",
       "      <td>0.106105</td>\n",
       "      <td>0.411754</td>\n",
       "      <td>0.836582</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 1777 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   Activity        D1        D2    D3   D4        D5        D6        D7  \\\n",
       "0         1  0.000000  0.497009  0.10  0.0  0.132956  0.678031  0.273166   \n",
       "1         1  0.366667  0.606291  0.05  0.0  0.111209  0.803455  0.106105   \n",
       "\n",
       "         D8        D9  ...  D1767  D1768  D1769  D1770  D1771  D1772  D1773  \\\n",
       "0  0.585445  0.743663  ...      0      0      0      0      0      0      0   \n",
       "1  0.411754  0.836582  ...      1      1      1      1      0      1      0   \n",
       "\n",
       "   D1774  D1775  D1776  \n",
       "0      0      0      0  \n",
       "1      0      1      0  \n",
       "\n",
       "[2 rows x 1777 columns]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.head(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Проверяем выборку на дисбаланс"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Activity is 0: 45.77%\n",
      "Activity is 1: 54.23%\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfgAAAFXCAYAAABOYlxEAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAb/UlEQVR4nO3de3BU9f3/8ddmFxCSjUlKYGCQqygCApMGYm2gnQITVBCLSC41OkKrIheBimCAACYQGDTMNFxE6D+NpSIaNFWrDlTMF0PBCQ2YKAgzSbi2DRIlWZRc9vz+cNxfUyAsmrOHfHg+/mLPnj37XmZ2nuec3Zx1WZZlCQAAGCXM6QEAAEDrI/AAABiIwAMAYCACDwCAgQg8AAAGIvAAABjI4/QAraW6utbpEQAACKnYWO8V7+MIHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxnza3IAzFOZH+f0CECr6J1+IOTPyRE8AAAGsu0IvqGhQRkZGTp16pTq6+s1ffp03XrrrVq4cKFcLpf69++vpUuXKiwsTOvWrdPu3bvl8XiUkZGhIUOGqKqq6rLrAgCAq7OtmIWFhYqKitLWrVu1ZcsWZWVlKScnR3PmzNHWrVtlWZZ27dql8vJy7d+/X9u3b1dubq6WL18uSZddFwAABMe2wI8bN05PP/20JMmyLLndbpWXl2vEiBGSpFGjRqm4uFglJSVKTEyUy+VS9+7d1dTUpHPnzl12XQAAEBzbTtGHh4dLkurq6jR79mzNmTNHq1evlsvlCtxfW1ururo6RUVFNXtcbW2tLMu6ZN2WREd3ksfjtufFAHBEpdMDAK0kNtYb8ue09Vv0Z86c0YwZM5SWlqYJEyZozZo1gft8Pp8iIyMVEREhn8/XbLnX6232efv367akpuZC678AAABaQXV1ywepP1RLOw62naI/e/aspk6dqvnz52vy5MmSpIEDB2rfvn2SpKKiIsXHxysuLk579uyR3+/X6dOn5ff7FRMTc9l1AQBAcFyWZVl2bDg7O1t/+9vf1Ldv38CyRYsWKTs7Ww0NDerbt6+ys7PldruVl5enoqIi+f1+Pffcc4qPj1dFRYWWLFlyybpXYtfeEQDn8HfwMIVdfwff0hG8bYEPNQIPmIfAwxROBJ4/LAcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAM5LFz4wcPHtQLL7yg/Px8zZ07V2fPnpUknTp1SkOHDtXatWs1ffp01dTUqF27durQoYO2bNmiqqoqLVy4UC6XS/3799fSpUsVFsa+CAAAwbIt8Js3b1ZhYaE6duwoSVq7dq0k6euvv9Yjjzyi5557TpJUVVWld955Ry6XK/DYnJwczZkzRwkJCcrMzNSuXbs0duxYu0YFAMA4th0W9+zZU3l5eZcsz8vL08MPP6wuXbro7NmzOn/+vJ588kmlpqbqww8/lCSVl5drxIgRkqRRo0apuLjYrjEBADCSbUfwSUlJOnnyZLNlX375pfbu3Rs4em9oaNDUqVP1yCOP6Ouvv1ZqaqqGDBkiy7ICR/Th4eGqra296vNFR3eSx+Nu/RcCwDGVTg8AtJLYWG/In9PWz+D/13vvvafx48fL7f4uxJ07d1ZKSoo8Ho9+8pOf6I477lBFRUWzz9t9Pp8iIyOvuu2amgu2zQ0AwI9RXX31A9UfoqUdh5B+c23v3r0aNWpU4HZxcbGefvppSd+F/OjRo+rbt68GDhyoffv2SZKKiooUHx8fyjEBAGjzQhr4iooK3XLLLYHbv/jFL9S7d29NmTJF06ZN07x58xQTE6MFCxYoLy9PycnJamhoUFJSUijHBACgzXNZlmU5PURrsOv0hyTlTtpg27aBUJlX8JTTI1yzyvw4p0cAWkXv9AO2bPe6OUUPAABCg8ADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIEIPAAABiLwAAAYiMADAGAgAg8AgIFsDfzBgweVnp4uSfrss880cuRIpaenKz09Xe+++64kad26dZo8ebJSUlJ06NAhSVJVVZVSU1OVlpampUuXyu/32zkmAADG8di14c2bN6uwsFAdO3aUJJWXl+uxxx7T1KlTA+uUl5dr//792r59u86cOaNZs2bpjTfeUE5OjubMmaOEhARlZmZq165dGjt2rF2jAgBgHNuO4Hv27Km8vLzA7bKyMu3evVu/+c1vlJGRobq6OpWUlCgxMVEul0vdu3dXU1OTzp07p/Lyco0YMUKSNGrUKBUXF9s1JgAARrLtCD4pKUknT54M3B4yZIgeeughDR48WBs3btT69evl9XoVFRUVWCc8PFy1tbWyLEsul6vZsquJju4kj8fd6q8DMEVsrNfpEa5ZpdMDAK3EifefbYH/X2PHjlVkZGTg31lZWRo9erR8Pl9gHZ/PJ6/Xq7CwsGbLvn9cS2pqLrT+0IBBqquvvqMMwB52vf9a2nEI2bfop02bFvgS3d69ezVo0CDFxcVpz5498vv9On36tPx+v2JiYjRw4EDt27dPklRUVKT4+PhQjQkAgBFCdgS/bNkyZWVlqV27durcubOysrIUERGh+Ph4JScny+/3KzMzU5K0YMECLVmyRLm5uerbt6+SkpJCNSYAAEZwWZZlOT1Ea7Dz9GPupA22bRsIlXkFTzk9wjWrzI9zegSgVfROP2DLdq+LU/QAACB0CDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGIjAAwBgIAIPAICBCDwAAAYi8AAAGMhj58YPHjyoF154Qfn5+fr888+VlZUlt9ut9u3ba/Xq1ercubOys7N14MABhYeHS5I2bNighoYGPfPMM/r222/VpUsX5eTkqGPHjnaOCgCAUWw7gt+8ebMWL16sixcvSpJWrFihJUuWKD8/X2PHjtXmzZslSeXl5dqyZYvy8/OVn58vr9erDRs2aPz48dq6dasGDhyobdu22TUmAABGsi3wPXv2VF5eXuB2bm6u7rjjDklSU1OTOnToIL/fr6qqKmVmZiolJUWvv/66JKmkpEQjR46UJI0aNUrFxcV2jQkAgJFsO0WflJSkkydPBm536dJFknTgwAG98sor+vOf/6wLFy7o4Ycf1mOPPaampiY98sgjGjx4sOrq6uT1eiVJ4eHhqq2tverzRUd3ksfjtufFAAaIjfU6PcI1q3R6AKCVOPH+s/Uz+P/17rvvauPGjXr55ZcVExMTiPr3n6/fddddOnz4sCIiIuTz+XTTTTfJ5/MpMjLyqtuuqblg9/hAm1ZdffUdZQD2sOv919KOQ8i+Rf/WW2/plVdeUX5+vm655RZJUmVlpVJTU9XU1KSGhgYdOHBAgwYNUlxcnD766CNJUlFRkX7605+GakwAAIwQkiP4pqYmrVixQt26ddOsWbMkScOHD9fs2bM1ceJETZkyRe3atdPEiRPVv39/TZ8+XQsWLNBrr72m6Ohovfjii6EYEwAAY7gsy7KcHqI12Hn6MXfSBtu2DYTKvIKnnB7hmlXmxzk9AtAqeqcfsGW718UpegAAEDoEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEBBBT4rK+uSZQsWLGj1YQAAQOto8Vr0ixYt0okTJ1RWVqajR48Gljc2Ngb1E64AAMAZLQZ++vTpOnXqlFasWKGZM2cGlrvdbvXr18/24QAAwA/TYuB79OihHj16qLCwUHV1daqtrdX3v01z4cIFRUVFhWJGAABwjYL6udhNmzZp06ZNzYLucrm0a9cuu+YCAAA/QlCB3759u3bu3KmYmBi75wEAAK0gqG/Rd+vWTTfffLPdswAAgFYS1BF87969lZaWpoSEBLVv3z6w/L+/eAcAAK4fQQW+a9eu6tq1q92zAACAVhJU4DlSBwCgbQkq8AMGDJDL5Wq2rEuXLvroo49sGQoAAPw4QQX+8OHDgX83NDRo586dKi0ttWsmAADwI13zj820a9dO99xzj/7xj3/YMQ8AAGgFQR3Bv/nmm4F/W5alo0ePql27dnbNBAAAfqSgAr9v375mt6Ojo7V27VpbBgIAAD9eUIHPyclRQ0ODKioq1NTUpP79+8vjCeqhAADAAUFVuqysTLNnz1ZUVJT8fr/Onj2r9evXa+jQoXbPBwAAfoCgAp+dna21a9cGgl5aWqqsrCy9/vrrtg4HAAB+mKC+RX/hwoVmR+vDhg3TxYsXr/q4gwcPKj09XZJUVVWl1NRUpaWlaenSpfL7/ZKkdevWafLkyUpJSdGhQ4daXBcAAAQnqMDffPPN2rlzZ+D2zp07r/pb8Js3b9bixYsDOwI5OTmaM2eOtm7dKsuytGvXLpWXl2v//v3avn27cnNztXz58iuuCwAAghdU4LOysrRmzRolJCQoISFBixYt0vPPP9/iY3r27Km8vLzA7fLyco0YMUKSNGrUKBUXF6ukpESJiYlyuVzq3r27mpqadO7cucuuCwAAghfUZ/BFRUXq2LGjduzYoePHj2vu3Lnav3+/+vTpc8XHJCUl6eTJk4HblmUFLncbHh6u2tpa1dXVNTsT8P3yy617NdHRneTxuIN5OcANKTbW6/QI16zS6QGAVuLE+y+owL/22mvavn27OnbsqAEDBqigoEBTpkxRcnJy0E8UFvb/Txb4fD5FRkYqIiJCPp+v2XKv13vZda+mpuZC0LMAN6Lq6qvvKAOwh13vv5Z2HII6Rd/Q0NDsynU/5Cp2AwcODFwwp6ioSPHx8YqLi9OePXvk9/t1+vRp+f1+xcTEXHZdAAAQvKCO4MeMGaNHH31U99xzjyTpgw8+0OjRo6/piRYsWKAlS5YoNzdXffv2VVJSktxut+Lj45WcnCy/36/MzMwrrgsAAILnsizLCmbF9957T5988ok8Ho+GDx+uMWPG2D3bNbHz9GPupA22bRsIlXkFTzk9wjWrzI9zegSgVfROP2DLdls6RR/09WbHjRuncePGtcpAAADAXtf8c7EAAOD6R+ABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADAQgQcAwEAEHgAAAxF4AAAMROABADCQJ5RPVlBQoB07dkiSLl68qM8//1y5ublavXq1unXrJkmaNWuW4uPjtWzZMh05ckTt27dXdna2evXqFcpRAQBo00Ia+EmTJmnSpEmSpOXLl+vBBx9UWVmZ5s+fr