ML_STN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 데이터 전처리\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util import preprocessing_simple_train\n",
    "from util import month_to_day, make_mean\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "mean_df = preprocessing_simple_train(\"fast\")\n",
    "tmp = month_to_day(mean_df['rainfall_train.ef_month']) + mean_df['rainfall_train.ef_day'] \n",
    "\n",
    "mean_df['daysin'] = np.sin(tmp)\n",
    "mean_df['daycos'] = np.cos(tmp)\n",
    "mean_df['day'] = tmp\n",
    "mean_df = make_mean(mean_df,15)\n",
    "r_df = pd.read_csv('./data/rainfall_train.csv')[['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour','rainfall_train.class_interval']]\n",
    "r_df = r_df[r_df['rainfall_train.class_interval'] != -999]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [],
   "source": [
    "_v=[f\"rainfall_train.v{format(i,'02')}\" for i in range(1,10)]\n",
    "mean_df[_v] = mean_df[_v].apply(lambda x: np.log(x+0.1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 테스트/ 검증 데이터 결정\n",
    "####  테스트데이터"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util import make_day2class, make_day2freqclass, make_day2vv\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "import pandas as pd \n",
    "\n",
    "_year = \"A\"\n",
    "\n",
    "train_df = mean_df[mean_df['rainfall_train.ef_year'] != _year].reset_index(drop=True)\n",
    "test_df = mean_df[mean_df['rainfall_train.ef_year'] == _year].reset_index(drop=True)\n",
    "rtrain_df = r_df[r_df['rainfall_train.ef_year'] != _year].reset_index(drop=True)\n",
    "rtest_df = r_df[r_df['rainfall_train.ef_year'] == _year].reset_index(drop=True)\n",
    "\n",
    "day2class = make_day2class(train_df)\n",
    "day2vv = make_day2vv(train_df)\n",
    "day2freqclass = make_day2freqclass(train_df)\n",
    "\n",
    "day2class=day2class.rename(columns={\"rainfall_train.class_interval\":\"mean_class\"})\n",
    "day2vv=day2vv.rename(columns={\"rainfall_train.vv\":\"mean_vv\"})\n",
    "day2class['mean_class'] = StandardScaler().fit_transform(day2class[['mean_class']])\n",
    "day2freqclass=day2freqclass.rename(columns={\"rainfall_train.class_interval\":\"freq_class\"})\n",
    "train_df = pd.merge(train_df,day2class,how=\"left\",on=['day']).copy()\n",
    "train_df = pd.merge(train_df,day2vv,how=\"left\",on=['day']).copy()\n",
    "train_df = pd.merge(train_df,day2freqclass,how=\"left\",on=['day']).copy()\n",
    "train_df['freq_class'] = train_df['freq_class'].fillna(0)\n",
    "test_df = pd.merge(test_df,day2class,how=\"left\",on=['day']).copy()\n",
    "test_df = pd.merge(test_df,day2vv,how=\"left\",on=['day']).copy()\n",
    "test_df = pd.merge(test_df,day2freqclass,how=\"left\",on=['day']).copy()\n",
    "test_df['freq_class'] = test_df['freq_class'].fillna(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 제출용 데이터"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util import make_day2class\n",
    "import pandas as pd \n",
    "day2class = make_day2class(mean_df)\n",
    "day2vv = make_day2vv(mean_df)\n",
    "day2class=day2class.rename(columns={\"rainfall_train.class_interval\":\"mean_class\"})\n",
    "day2vv=day2vv.rename(columns={\"rainfall_train.vv\":\"mean_vv\"})\n",
    "day2class['mean_class'] = StandardScaler().fit_transform(day2class[['mean_class']])\n",
    "\n",
    "day2freqclass = make_day2freqclass(train_df)\n",
    "day2freqclass=day2freqclass.rename(columns={\"rainfall_train.class_interval\":\"freq_class\"})\n",
    "train_df = pd.merge(mean_df,day2class,how=\"left\",on=['day']).copy()\n",
    "train_df = pd.merge(train_df,day2vv,how=\"left\",on=['day']).copy()\n",
    "\n",
    "train_df = pd.merge(train_df,day2freqclass,how=\"left\",on=['day']).copy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 무강수/강수 데이터 만들기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = train_df.drop(columns=[\"rainfall_train.stn4contest\",\"rainfall_train.ef_year\",\"rainfall_train.ef_month\",\"rainfall_train.ef_day\",\"rainfall_train.vv\",\"rainfall_train.class_interval\",'v_max','v_median','day','freq_class','rainfall_train.dh_x','mean_vv'])\n",
    "X['rainfall_train.ef_hour'] /= 24.0\n",
    "Y = train_df[\"rainfall_train.class_interval\"]\n",
