diff --git "a/examples/model_report/\345\206\263\347\255\226\346\240\221\347\273\204\345\220\210\347\255\226\347\225\245\346\214\226\346\216\230.xlsx" "b/examples/model_report/\345\206\263\347\255\226\346\240\221\347\273\204\345\220\210\347\255\226\347\225\245\346\214\226\346\216\230.xlsx"
new file mode 100644
index 0000000..0b51582
Binary files /dev/null and "b/examples/model_report/\345\206\263\347\255\226\346\240\221\347\273\204\345\220\210\347\255\226\347\225\245\346\214\226\346\216\230.xlsx" differ
diff --git a/examples/rule_extraction.ipynb b/examples/rule_extraction.ipynb
new file mode 100644
index 0000000..772f6cf
--- /dev/null
+++ b/examples/rule_extraction.ipynb
@@ -0,0 +1,141 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "import sys\n",
+    "sys.path.append(\"../\")\n",
+    "\n",
+    "import os\n",
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "import matplotlib.pyplot as plt\n",
+    "from sklearn.model_selection import train_test_split\n",
+    "from scorecardpipeline import *\n",
+    "from scorecardpipeline.rule_extraction import DecisionTreeRuleExtractor\n",
+    "\n",
+    "\n",
+    "logger = init_setting(seed=8888, logger=True)\n",
+    "\n",
+    "\n",
+    "feature_map = {}\n",
+    "n_samples = 10000\n",
+    "ab = np.array(list('ABCDEFG'))\n",
+    "\n",
+    "data = pd.DataFrame({\n",
+    "    'A': np.random.randint(10, size = n_samples),\n",
+    "    'B': ab[np.random.choice(7, n_samples)],\n",
+    "    'C': ab[np.random.choice(2, n_samples)],\n",
+    "    '时间': np.random.random(size = n_samples),\n",
+    "    'target': np.random.randint(2, size = n_samples)\n",
+    "})\n",
+    "\n",
+    "\n",
+    "train, test = train_test_split(data, test_size=0.3, stratify=data[\"target\"])\n",
+    "\n",
+    "\n",
+    "pdtr = DecisionTreeRuleExtractor(target=\"target\", feature_map=feature_map, max_iter=8)\n",
+    "pdtr.fit(train, lift=0., max_depth=2, max_samples=1., verbose=False, min_samples_split=8, min_samples_leaf=5)\n",
+    "report = pdtr.report(valid=[test, train, data], save=\"model_report/决策树组合策略挖掘.xlsx\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "outputs": [
+    {
+     "data": {
+      "text/plain": "                          组合策略   命中数    命中率  好样本数  好样本占比  坏样本数  坏样本占比     坏率  样本整体坏率  LIFT值\n0     B <= 0.495  & 时间 > 0.943   243 0.0347   142 0.0401   101 0.0292 0.4156  0.4941 0.8411\n1   B <= 0.495  & 时间 <= 0.943   3758 0.5369  1933 0.5459  1825 0.5276 0.4856  0.4941 0.9828\n2     B > 0.495 & 时间 <= 0.877   2669 0.3813  1321 0.3731  1348 0.3897 0.5051  0.4941 1.0221\n3       B > 0.495 & 时间 > 0.877   330 0.0471   145 0.0409   185 0.0535 0.5606  0.4941 1.1345\n4      B <= 0.495  & A <= 8.5   3605 0.5150  1889 0.5335  1716 0.4961 0.4760  0.4941 0.9633\n5          B > 0.495 & A > 4.5  1496 0.2137   749 0.2115   747 0.2160 0.4993  0.4941 1.0105\n6        B > 0.495 & A <= 4.5   1503 0.2147   717 0.2025   786 0.2272 0.5230  0.4941 1.0583\n7        B <= 0.495  & A > 8.5   396 0.0566   186 0.0525   210 0.0607 0.5303  0.4941 1.0732\n8        A <= 8.5  & A <= 3.5   2772 0.3960  1432 0.4044  1340 0.3874 0.4834  0.4941 0.9783\n9          A <= 8.5  & A > 3.5  3526 0.5037  1775 0.5013  1751 0.5062 0.4966  0.4941 1.0050\n10       A > 8.5 & C <= 0.494    358 0.0511   174 0.0491   184 0.0532 0.5140  0.4941 1.0401\n11         A > 8.5 & C > 0.494   344 0.0491   160 0.0452   184 0.0532 0.5349  0.4941 1.0824",