6SkpMB6H3zwgerr67Vt2zaVlpZq1apV2rhxYyhHBQCgTXPkFP2nn36qY8eOKTk5WeXl5XrjjTeUlpamVatWqbGxUSUlJRo5cqQkadiwYSorK3NiTAAA2qyQHsF/b9OmTZoxY4Yk6ec//7nGjBmjHj16aOnSpXr11VdVV1eniIiIwPput1uNjY3yeK48bnR0J3k8bttnB9qq2Fiv0yNcs0qnBwBaiRPvv5AH/vz586qoqNBdd90lSXrwwQcVGRkpSRo9erTef/99eb1e+Xy+wGP8fn+LcZekmpoL9g0NGKC6utbpEYAbll3vv5Z2HEJ+iv6TTz7Rz372M0mSZVm6//779a9//UuStHfvXg0aNEhxcXEqKiqSJJWWluq2224L9ZgAALRpIT+Cr6ioUI8ePSRJLpdL2dnZmjlzpm666Sb169dPU6ZMkdvt1scff6yUlBRZlqWVK1eGekwAANq0kAf+t7/9bbPbiYmJSkxMvGS9559/PlQjAQBgHC50AwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABjIE+on/PWvf62IiAhJUo8ePZScnKwVK1bI7XYrMTFRM2fOlN/v17Jly3TkyBG1b99e2dnZ6tWrV6hHBQCgzQpp4C9evCjLspSfnx9YNnHiROXl5emWW27R448/rs8++0wnT55UfX29tm3bptLSUq1atUobN24M5agAALRpIQ384cOH9c0332jq1KlqbGzUrFmzVF9fr549e0qSEhMTVVxcrOrqao0cOVKSNGzYMJWVlYVyTAAA2ryQBv6mm27StGnT9NBDD6myslK/+93vFBkZGbg/PDxcJ06cUF1dXeA0viS53W41NjbK47nyuNHRneTxuG2dH2jLYmO9To9wzSqdHgBoJU68/0Ia+D59+qhXr15yuVzq06ePvF6vvvrqq8D9Pp9PkZGR+vbbb+Xz+QLL/X5/i3GXpJqaC3aNDRihurrW6RGAG5Zd77+WdhxC+i36119/XatWrZIk/fvf/9Y333yjTp066fjx47IsS3v27FF8fLzi4uJUVFQkSSotLdVtt90WyjEBAGjzQnoEP3nyZD333HNKTU2Vy+XSypUrFRYWpmeeeUZNTU1KTEzU0KFDdeedd+rjjz9WSkqKLMvSypUrQzkmAABtXkgD3759e7344ouXLH/ttdea3Q4LC9Pzzz8fqrEAADAOF7oBAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADOQJ5ZM1NDQoIyNDp06dUn19vaZPn65u3brpiSeeUO/evSVJqampuvfee7Vu3Trt3r1bHo9HGRkZGjJkSChHBQCgTQtp4AsLCxUVFaU1a9boq6++0gMPPKAZM2boscce09SpUwPrlZeXa//+/dq+fbvOnDmjWbNm6Y033gjlqAAAtGkhDfy4ceOUlJQkSbIsS263W2VlZaqoqNCuXbvUq1cvZWRkqKSkRImJiXK5XOrevbuampp07tw5xcTEhHJcAADarJAGPjw8XJJUV1en2bNna86cOaqvr9dDDz2kwYMHa+PGjVq/fr28Xq+ioqKaPa62trbFwEdHd5LH47b7JQBtVmys1+kRrlml0wMArcSJ919IAy9JZ86c0YwZM5SWlqYJEybo/PnzioyMlCSNHTtWWVlZGj16tHw+X+AxPp9PXm/L/zk1NRdsnRto66qra50eAbhh2fX+a2nHIaTfoj979qymTp2q+fPna/LkyZKkadOm6dChQ5KkvXv3atCgQYqLi9OePXvk9/t1+vRp+f1+Ts8DAHANQnoE/9JLL+n8+fPasGGDNmzYIElauHChVq5cqXbt2qlz587KyspSRESE4uPjlZycLL/fr8zMzFCOCQBAm+eyLMtyeojWYOfpx9xJG2zbNhAq8wqecnqEa1aZH+f0CECr6J1+wJbtXjen6AEAQGgQeAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwEIEHAMBABB4AAAMReAAADETgAQAwkMfpAa7E7/dr2bJlOnLkiNq3b6/s7Gz16tXL6bEAAGgTrtsj+J07d6q+vl7btm3T73//e61atcrpkQAAaDOu28CXlJRo5MiRkqRhw4aprKzM4YkAAGg7rttT9HV1dYqIiAjcdrvdamxslMdz+ZFjY722zZLzfwts2zaAK4udd9TpEYA267o9go+IiJDP5wvc9vv9V4w7AABo7roNfFxcnIqKiiRJpaWluu222xyeCACAtsNlWZbl9BCX8/236L/44gtZlqWVK1eqX79+To8FAECbcN0GHgAA/HDX7Sl6AADwwxF4AAAMRODhKL/fr8zMTCUnJys9PV1VVVVOjwTcUA4ePKj09HSnx4AN+LszOOq/r1hYWlqqVatWaePGjU6PBdwQNm/erMLCQnXs2NHpUWADjuDhKK5YCDinZ8+eysvLc3oM2ITAw1FXumIhAPslJSVxATGDEXg4iisWAoA9CDwcxRULAcAeHCrBUWPHjtXHH3+slJSUwBULAQA/HleyAwDAQJyiBwDAQAQeAAADEXgAAAxE4AEAMBCBBwDAQAQegCTpiy++0O23367333+/xfVOnDihjIwMSdKnn36qRYsWXXHd/75/27Ztevvtt1tvYAAt4u/gAUiSCgoKlJSUpFdffVVJSUlXXO/06dM6ceKEJOnOO+/UnXfeecV1//v+f/7znxoxYkTrDg3gijiCB6DGxkYVFhZq7ty5+uyzz3T8+HFJUnFxse6//35NmDBBTzzxhOrq6pSdna2ysjItX75c+/btU3p6ug4fPqzx48cHtvfhhx/qySefDNxfXFysv//97/rDH/6gnTt3KiEhQXV1dZKkkydP6r777nPkdQMmI/AAtHv3bnXv3l19+vTRmDFj9Oqrr6q+vl7PPPOMVq9erb/+9a+6/fbbtWPHDi1evFiDBw/W0qVLA48fMGCAwsLC9MUXX0iS3n77bd1///2B+++++2796le/0uzZszVmzBj98pe/1HvvvSdJevPNNzVx4sTQvmDgBkDgAaigoCBwBH7vvfdqx44dOnz4sLp27ao77rhDkjRv3jylp6dfcRsTJ07UO++8o2+++Ub79+/X6NGjr7jugw8+qLfeekvSdzsDBB5ofXwGD9zgvvzySxUVFamsrEx/+tOfZFmWzp8/H/gRoO/V1tY2++W//zV+/Hg9+uijGjBggBITE9WhQ4crrjt8+HD95z//0QcffKAePXqoa9eurfZ6AHyHwAM3uMLCQt11113asmVLYFleXp6Kiop07tw5HTt2TLfeemvg/pEjR6qxsfGS7XTt2lXdunXTyy+/rGefffaS+91ut5qamiRJLpdLDzzwgLKzs7Vw4UKbXhlwY+MUPXCDKygoUFpaWrNlaWlpOnLkiNasWaNnn31WEyZM0LFjx/T444+rX79+qq2t1fz58y/Z1sSJE3Xu3DklJCRcct/dd9+tl156KfDZ+3333advv/1WY8aMseeFATc4fk0OQMj5/X795S9/UUVFhRYvXuz0OICROEUPIORmzpypM2fO6I9//KPTowDG4ggeAAAD8Rk8AAAGIvAAABiIwAMAYCACDwCAgQg8AAAGIvAAABjo/wEkI8+5yKypjwAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 576x396 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.countplot(data=data, x='Activity', palette='CMRmap')\n",
    "activity_percent = data['Activity'].value_counts(normalize=True) * 100\n",
    "print('Activity is 0: {:.2f}%'.format(activity_percent[0]))\n",
    "print('Activity is 1: {:.2f}%'.format(activity_percent[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Выборку можно считать сбалансированной.  \n",
    "Разделим выборку на тренировочную и тестовую в соотношении 80/20."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "X, y = data.drop('Activity', axis=1), data['Activity']\n",
    "X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y, stratify=y, test_size=0.2, random_state=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Оптимизация гиперпараметров модели"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **Логистическая регрессия**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Зафиксируем метрику F1-score, полученную без дополнительной настройки, т.е со значениями гиперпараметров, установленных по умолчанию:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "f1_score на тестовом наборе: 0.78\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\sklearn\\linear_model\\_logistic.py:444: ConvergenceWarning: lbfgs failed to converge (status=1):\n",
      "STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.\n",
      "\n",
      "Increase the number of iterations (max_iter) or scale the data as shown in:\n",
      "    https://scikit-learn.org/stable/modules/preprocessing.html\n",
      "Please also refer to the documentation for alternative solver options:\n",
      "    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression\n",
      "  n_iter_i = _check_optimize_result(\n"
     ]
    }
   ],
   "source": [
    "#Создаем объект класса логистическая регрессия\n",
    "log_reg = linear_model.LogisticRegression(random_state=42, max_iter=50)\n",
    "#Обучаем модель, минимизируя logloss\n",
    "log_reg.fit(X_train, y_train)\n",
    "y_test_pred = log_reg.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **GridSearchCV**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Пробуем подобрать оптимальные параметры с помощью поиска по сетке"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 3.42 s\n",
      "Wall time: 53.5 s\n",
      "f1_score на тестовом наборе: 0.78\n",
      "Наилучшие значения гиперпараметров: {'C': 0.35, 'penalty': 'l1', 'solver': 'saga'}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\sklearn\\linear_model\\_sag.py:350: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "param_grid = [{'penalty': ['l1', 'l2', 'none'] , # тип регуляризации\n",