    "X_V = test_df.drop(columns=[\"rainfall_train.stn4contest\",\"rainfall_train.ef_year\",\"rainfall_train.ef_month\",\"rainfall_train.ef_day\",\"rainfall_train.vv\",\"rainfall_train.class_interval\",'v_max','v_median','day','freq_class','rainfall_train.dh_x','mean_vv'])\n",
    "X_V['rainfall_train.ef_hour'] /= 24.0\n",
    "Y_V = test_df[\"rainfall_train.class_interval\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 강수/무강수 분류"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "무강수/강수 분류 accuracy_score: 0.9080343999380182\n"
     ]
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.ensemble import ExtraTreesClassifier\n",
    "from sklearn.metrics import accuracy_score\n",
    "m_b = ExtraTreesClassifier(n_estimators= 175, criterion = 'log_loss', max_depth = 12,max_leaf_nodes = 45, random_state=42) \n",
    "m_b.fit(X,Y==0)\n",
    "Y_prev = m_b.predict(X_V)\n",
    "print(f\"무강수/강수 분류 accuracy_score: {accuracy_score(Y_V==0,Y_prev)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 강수 무강수 데이터"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "metadata": {},
   "outputs": [],
   "source": [
    "ctrain_df = train_df[train_df['rainfall_train.class_interval'] != 0]\n",
    "ctest_df = test_df[test_df['rainfall_train.class_interval'] != 0]\n",
    "\n",
    "\n",
    "CX = ctrain_df.drop(columns=[\"rainfall_train.stn4contest\",\"rainfall_train.ef_year\",\"rainfall_train.ef_month\",\"rainfall_train.ef_day\",\"rainfall_train.vv\",\"rainfall_train.class_interval\",'v_max','day','daysin','daycos','mean_vmax','v_median','rainfall_train.v00','freq_class','rainfall_train.dh_x','mean_vv'])\n",
    "CX['rainfall_train.ef_hour'] /= 24.0\n",
    "CY = ctrain_df[\"rainfall_train.class_interval\"]\n",
    "CX_V = ctest_df.drop(columns=[\"rainfall_train.stn4contest\",\"rainfall_train.ef_year\",\"rainfall_train.ef_month\",\"rainfall_train.ef_day\",\"rainfall_train.vv\",\"rainfall_train.class_interval\",'v_max','day','daysin','daycos','mean_vmax','v_median','rainfall_train.v00','freq_class','rainfall_train.dh_x','mean_vv'])\n",
    "CX_V['rainfall_train.ef_hour'] /= 24.0\n",
    "CY_V = ctest_df[\"rainfall_train.class_interval\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 173,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train_DH 최소 : 0.5583180377106625\n",
      "valid_DH 최소 : 0.49503239740820737\n"
     ]
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier\n",
    "from util import compute_csi\n",
    "m_c = RandomForestClassifier(n_estimators=150,random_state=42,max_depth=10,criterion='gini')\n",
    "m_c.fit(CX,CY)\n",
    "CY_prev =m_c.predict(CX_V)\n",
    "print(f\"train_DH 최소 : {compute_csi(CY,m_c.predict(CX))}\")\n",
    "print(f\"valid_DH 최소 : {compute_csi(CY_V,CY_prev)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DH 최소 : 0.49503239740820737\n"
     ]
    }
   ],
   "source": [
    "CY_prev =np.where(ctest_df['rainfall_train.dh_x'] >200,ctest_df['freq_class'],CY_prev)\n",
    "print(f\"DH 최소 : {compute_csi(CY_V,CY_prev)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 전체 CSI 점수 검증\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_20821/503635885.py:2: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  ctest_df['cprev'] = CY_prev\n"
     ]
    }
   ],
   "source": [
    "test_df['prev'] = 0\n",
    "ctest_df['cprev'] = CY_prev\n",
    "rtest_df = pd.merge(rtest_df, test_df.loc[Y_prev,['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour','prev']], how=\"left\" , on=['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour'])\n",
    "rtest_df = pd.merge(rtest_df, ctest_df[['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour','cprev']], how=\"left\" , on=['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour'])\n",
    "rtest_df['prev'] = rtest_df['cprev'].combine_first(rtest_df['prev'])\n",
    "rtest_df['prev'] = rtest_df['prev'].fillna(1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "최종 CSI : 0.421360918418113\n"
     ]
    }
   ],
   "source": [
    "print(f\"최종 CSI : {compute_csi(rtest_df['rainfall_train.class_interval'], rtest_df['prev'])}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['rainfall_train.ef_hour' 'rainfall_train.v01' 'rainfall_train.v02'\n",
      " 'rainfall_train.v03' 'rainfall_train.v04' 'rainfall_train.v05'\n",
      " 'rainfall_train.v06' 'rainfall_train.v07' 'rainfall_train.v08'\n",
      " 'rainfall_train.v09' 'mean_vmedian' 'mean_class']\n",