+      "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>组合策略</th>\n      <th>命中数</th>\n      <th>命中率</th>\n      <th>好样本数</th>\n      <th>好样本占比</th>\n      <th>坏样本数</th>\n      <th>坏样本占比</th>\n      <th>坏率</th>\n      <th>样本整体坏率</th>\n      <th>LIFT值</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>B &lt;= 0.495  &amp; 时间 &gt; 0.943</td>\n      <td>243</td>\n      <td>0.0347</td>\n      <td>142</td>\n      <td>0.0401</td>\n      <td>101</td>\n      <td>0.0292</td>\n      <td>0.4156</td>\n      <td>0.4941</td>\n      <td>0.8411</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>B &lt;= 0.495  &amp; 时间 &lt;= 0.943</td>\n      <td>3758</td>\n      <td>0.5369</td>\n      <td>1933</td>\n      <td>0.5459</td>\n      <td>1825</td>\n      <td>0.5276</td>\n      <td>0.4856</td>\n      <td>0.4941</td>\n      <td>0.9828</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>B &gt; 0.495 &amp; 时间 &lt;= 0.877</td>\n      <td>2669</td>\n      <td>0.3813</td>\n      <td>1321</td>\n      <td>0.3731</td>\n      <td>1348</td>\n      <td>0.3897</td>\n      <td>0.5051</td>\n      <td>0.4941</td>\n      <td>1.0221</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>B &gt; 0.495 &amp; 时间 &gt; 0.877</td>\n      <td>330</td>\n      <td>0.0471</td>\n      <td>145</td>\n      <td>0.0409</td>\n      <td>185</td>\n      <td>0.0535</td>\n      <td>0.5606</td>\n      <td>0.4941</td>\n      <td>1.1345</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>B &lt;= 0.495  &amp; A &lt;= 8.5</td>\n      <td>3605</td>\n      <td>0.5150</td>\n      <td>1889</td>\n      <td>0.5335</td>\n      <td>1716</td>\n      <td>0.4961</td>\n      <td>0.4760</td>\n      <td>0.4941</td>\n      <td>0.9633</td>\n    </tr>\n    <tr>\n      <th>5</th>\n      <td>B &gt; 0.495 &amp; A &gt; 4.5</td>\n      <td>1496</td>\n      <td>0.2137</td>\n      <td>749</td>\n      <td>0.2115</td>\n      <td>747</td>\n      <td>0.2160</td>\n      <td>0.4993</td>\n      <td>0.4941</td>\n      <td>1.0105</td>\n    </tr>\n    <tr>\n      <th>6</th>\n      <td>B &gt; 0.495 &amp; A &lt;= 4.5</td>\n      <td>1503</td>\n      <td>0.2147</td>\n      <td>717</td>\n      <td>0.2025</td>\n      <td>786</td>\n      <td>0.2272</td>\n      <td>0.5230</td>\n      <td>0.4941</td>\n      <td>1.0583</td>\n    </tr>\n    <tr>\n      <th>7</th>\n      <td>B &lt;= 0.495  &amp; A &gt; 8.5</td>\n      <td>396</td>\n      <td>0.0566</td>\n      <td>186</td>\n      <td>0.0525</td>\n      <td>210</td>\n      <td>0.0607</td>\n      <td>0.5303</td>\n      <td>0.4941</td>\n      <td>1.0732</td>\n    </tr>\n    <tr>\n      <th>8</th>\n      <td>A &lt;= 8.5  &amp; A &lt;= 3.5</td>\n      <td>2772</td>\n      <td>0.3960</td>\n      <td>1432</td>\n      <td>0.4044</td>\n      <td>1340</td>\n      <td>0.3874</td>\n      <td>0.4834</td>\n      <td>0.4941</td>\n      <td>0.9783</td>\n    </tr>\n    <tr>\n      <th>9</th>\n      <td>A &lt;= 8.5  &amp; A &gt; 3.5</td>\n      <td>3526</td>\n      <td>0.5037</td>\n      <td>1775</td>\n      <td>0.5013</td>\n      <td>1751</td>\n      <td>0.5062</td>\n      <td>0.4966</td>\n      <td>0.4941</td>\n      <td>1.0050</td>\n    </tr>\n    <tr>\n      <th>10</th>\n      <td>A &gt; 8.5 &amp; C &lt;= 0.494</td>\n      <td>358</td>\n      <td>0.0511</td>\n      <td>174</td>\n      <td>0.0491</td>\n      <td>184</td>\n      <td>0.0532</td>\n      <td>0.5140</td>\n      <td>0.4941</td>\n      <td>1.0401</td>\n    </tr>\n    <tr>\n      <th>11</th>\n      <td>A &gt; 8.5 &amp; C &gt; 0.494</td>\n      <td>344</td>\n      <td>0.0491</td>\n      <td>160</td>\n      <td>0.0452</td>\n      <td>184</td>\n      <td>0.0532</td>\n      <td>0.5349</td>\n      <td>0.4941</td>\n      <td>1.0824</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "report[0]"
+   ],
+   "metadata": {
+    "collapsed": false
+   }
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "outputs": [
+    {
+     "data": {
+      "text/plain": "                          组合策略   命中数    命中率 好样本数  好样本占比 坏样本数  坏样本占比     坏率 样本整体坏率  LIFT值\n0     B <= 0.495  & 时间 > 0.943    96 0.0320   50 0.0329   46 0.0310 0.4792 0.4940 0.9700\n1   B <= 0.495  & 时间 <= 0.943   1604 0.5347  825 0.5435  779 0.5256 0.4857 0.4940 0.9831\n2     B > 0.495 & 时间 <= 0.877   1146 0.3820  570 0.3755  576 0.3887 0.5026 0.4940 1.0174\n3       B > 0.495 & 时间 > 0.877   154 0.0513   73 0.0481   81 0.0547 0.5260 0.4940 1.0647\n4      B <= 0.495  & A <= 8.5   1535 0.5117  791 0.5211  744 0.5020 0.4847 0.4940 0.9812\n5          B > 0.495 & A > 4.5   617 0.2057  306 0.2016  311 0.2099 0.5041 0.4940 1.0203\n6        B > 0.495 & A <= 4.5    683 0.2277  337 0.2220  346 0.2335 0.5066 0.4940 1.0255\n7        B <= 0.495  & A > 8.5   165 0.0550   84 0.0553   81 0.0547 0.4909 0.4940 0.9937\n8        A <= 8.5  & A <= 3.5   1214 0.4047  615 0.4051  599 0.4042 0.4934 0.4940 0.9988\n9          A <= 8.5  & A > 3.5  1487 0.4957  744 0.4901  743 0.5013 0.4997 0.4940 1.0115\n10       A > 8.5 & C <= 0.494    147 0.0490   81 0.0534   66 0.0445 0.4490 0.4940 0.9089\n11         A > 8.5 & C > 0.494   152 0.0507   78 0.0514   74 0.0499 0.4868 0.4940 0.9855",