    "              'solver': ['saga'], # алгоритм оптимизации\n",
    "               'C': [0.01, 0.35, 0.7, 1]}, # уровень силы регуляризации\n",
    "                {'penalty': ['l2', 'none'] ,\n",
    "              'solver': ['lbfgs', 'sag'], \n",
    "               'C': [0.01, 0.35, 0.7, 1]}, \n",
    "               {'penalty': ['l1', 'l2'] ,\n",
    "              'solver': ['liblinear'], \n",
    "               'C': [0.01, 0.35, 0.7, 1]}\n",
    "             ]\n",
    "\n",
    "grid_search_1 = GridSearchCV(\n",
    "    estimator=log_reg, \n",
    "    param_grid=param_grid,\n",
    "    scoring = 'f1', \n",
    "    cv=5, \n",
    "    n_jobs = -1\n",
    ")  \n",
    "%time grid_search_1.fit(X_train, y_train) \n",
    "y_test_pred = grid_search_1.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))\n",
    "print(\"Наилучшие значения гиперпараметров: {}\".format(grid_search_1.best_params_))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "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>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_C</th>\n",
       "      <th>param_penalty</th>\n",
       "      <th>param_solver</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.566247</td>\n",
       "      <td>0.230019</td>\n",
       "      <td>0.020557</td>\n",
       "      <td>0.001625</td>\n",
       "      <td>0.01</td>\n",
       "      <td>l1</td>\n",
       "      <td>saga</td>\n",
       "      <td>{'C': 0.01, 'penalty': 'l1', 'solver': 'saga'}</td>\n",
       "      <td>0.774691</td>\n",
       "      <td>0.773885</td>\n",
       "      <td>0.756672</td>\n",
       "      <td>0.745710</td>\n",
       "      <td>0.737679</td>\n",
       "      <td>0.757727</td>\n",
       "      <td>0.014808</td>\n",
       "      <td>31</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>3.207741</td>\n",
       "      <td>0.084760</td>\n",
       "      <td>0.020197</td>\n",
       "      <td>0.002444</td>\n",
       "      <td>0.01</td>\n",
       "      <td>l2</td>\n",
       "      <td>saga</td>\n",
       "      <td>{'C': 0.01, 'penalty': 'l2', 'solver': 'saga'}</td>\n",
       "      <td>0.815362</td>\n",
       "      <td>0.791988</td>\n",
       "      <td>0.769006</td>\n",
       "      <td>0.791728</td>\n",
       "      <td>0.765766</td>\n",
       "      <td>0.786770</td>\n",
       "      <td>0.018034</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time param_C  \\\n",
       "0       1.566247      0.230019         0.020557        0.001625    0.01   \n",
       "1       3.207741      0.084760         0.020197        0.002444    0.01   \n",
       "\n",
       "  param_penalty param_solver                                          params  \\\n",
       "0            l1         saga  {'C': 0.01, 'penalty': 'l1', 'solver': 'saga'}   \n",
       "1            l2         saga  {'C': 0.01, 'penalty': 'l2', 'solver': 'saga'}   \n",
       "\n",
       "   split0_test_score  split1_test_score  split2_test_score  split3_test_score  \\\n",
       "0           0.774691           0.773885           0.756672           0.745710   \n",
       "1           0.815362           0.791988           0.769006           0.791728   \n",
       "\n",
       "   split4_test_score  mean_test_score  std_test_score  rank_test_score  \n",
       "0           0.737679         0.757727        0.014808               31  \n",
       "1           0.765766         0.786770        0.018034                5  "
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Результаты кросс-валидации\n",
    "result_cv = pd.DataFrame(grid_search_1.cv_results_)\n",
    "result_cv.head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time',\n",
       "       'param_C', 'param_penalty', 'param_solver', 'params',\n",
       "       'split0_test_score', 'split1_test_score', 'split2_test_score',\n",
       "       'split3_test_score', 'split4_test_score', 'mean_test_score',\n",
       "       'std_test_score', 'rank_test_score'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result_cv.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x396 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Отрисуем, как менялась точность при различных гиперпараметрах\n",
    "visual = pd.pivot_table(pd.DataFrame(grid_search_1.cv_results_),\n",
    "                        values='mean_test_score', index='param_C',\n",
    "                        columns=['param_penalty', 'param_solver'])\n",
    "sns.heatmap(visual)\n",
    "plt.title('Тепловая карта зависимости метрики f1 от solver, penalty и С') # подпись графика\n",
    "sns.set(rc={'figure.figsize':(12, 8)}) #задаем размер графика"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Изучим область предполагаемого максимума метрики подробнее с помощью новой сетки."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 3.09 s\n",
      "Wall time: 32.9 s\n",
      "f1_score на тестовом наборе: 0.78\n",
      "Наилучшие значения гиперпараметров: {'C': 0.2, 'penalty': 'l1', 'solver': 'saga'}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\sklearn\\linear_model\\_sag.py:350: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "param_grid = {'penalty': ['l1'] , # тип регуляризации\n",
    "              'solver': ['saga', 'liblinear'], # алгоритм оптимизации\n",
    "               'C': [0.1, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5]}, # уровень силы регуляризации\n",
    "grid_search_2 = GridSearchCV(\n",
    "    estimator=log_reg, \n",
    "    param_grid=param_grid,\n",
    "    scoring = 'f1', \n",
    "    cv=5, \n",
    "    n_jobs = -1\n",
    ")  \n",
    "%time grid_search_2.fit(X_train, y_train) \n",
    "y_test_pred = grid_search_2.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))\n",
    "print(\"Наилучшие значения гиперпараметров: {}\".format(grid_search_2.best_params_))                "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_C</th>\n",
       "      <th>param_penalty</th>\n",
       "      <th>param_solver</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5.046076</td>\n",
       "      <td>0.015036</td>\n",
       "      <td>0.026692</td>\n",
       "      <td>0.001697</td>\n",
       "      <td>0.1</td>\n",
       "      <td>l1</td>\n",
       "      <td>saga</td>\n",
       "      <td>{'C': 0.1, 'penalty': 'l1', 'solver': 'saga'}</td>\n",
       "      <td>0.807692</td>\n",
       "      <td>0.799392</td>\n",