      "[0.06942132 0.06470551 0.0860957  0.07893956 0.08986504 0.110723\n",
      " 0.11794296 0.08909581 0.04169541 0.02019479 0.12744075 0.10388015]\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "sns.barplot(x=CX_V.columns.values,y=m_c.feature_importances_)\n",
    "plt.xticks(rotation=30)\n",
    "print(CX_V.columns.values)\n",
    "print(m_c.feature_importances_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 제출"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "metadata": {},
   "outputs": [],
   "source": [
    "from util import preprocessing_simple_test\n",
    "\n",
    "TEST =preprocessing_simple_test()\n",
    "TEST['rainfall_test.ef_hour'] /= 24.0\n",
    "TEST['day'] = month_to_day(TEST[\"rainfall_test.ef_month\"]) + TEST[\"rainfall_test.ef_day\"]\n",
    "TEST['daysin'] = np.sin(np.pi*2*(TEST['day']/365.0))\n",
    "TEST['daycos'] = np.cos(np.pi*2*(TEST['day']/365.0))\n",
    "\n",
    "TEST.columns = TEST.columns.str.replace(\"rainfall_test\",\"rainfall_train\")\n",
    "TEST = make_mean(TEST)\n",
    "_v=[f\"rainfall_train.v{format(i,'02')}\" for i in range(1,10)]\n",
    "TEST[_v] = TEST[_v].apply(lambda x: np.log(x+0.1))\n",
    "\n",
    "TEST = pd.merge(TEST,day2class,how=\"left\",on=['day'])\n",
    "TEST = pd.merge(TEST,day2vv,how=\"left\",on=['day'])\n",
    "\n",
    "X=TEST.drop(columns=['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','v_max','v_median','day','rainfall_train.class_interval','rainfall_train.dh_x','mean_vv'])\n",
    "b_pred = m_b.predict(X)\n",
    "CTEST = TEST[~b_pred].reset_index(drop=True)\n",
    "X= X[~b_pred].drop(columns=['mean_vmax','rainfall_train.v00','daycos','daysin'])\n",
    "CTEST['class'] = m_c.predict(X)\n",
    "TEST.loc[b_pred, 'rainfall_train.class_interval'] = 0\n",
    "tmp_df = pd.merge(TEST,CTEST[['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour','class']],how='left',on=['rainfall_train.stn4contest','rainfall_train.ef_year','rainfall_train.ef_month','rainfall_train.ef_day','rainfall_train.ef_hour'])\n",
    "tmp_df['rainfall_train.class_interval'] = tmp_df['class'].combine_first(tmp_df['rainfall_train.class_interval'])\n",
    "tmp_df['rainfall_train.ef_hour'] *= 24\n",
    "tmp_df['rainfall_train.ef_hour'] = tmp_df['rainfall_train.ef_hour'].astype(int)\n",
    "tmp_df.drop(columns=['class'],inplace=True)\n",
    "tmp_df = tmp_df.rename(columns={'rainfall_train.class_interval':'class'})\n",
    "tmp_df.columns = tmp_df.columns.str.replace(\"rainfall_train\", \"rainfall_test\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "metadata": {},
   "outputs": [],
   "source": [
    "RAW= pd.read_csv(\"./data/rainfall_test.csv\").drop(columns=[\"Unnamed: 0\"])\n",
    "ind = RAW['rainfall_test.class_interval'] == -999\n",
    "tt_df = pd.merge(RAW,tmp_df[['rainfall_test.stn4contest','rainfall_test.ef_year','rainfall_test.ef_month','rainfall_test.ef_day','rainfall_test.ef_hour','class']],how='left',on=['rainfall_test.stn4contest','rainfall_test.ef_year','rainfall_test.ef_month','rainfall_test.ef_day','rainfall_test.ef_hour'])\n",
    "RAW['rainfall_test.class_interval']=tt_df['class'].combine_first(tt_df['rainfall_test.class_interval'])\n",
    "RAW['rainfall_test.class_interval'] = RAW['rainfall_test.class_interval'].astype(int)\n",
    "RAW.loc[ind,'rainfall_test.class_interval'] = -999"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 제출파일 검증"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 179,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "rainfall_test.class_interval\n",
       " 0      103360\n",
       " 5       12300\n",
       " 6        1934\n",
       " 7        1789\n",
       " 2        1019\n",
       " 4         848\n",
       " 3         260\n",
       " 9         180\n",
       " 8         180\n",
       "-999       130\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 179,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "RAW['rainfall_test.class_interval'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 180,
   "metadata": {},
   "outputs": [],
   "source": [
    "RAW.to_csv('240198.csv',index=False,encoding='euc-kr')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "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": "3.10.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}