+      "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>组合策略</th>\n      <th>命中数</th>\n      <th>命中率</th>\n      <th>好样本数</th>\n      <th>好样本占比</th>\n      <th>坏样本数</th>\n      <th>坏样本占比</th>\n      <th>坏率</th>\n      <th>样本整体坏率</th>\n      <th>LIFT值</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>B &lt;= 0.495  &amp; 时间 &gt; 0.943</td>\n      <td>96</td>\n      <td>0.0320</td>\n      <td>50</td>\n      <td>0.0329</td>\n      <td>46</td>\n      <td>0.0310</td>\n      <td>0.4792</td>\n      <td>0.4940</td>\n      <td>0.9700</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>B &lt;= 0.495  &amp; 时间 &lt;= 0.943</td>\n      <td>1604</td>\n      <td>0.5347</td>\n      <td>825</td>\n      <td>0.5435</td>\n      <td>779</td>\n      <td>0.5256</td>\n      <td>0.4857</td>\n      <td>0.4940</td>\n      <td>0.9831</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>B &gt; 0.495 &amp; 时间 &lt;= 0.877</td>\n      <td>1146</td>\n      <td>0.3820</td>\n      <td>570</td>\n      <td>0.3755</td>\n      <td>576</td>\n      <td>0.3887</td>\n      <td>0.5026</td>\n      <td>0.4940</td>\n      <td>1.0174</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>B &gt; 0.495 &amp; 时间 &gt; 0.877</td>\n      <td>154</td>\n      <td>0.0513</td>\n      <td>73</td>\n      <td>0.0481</td>\n      <td>81</td>\n      <td>0.0547</td>\n      <td>0.5260</td>\n      <td>0.4940</td>\n      <td>1.0647</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>B &lt;= 0.495  &amp; A &lt;= 8.5</td>\n      <td>1535</td>\n      <td>0.5117</td>\n      <td>791</td>\n      <td>0.5211</td>\n      <td>744</td>\n      <td>0.5020</td>\n      <td>0.4847</td>\n      <td>0.4940</td>\n      <td>0.9812</td>\n    </tr>\n    <tr>\n      <th>5</th>\n      <td>B &gt; 0.495 &amp; A &gt; 4.5</td>\n      <td>617</td>\n      <td>0.2057</td>\n      <td>306</td>\n      <td>0.2016</td>\n      <td>311</td>\n      <td>0.2099</td>\n      <td>0.5041</td>\n      <td>0.4940</td>\n      <td>1.0203</td>\n    </tr>\n    <tr>\n      <th>6</th>\n      <td>B &gt; 0.495 &amp; A &lt;= 4.5</td>\n      <td>683</td>\n      <td>0.2277</td>\n      <td>337</td>\n      <td>0.2220</td>\n      <td>346</td>\n      <td>0.2335</td>\n      <td>0.5066</td>\n      <td>0.4940</td>\n      <td>1.0255</td>\n    </tr>\n    <tr>\n      <th>7</th>\n      <td>B &lt;= 0.495  &amp; A &gt; 8.5</td>\n      <td>165</td>\n      <td>0.0550</td>\n      <td>84</td>\n      <td>0.0553</td>\n      <td>81</td>\n      <td>0.0547</td>\n      <td>0.4909</td>\n      <td>0.4940</td>\n      <td>0.9937</td>\n    </tr>\n    <tr>\n      <th>8</th>\n      <td>A &lt;= 8.5  &amp; A &lt;= 3.5</td>\n      <td>1214</td>\n      <td>0.4047</td>\n      <td>615</td>\n      <td>0.4051</td>\n      <td>599</td>\n      <td>0.4042</td>\n      <td>0.4934</td>\n      <td>0.4940</td>\n      <td>0.9988</td>\n    </tr>\n    <tr>\n      <th>9</th>\n      <td>A &lt;= 8.5  &amp; A &gt; 3.5</td>\n      <td>1487</td>\n      <td>0.4957</td>\n      <td>744</td>\n      <td>0.4901</td>\n      <td>743</td>\n      <td>0.5013</td>\n      <td>0.4997</td>\n      <td>0.4940</td>\n      <td>1.0115</td>\n    </tr>\n    <tr>\n      <th>10</th>\n      <td>A &gt; 8.5 &amp; C &lt;= 0.494</td>\n      <td>147</td>\n      <td>0.0490</td>\n      <td>81</td>\n      <td>0.0534</td>\n      <td>66</td>\n      <td>0.0445</td>\n      <td>0.4490</td>\n      <td>0.4940</td>\n      <td>0.9089</td>\n    </tr>\n    <tr>\n      <th>11</th>\n      <td>A &gt; 8.5 &amp; C &gt; 0.494</td>\n      <td>152</td>\n      <td>0.0507</td>\n      <td>78</td>\n      <td>0.0514</td>\n      <td>74</td>\n      <td>0.0499</td>\n      <td>0.4868</td>\n      <td>0.4940</td>\n      <td>0.9855</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