       "      <td>0.78056</td>\n",
       "      <td>0.772660</td>\n",
       "      <td>0.766412</td>\n",
       "      <td>0.785343</td>\n",
       "      <td>0.015742</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.270175</td>\n",
       "      <td>0.020592</td>\n",
       "      <td>0.025792</td>\n",
       "      <td>0.002671</td>\n",
       "      <td>0.1</td>\n",
       "      <td>l1</td>\n",
       "      <td>liblinear</td>\n",
       "      <td>{'C': 0.1, 'penalty': 'l1', 'solver': 'libline...</td>\n",
       "      <td>0.807122</td>\n",
       "      <td>0.804249</td>\n",
       "      <td>0.77037</td>\n",
       "      <td>0.773333</td>\n",
       "      <td>0.766412</td>\n",
       "      <td>0.784297</td>\n",
       "      <td>0.017624</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time param_C  \\\n",
       "0       5.046076      0.015036         0.026692        0.001697     0.1   \n",
       "1       0.270175      0.020592         0.025792        0.002671     0.1   \n",
       "\n",
       "  param_penalty param_solver  \\\n",
       "0            l1         saga   \n",
       "1            l1    liblinear   \n",
       "\n",
       "                                              params  split0_test_score  \\\n",
       "0      {'C': 0.1, 'penalty': 'l1', 'solver': 'saga'}           0.807692   \n",
       "1  {'C': 0.1, 'penalty': 'l1', 'solver': 'libline...           0.807122   \n",
       "\n",
       "   split1_test_score  split2_test_score  split3_test_score  split4_test_score  \\\n",
       "0           0.799392            0.78056           0.772660           0.766412   \n",
       "1           0.804249            0.77037           0.773333           0.766412   \n",
       "\n",
       "   mean_test_score  std_test_score  rank_test_score  \n",
       "0         0.785343        0.015742               15  \n",
       "1         0.784297        0.017624               16  "
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Результаты кросс-валидации\n",
    "result_cv = pd.DataFrame(grid_search_2.cv_results_)\n",
    "result_cv.head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 864x576 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Отрисуем, как менялась точность при различных гиперпараметрах\n",
    "visual = pd.pivot_table(pd.DataFrame(grid_search_2.cv_results_),\n",
    "                        values='mean_test_score', index='param_C',\n",
    "                        columns=['param_penalty', 'param_solver'])\n",
    "sns.heatmap(visual)\n",
    "plt.title('Тепловая карта зависимости метрики f1 от solver, penalty и С') # подпись графика\n",
    "sns.set(rc={'figure.figsize':(12, 8)}) #задаем размер графика"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Значения метрики f1_score на тестовом наборе не изменились, комбинация внешних параметров лучше, чем заданы по умолчанию не найдена."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **RandomizedSearchCV**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Проведем поиск оптимальных параметров среди тех же значений, что и для GridSearchCV. Количество итераций выберем равным числу комбинаций параметров для GridSearchCV. Значительного выигрыша по времени не ожидается, но можно не использовать повторный поиск с более мелкой сеткой."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 766 ms\n",
      "Wall time: 59.7 s\n",
      "f1_score на тестовом наборе: 0.78\n",
      "Наилучшие значения гиперпараметров: {'solver': 'liblinear', 'penalty': 'l1', 'C': 0.34}\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "\n",
    "param_distributions = [{'penalty': ['l1', 'l2', 'none'] , \n",
    "              'solver': ['saga'], \n",
    "               'C': list(np.linspace(0.01, 1, 10, dtype=float))}, \n",
    "                {'penalty': ['l2', 'none'] ,\n",
    "              'solver': ['lbfgs', 'sag'], \n",
    "               'C': list(np.linspace(0.01, 1, 10, dtype=float))}, \n",
    "               {'penalty': ['l1', 'l2'] ,\n",
    "              'solver': ['liblinear'],\n",
    "               'C': list(np.linspace(0.01, 1, 10, dtype=float))}]\n",
    "            \n",
    "random_search = RandomizedSearchCV(\n",
    "    estimator=log_reg, \n",
    "    param_distributions=param_distributions,\n",
    "    scoring = 'f1', \n",
    "    cv=5, \n",
    "    n_iter = 36, \n",
    "    n_jobs = -1\n",
    ")  \n",
    "%time random_search.fit(X_train, y_train) \n",
    "y_test_pred = random_search.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))\n",
    "print(\"Наилучшие значения гиперпараметров: {}\".format(random_search.best_params_))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Метрику f1_score на тестовом наборе улучшить не удалось."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **Hyperopt**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import cross_val_score\n",
    "import hyperopt\n",
    "from hyperopt import hp, fmin, tpe, Trials\n",
    "# fmin - основная функция, она будет минимизировать наш функционал\n",
    "# tpe - алгоритм оптимизации\n",
    "# hp - включает набор методов для объявления пространства поиска гиперпараметров\n",
    "# trails - используется для логирования результатов"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Настроим оптимизацию гиперпараметров для алгоритма логистической регрессии."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# зададим пространство поиска гиперпараметров\n",
    "\n",
    "space = hp.choice('parameter_combinations', [\n",
    "        {'solver': 'saga',\n",
    "         'penalty': 'l1',\n",
    "         'C': hp.uniform('C_saga_l1', 0.01, 1)\n",
    "        },\n",
    "        {'solver': 'saga',\n",
    "         'penalty': 'l2',\n",
    "         'C': hp.uniform('C_saga_l2', 0.01, 1)\n",
    "        }, \n",
    "        {'solver': 'saga',\n",
    "         'penalty': 'none',\n",
    "         'C': hp.uniform('C_saga_none', 0.01, 1)\n",
    "        }, \n",
    "         \n",
    "         \n",
    "        \n",
    "        {'solver': 'lbfgs',\n",
    "         'penalty':'l2',\n",
    "         'C': hp.uniform('C_lbfgs_l2', 0.01, 1)\n",
    "        },\n",
    "        {'solver':'lbfgs',\n",
    "         'penalty': 'none',\n",
    "         'C': hp.uniform('C_lbfgs_none', 0.01, 1)\n",
    "        },\n",
    "        \n",
    "        \n",
    "        {'solver':'sag',\n",
    "         'penalty': 'l2',\n",
    "         'C': hp.uniform('C_sag_l2', 0.01, 1)\n",
    "        },\n",
    "        {'solver': 'sag',\n",
    "         'penalty': 'none',\n",
    "         'C': hp.uniform('C_sag_none', 0.01, 1)\n",
    "        },\n",
    "        \n",
    "        \n",
    "        {'solver': 'liblinear',\n",
    "        'penalty': 'l1',\n",
    "        'C': hp.uniform('C_liblinear_l1', 0.01, 1)\n",
    "        },\n",
    "        {'solver': 'liblinear',\n",
    "        'penalty': 'l2',\n",
    "        'C': hp.uniform('C_liblinear_l2', 0.01, 1)\n",
    "        },\n",
    "        ]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "random_state = 42\n",
    "\n",
    "def hyperopt_lr(params, cv=5, X=X_train, y=y_train, random_state=random_state):\n",
    "    params = {'solver': params['solver'], \n",
    "              'penalty': params['penalty'], \n",
    "              'C': params['C']\n",
    "             }\n",
    "      \n",
    "    model = linear_model.LogisticRegression(**params, random_state=random_state, max_iter=50)\n",
    "\n",
    "    # Обучаем модель с помощью кросс-валидации\n",
    "    score = cross_val_score(model, X, y, cv=cv, scoring='f1', n_jobs=-1).mean()\n",
    "\n",
    "    return -score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 0 ns\n",
      "Wall time: 0 ns\n",
      "100%|██████████| 20/20 [00:55<00:00,  2.79s/trial, best loss: -0.787628046912189]\n",
      "Наилучшие значения гиперпараметров {'C_saga_l2': 0.08292441519601887, 'parameter_combinations': 1}\n"
     ]
    }
   ],
   "source": [
    "# начинаем подбор гиперпараметров\n",
    "%time \n",
    " \n",
    "trials = Trials() # используется для логирования результатов\n",
    "\n",
    "best=fmin(hyperopt_lr, # наша функция \n",
    "          space=space, # пространство гиперпараметров\n",
    "          algo=tpe.suggest, # алгоритм оптимизации, установлен по умолчанию, задавать необязательно\n",
    "          max_evals=20, # максимальное количество итераций\n",
    "          trials=trials, # логирование результатов\n",
    "          rstate=np.random.default_rng(random_state)# фиксируем для повторяемости результата\n",
    "          )\n",
    "print(\"Наилучшие значения гиперпараметров {}\".format(best))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "f1_score на обучающем наборе: 0.85\n",
      "f1_score на тестовом наборе: 0.79\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\sklearn\\linear_model\\_sag.py:350: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "# рассчитаем точность для тестовой выборки\n",
    "model_HO_best = linear_model.LogisticRegression(\n",
    "    solver = 'saga', \n",
    "    penalty = 'l2',\n",
    "    C = 0.08292441519601887,\n",
    "    random_state=random_state, \n",
    "    max_iter=50\n",
    "    )\n",
    "model_HO_best.fit(X_train, y_train)\n",
    "y_train_pred = model_HO_best.predict(X_train)\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "y_test_pred = model_HO_best.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "С помощью Hyperopt метрику f1_score на тестовом наборе удалось улучшить."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> Optuna"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "import optuna"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "random_state = 42\n",
    "\n",
    "def optuna_lr(trial):\n",
    "    # пространство гиперпараметров\n",
    "    solver = trial.suggest_categorical('solver', ['saga', 'lbfgs', 'sag', 'liblinear'])\n",
    "    if solver == 'saga':\n",
    "        penalty = trial.suggest_categorical('penalty', ['l1', 'l2', 'none'])\n",
    "    else:\n",
    "        penalty = 'l2'\n",
    "        \n",
    "    C = trial.suggest_uniform('C', 0.01, 1)\n",
    "    \n",
    "    model = linear_model.LogisticRegression(\n",
    "        solver=solver,\n",
    "        penalty=penalty,\n",
    "        C=C,\n",
    "        random_state=random_state, \n",
    "        max_iter=50\n",
    "    )    \n",
    "    \n",
    "    # Обучаем модель с помощью кросс-валидации    \n",
    "    score = cross_val_score(\n",
    "        model, X_train, y_train, cv=5, scoring='f1', n_jobs=-1).mean()\n",
    "    \n",
    "    return score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[32m[I 2022-08-26 16:19:12,948]\u001b[0m A new study created in memory with name: LogisticRegression\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:14,064]\u001b[0m Trial 0 finished with value: 0.7834178997770478 and parameters: {'solver': 'lbfgs', 'C': 0.16445845403801215}. Best is trial 0 with value: 0.7834178997770478.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:17,170]\u001b[0m Trial 1 finished with value: 0.7793051656527762 and parameters: {'solver': 'sag', 'C': 0.710991852018085}. Best is trial 0 with value: 0.7834178997770478.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:18,381]\u001b[0m Trial 2 finished with value: 0.7814287466066778 and parameters: {'solver': 'lbfgs', 'C': 0.19000671753502962}. Best is trial 0 with value: 0.7834178997770478.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:21,697]\u001b[0m Trial 3 finished with value: 0.7784773740869497 and parameters: {'solver': 'sag', 'C': 0.2983168487960615}. Best is trial 0 with value: 0.7834178997770478.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:25,567]\u001b[0m Trial 4 finished with value: 0.7796530883762236 and parameters: {'solver': 'saga', 'penalty': 'l2', 'C': 0.5190920940294755}. Best is trial 0 with value: 0.7834178997770478.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:29,193]\u001b[0m Trial 5 finished with value: 0.7886791388489908 and parameters: {'solver': 'sag', 'C': 0.07440107705542671}. Best is trial 5 with value: 0.7886791388489908.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:30,461]\u001b[0m Trial 6 finished with value: 0.7854048992368717 and parameters: {'solver': 'lbfgs', 'C': 0.10669539286632003}. Best is trial 5 with value: 0.7886791388489908.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:34,294]\u001b[0m Trial 7 finished with value: 0.7791666903993274 and parameters: {'solver': 'saga', 'penalty': 'l2', 'C': 0.6658970615104421}. Best is trial 5 with value: 0.7886791388489908.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:37,910]\u001b[0m Trial 8 finished with value: 0.7795521582233507 and parameters: {'solver': 'sag', 'C': 0.9698887814869129}. Best is trial 5 with value: 0.7886791388489908.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:39,030]\u001b[0m Trial 9 finished with value: 0.7713060285701678 and parameters: {'solver': 'lbfgs', 'C': 0.9226554926728857}. Best is trial 5 with value: 0.7886791388489908.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:39,790]\u001b[0m Trial 10 finished with value: 0.790609157811635 and parameters: {'solver': 'liblinear', 'C': 0.015937211126258566}. Best is trial 10 with value: 0.790609157811635.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:40,555]\u001b[0m Trial 11 finished with value: 0.7908379790380127 and parameters: {'solver': 'liblinear', 'C': 0.015244149605851878}. Best is trial 11 with value: 0.7908379790380127.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:41,451]\u001b[0m Trial 12 finished with value: 0.7772516768896305 and parameters: {'solver': 'liblinear', 'C': 0.35042886867188444}. Best is trial 11 with value: 0.7908379790380127.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:42,117]\u001b[0m Trial 13 finished with value: 0.7911749231867716 and parameters: {'solver': 'liblinear', 'C': 0.020783734592852925}. Best is trial 13 with value: 0.7911749231867716.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:43,107]\u001b[0m Trial 14 finished with value: 0.7773860802149837 and parameters: {'solver': 'liblinear', 'C': 0.341353575497707}. Best is trial 13 with value: 0.7911749231867716.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:43,992]\u001b[0m Trial 15 finished with value: 0.7791966528547652 and parameters: {'solver': 'liblinear', 'C': 0.2531652323662312}. Best is trial 13 with value: 0.7911749231867716.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:45,037]\u001b[0m Trial 16 finished with value: 0.7786966068976742 and parameters: {'solver': 'liblinear', 'C': 0.4400127090388472}. Best is trial 13 with value: 0.7911749231867716.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:45,799]\u001b[0m Trial 17 finished with value: 0.7925778470843667 and parameters: {'solver': 'liblinear', 'C': 0.029834922551910417}. Best is trial 17 with value: 0.7925778470843667.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:46,789]\u001b[0m Trial 18 finished with value: 0.7813216668290741 and parameters: {'solver': 'liblinear', 'C': 0.5334038771422205}. Best is trial 17 with value: 0.7925778470843667.\u001b[0m\n",
      "\u001b[32m[I 2022-08-26 16:19:47,621]\u001b[0m Trial 19 finished with value: 0.7824588808976449 and parameters: {'solver': 'liblinear', 'C': 0.15265035504437724}. Best is trial 17 with value: 0.7925778470843667.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "# Поиск оптимальных гиперпараметров\n",
    "sampler = optuna.samplers.TPESampler(seed=random_state)\n",
    "study_lr = optuna.create_study(\n",
    "    sampler=sampler, study_name='LogisticRegression', direction='maximize')\n",
    "study_lr.optimize(optuna_lr, n_trials=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\sklearn\\linear_model\\_logistic.py:1153: UserWarning: 'n_jobs' > 1 does not have any effect when 'solver' is set to 'liblinear'. Got 'n_jobs' = 8.\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "# Модель с лучшими параметрами\n",
    "optuna_lr = linear_model.LogisticRegression(\n",
    "    **study_lr.best_params,  n_jobs=-1,\n",
    "    random_state=random_state, max_iter=50\n",
    ")\n",
    "optuna_lr.fit(X_train, y_train)\n",
    "y_train_pred = optuna_lr.predict(X_train)\n",
    "y_test_pred = optuna_lr.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Наилучшие значения гиперпараметров: {'solver': 'liblinear', 'C': 0.029834922551910417}\n",
      "f1_score на обучающем наборе: 0.84\n",
      "f1_score на тестовом наборе: 0.78\n"
     ]
    }
   ],
   "source": [
    "print(f'Наилучшие значения гиперпараметров: {study_lr.best_params}')\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "С помощью Optuna метрику f1_score на тестовом наборе улучшить не удалось."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **Случайный лес**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Проделаем аналогичные действия для RandomForestClassifier().\n",
    "Сначала посчитаем модель с параметрами по умолчанию и оценим метрику:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "f1_score на обучающем наборе: 1.00\n",
      "f1_score на тестовом наборе: 0.80\n"
     ]
    }
   ],
   "source": [
    "#Создаем объект класса случайный лес\n",
    "rf = ensemble.RandomForestClassifier(random_state=42)\n",
    "\n",
    "#Обучаем модель\n",
    "rf.fit(X_train, y_train)\n",
    "#Выводим значения метрики \n",
    "y_train_pred = rf.predict(X_train)\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "y_test_pred = rf.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **GridSearchCV**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 4.67 s\n",
      "Wall time: 3min 50s\n",
      "f1_score на обучающем наборе: 0.93\n",
      "f1_score на тестовом наборе: 0.80\n",
      "Наилучшие значения гиперпараметров: {'max_depth': 30, 'min_samples_leaf': 6, 'n_estimators': 300}\n"
     ]
    }
   ],
   "source": [
    "# Сетка параметров\n",
    "param_grid = {'n_estimators': list(range(100, 400, 100)),\n",
    "              'min_samples_leaf': list(range(4, 10, 2)),\n",
    "              'max_depth': list(np.linspace(5, 50, 10, dtype=int))\n",
    "              }\n",
    "# Поиск по сетке параметров с кросс-валидацией             \n",
    "grid_search_forest = GridSearchCV(\n",
    "    estimator=ensemble.RandomForestClassifier(random_state=42), \n",
    "    param_grid=param_grid, \n",
    "    cv=5, \n",
    "    n_jobs = -1\n",
    ")  \n",
    "%time grid_search_forest.fit(X_train, y_train) \n",
    "y_train_pred = grid_search_forest.predict(X_train)\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "y_test_pred = grid_search_forest.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))\n",
    "print(\"Наилучшие значения гиперпараметров: {}\".format(grid_search_forest.best_params_))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Значения метрики f1_score на тестовом наборе не изменились, комбинация внешних параметров лучше, чем заданы по умолчанию не найдена."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **RandomizedSearchCV**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Проведем поиск оптимальных параметров среди тех же значений, что и для GridSearchCV."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 3.06 s\n",
      "Wall time: 48.2 s\n",
      "f1_score на обучающем наборе: 0.93\n",
      "f1_score на тестовом наборе: 0.80\n",
      "Наилучшие значения гиперпараметров: {'n_estimators': 300, 'min_samples_leaf': 6, 'max_depth': 40}\n"
     ]
    }
   ],
   "source": [
    "# Сетка параметров\n",
    "param_distributions = {'n_estimators': list(range(100, 400, 100)),\n",
    "              'min_samples_leaf': list(range(4, 10, 2)),\n",
    "              'max_depth': list(np.linspace(5, 50, 10, dtype=int))\n",
    "              }\n",
    "\n",
    "# Поиск по сетке параметров с кросс-валидацией            \n",
    "random_search_forest = RandomizedSearchCV(\n",
    "    estimator=ensemble.RandomForestClassifier(random_state=42), \n",
    "    param_distributions=param_distributions, \n",
    "    cv=5,\n",
    "    n_iter = 20, \n",
    "    n_jobs = -1\n",
    ")  \n",
    "%time random_search_forest.fit(X_train, y_train) \n",
    "y_train_pred = random_search_forest.predict(X_train)\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "y_test_pred = random_search_forest.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))\n",
    "print(\"Наилучшие значения гиперпараметров: {}\".format(random_search_forest.best_params_))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Значения метрики f1_score на тестовом наборе не изменились, комбинация внешних параметров лучше, чем заданы по умолчанию не найдена."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> **Hyperopt**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Настроим оптимизацию гиперпараметров для алгоритма случайного леса."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# зададим пространство поиска гиперпараметров\n",
    "space={'n_estimators': hp.quniform('n_estimators', 100, 400, 10),\n",
    "       'max_depth' : hp.quniform('max_depth', 5, 50, 10),\n",
    "       'min_samples_leaf': hp.quniform('min_samples_leaf', 4, 10, 1)\n",
    "      }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "# зафиксируем random_state\n",
    "random_state = 42\n",
    "def hyperopt_rf(params, cv=5, X=X_train, y=y_train, random_state=random_state):\n",
    "    # функция получает комбинацию гиперпараметров в \"params\"\n",
    "    params = {'n_estimators': int(params['n_estimators']), \n",
    "              'max_depth': int(params['max_depth']), \n",
    "             'min_samples_leaf': int(params['min_samples_leaf'])\n",
    "              }\n",
    "  \n",
    "    # используем эту комбинацию для построения модели\n",
    "    model = ensemble.RandomForestClassifier(**params, \n",
    "                                            random_state=random_state,\n",
    "                                            n_jobs=-1\n",
    "                                            )\n",
    "\n",
    "    \n",
    "    # Обучаем модель с помощью кросс-валидации    \n",
    "    score = cross_val_score(model, X, y, cv=cv, scoring='f1', n_jobs=-1).mean()\n",
    "    # метрику необходимо минимизировать, поэтому ставим знак минус\n",
    "    return -score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 0 ns\n",
      "Wall time: 0 ns\n",
      "100%|██████████| 20/20 [01:03<00:00,  3.20s/trial, best loss: -0.8124706654030968]\n",
      "Наилучшие значения гиперпараметров {'max_depth': 20.0, 'min_samples_leaf': 4.0, 'n_estimators': 110.0}\n"
     ]
    }
   ],
   "source": [
    "# начинаем подбор гиперпараметров\n",
    "%time \n",
    " \n",
    "trials = Trials() # используется для логирования результатов\n",
    "\n",
    "best=fmin(hyperopt_rf, # наша функция \n",
    "          space=space, # пространство гиперпараметров\n",
    "          algo=tpe.suggest, # алгоритм оптимизации, установлен по умолчанию, задавать необязательно\n",
    "          max_evals=20, # максимальное количество итераций\n",
    "          trials=trials, # логирование результатов\n",
    "          rstate=np.random.default_rng(random_state)# фиксируем для повторяемости результата\n",
    "          # rstate=np.random.default_rng(random_state)\n",
    "         )\n",
    "print(\"Наилучшие значения гиперпараметров {}\".format(best))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "f1_score на обучающем наборе: 0.96\n",
      "f1_score на тестовом наборе: 0.80\n"
     ]
    }
   ],
   "source": [
    "# рассчитаем точность для тестовой выборки\n",
    "model = ensemble.RandomForestClassifier(\n",
    "    random_state=random_state, \n",
    "    n_estimators=int(best['n_estimators']),\n",
    "    max_depth=int(best['max_depth']),\n",
    "    min_samples_leaf=int(best['min_samples_leaf'])\n",
    ")\n",
    "model.fit(X_train, y_train)\n",
    "y_train_pred = model.predict(X_train)\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "y_test_pred = model.predict(X_test)\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Значения метрики f1_score на тестовом наборе не изменились, комбинация внешних параметров лучше, чем заданы по умолчанию не найдена."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <center> Optuna"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "random_state = 42\n",
    "\n",
    "def optuna_rf(trial):\n",
    "  # задаем пространства поиска гиперпараметров\n",
    "    n_estimators = trial.suggest_int('n_estimators', 100, 400, 10)\n",
    "    max_depth = trial.suggest_int('max_depth', 5, 50, 10)\n",
    "    min_samples_leaf = trial.suggest_int('min_samples_leaf', 4, 10, 1)\n",
    "\n",
    "  # создаем модель\n",
    "    model = ensemble.RandomForestClassifier(n_estimators=n_estimators,\n",
    "                                          max_depth=max_depth,\n",
    "                                          min_samples_leaf=min_samples_leaf,\n",
    "                                          n_jobs=-1,\n",
    "                                          random_state=random_state)\n",
    "  # Обучаем модель с помощью кросс-валидации    \n",
    "    score = cross_val_score(\n",
    "        model, X_train, y_train, cv=5, scoring='f1', n_jobs=-1).mean()\n",
    "    return score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[32m[I 2022-08-26 16:25:34,197]\u001b[0m A new study created in memory with name: RandomForestClassifier\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:37,004]\u001b[0m Trial 0 finished with value: 0.808715793761546 and parameters: {'n_estimators': 210, 'max_depth': 45, 'min_samples_leaf': 9}. Best is trial 0 with value: 0.808715793761546.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:39,430]\u001b[0m Trial 1 finished with value: 0.7661087259188779 and parameters: {'n_estimators': 280, 'max_depth': 5, 'min_samples_leaf': 5}. Best is trial 0 with value: 0.808715793761546.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:41,146]\u001b[0m Trial 2 finished with value: 0.8042089648798179 and parameters: {'n_estimators': 110, 'max_depth': 45, 'min_samples_leaf': 8}. Best is trial 0 with value: 0.808715793761546.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:43,727]\u001b[0m Trial 3 finished with value: 0.7665822027859597 and parameters: {'n_estimators': 310, 'max_depth': 5, 'min_samples_leaf': 10}. Best is trial 0 with value: 0.808715793761546.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:48,683]\u001b[0m Trial 4 finished with value: 0.8105875347794583 and parameters: {'n_estimators': 350, 'max_depth': 15, 'min_samples_leaf': 5}. Best is trial 4 with value: 0.8105875347794583.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:51,071]\u001b[0m Trial 5 finished with value: 0.8064795592672265 and parameters: {'n_estimators': 150, 'max_depth': 15, 'min_samples_leaf': 7}. Best is trial 4 with value: 0.8105875347794583.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:54,299]\u001b[0m Trial 6 finished with value: 0.8076301172996345 and parameters: {'n_estimators': 230, 'max_depth': 15, 'min_samples_leaf': 8}. Best is trial 4 with value: 0.8105875347794583.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:25:56,613]\u001b[0m Trial 7 finished with value: 0.8108034426229004 and parameters: {'n_estimators': 140, 'max_depth': 15, 'min_samples_leaf': 6}. Best is trial 7 with value: 0.8108034426229004.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:00,337]\u001b[0m Trial 8 finished with value: 0.8107918064263029 and parameters: {'n_estimators': 240, 'max_depth': 35, 'min_samples_leaf': 5}. Best is trial 7 with value: 0.8108034426229004.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:04,311]\u001b[0m Trial 9 finished with value: 0.8128067740254483 and parameters: {'n_estimators': 250, 'max_depth': 25, 'min_samples_leaf': 4}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:10,150]\u001b[0m Trial 10 finished with value: 0.8114580340888675 and parameters: {'n_estimators': 390, 'max_depth': 35, 'min_samples_leaf': 4}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:15,608]\u001b[0m Trial 11 finished with value: 0.8117136027108843 and parameters: {'n_estimators': 380, 'max_depth': 35, 'min_samples_leaf': 4}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:21,385]\u001b[0m Trial 12 finished with value: 0.8108799537972627 and parameters: {'n_estimators': 400, 'max_depth': 35, 'min_samples_leaf': 4}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:26,016]\u001b[0m Trial 13 finished with value: 0.8120837820219862 and parameters: {'n_estimators': 320, 'max_depth': 25, 'min_samples_leaf': 4}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:30,268]\u001b[0m Trial 14 finished with value: 0.8123469047828469 and parameters: {'n_estimators': 310, 'max_depth': 25, 'min_samples_leaf': 6}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:33,134]\u001b[0m Trial 15 finished with value: 0.8125103706538708 and parameters: {'n_estimators': 190, 'max_depth': 25, 'min_samples_leaf': 6}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:35,999]\u001b[0m Trial 16 finished with value: 0.8125103706538708 and parameters: {'n_estimators': 190, 'max_depth': 25, 'min_samples_leaf': 6}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:38,483]\u001b[0m Trial 17 finished with value: 0.8054853371361057 and parameters: {'n_estimators': 170, 'max_depth': 25, 'min_samples_leaf': 7}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:42,547]\u001b[0m Trial 18 finished with value: 0.8118792777526114 and parameters: {'n_estimators': 270, 'max_depth': 15, 'min_samples_leaf': 5}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n",
      "c:\\Users\\Rui\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\optuna\\distributions.py:560: UserWarning: The distribution is specified by [5, 50] and step=10, but the range is not divisible by `step`. It will be replaced by [5, 45].\n",
      "  warnings.warn(\n",
      "\u001b[32m[I 2022-08-26 16:26:45,497]\u001b[0m Trial 19 finished with value: 0.8067635919619219 and parameters: {'n_estimators': 200, 'max_depth': 35, 'min_samples_leaf': 7}. Best is trial 9 with value: 0.8128067740254483.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "# Поиск оптимальных гиперпараметров\n",
    "sampler = optuna.samplers.TPESampler(seed=random_state)\n",
    "study_rf = optuna.create_study(\n",
    "    sampler=sampler, study_name='RandomForestClassifier', direction='maximize')\n",
    "study_rf.optimize(optuna_rf, n_trials=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Модель с лучшими параметрами  \n",
    "optuna_rf = ensemble.RandomForestClassifier(\n",
    "    **study_rf.best_params, class_weight='balanced', n_jobs=-1,\n",
    "    random_state=random_state\n",
    ")\n",
    "optuna_rf.fit(X_train, y_train)\n",
    "y_train_pred = optuna_rf.predict(X_train)\n",
    "y_test_pred = optuna_rf.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Наилучшие значения гиперпараметров: {'n_estimators': 250, 'max_depth': 25, 'min_samples_leaf': 4}\n",
      "f1_score на обучающем наборе: 0.96\n",
      "f1_score на тестовом наборе: 0.80\n"
     ]
    }
   ],
   "source": [
    "print(f'Наилучшие значения гиперпараметров: {study_rf.best_params}')\n",
    "print('f1_score на обучающем наборе: {:.2f}'.format(metrics.f1_score(y_train, y_train_pred)))\n",
    "print('f1_score на тестовом наборе: {:.2f}'.format(metrics.f1_score(y_test, y_test_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Значения метрики f1_score на тестовом наборе не изменились, комбинация внешних параметров лучше, чем заданы по умолчанию не найдена."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9.12 64-bit",
   "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": "3.9.12"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "93e5499867a482d1f69126f65754e04e21327fbe16c28c259dff18a59fe0a8c0"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}