+     },
+     "execution_count": 3,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "report[1]"
+   ],
+   "metadata": {
+    "collapsed": false
+   }
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "outputs": [
+    {
+     "data": {
+      "text/plain": "                          组合策略   命中数    命中率  好样本数  好样本占比  坏样本数  坏样本占比     坏率 样本整体坏率  LIFT值\n0     B <= 0.495  & 时间 > 0.943   242 0.0346   141 0.0398   101 0.0292 0.4174 0.4941 0.8446\n1   B <= 0.495  & 时间 <= 0.943   3759 0.5370  1934 0.5462  1825 0.5276 0.4855 0.4941 0.9825\n2     B > 0.495 & 时间 <= 0.877   2668 0.3811  1320 0.3728  1348 0.3897 0.5052 0.4941 1.0225\n3       B > 0.495 & 时间 > 0.877   331 0.0473   146 0.0412   185 0.0535 0.5589 0.4941 1.1311\n4      B <= 0.495  & A <= 8.5   3605 0.5150  1889 0.5335  1716 0.4961 0.4760 0.4941 0.9633\n5          B > 0.495 & A > 4.5  1496 0.2137   749 0.2115   747 0.2160 0.4993 0.4941 1.0105\n6        B > 0.495 & A <= 4.5   1503 0.2147   717 0.2025   786 0.2272 0.5230 0.4941 1.0583\n7        B <= 0.495  & A > 8.5   396 0.0566   186 0.0525   210 0.0607 0.5303 0.4941 1.0732\n8        A <= 8.5  & A <= 3.5   2772 0.3960  1432 0.4044  1340 0.3874 0.4834 0.4941 0.9783\n9          A <= 8.5  & A > 3.5  3526 0.5037  1775 0.5013  1751 0.5062 0.4966 0.4941 1.0050\n10       A > 8.5 & C <= 0.494    358 0.0511   174 0.0491   184 0.0532 0.5140 0.4941 1.0401\n11         A > 8.5 & C > 0.494   344 0.0491   160 0.0452   184 0.0532 0.5349 0.4941 1.0824",
+      "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>组合策略</th>\n      <th>命中数</th>\n      <th>命中率</th>\n      <th>好样本数</th>\n      <th>好样本占比</th>\n      <th>坏样本数</th>\n      <th>坏样本占比</th>\n      <th>坏率</th>\n      <th>样本整体坏率</th>\n      <th>LIFT值</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>B &lt;= 0.495  &amp; 时间 &gt; 0.943</td>\n      <td>242</td>\n      <td>0.0346</td>\n      <td>141</td>\n      <td>0.0398</td>\n      <td>101</td>\n      <td>0.0292</td>\n      <td>0.4174</td>\n      <td>0.4941</td>\n      <td>0.8446</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>B &lt;= 0.495  &amp; 时间 &lt;= 0.943</td>\n      <td>3759</td>\n      <td>0.5370</td>\n      <td>1934</td>\n      <td>0.5462</td>\n      <td>1825</td>\n      <td>0.5276</td>\n      <td>0.4855</td>\n      <td>0.4941</td>\n      <td>0.9825</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>B &gt; 0.495 &amp; 时间 &lt;= 0.877</td>\n      <td>2668</td>\n      <td>0.3811</td>\n      <td>1320</td>\n      <td>0.3728</td>\n      <td>1348</td>\n      <td>0.3897</td>\n      <td>0.5052</td>\n      <td>0.4941</td>\n      <td>1.0225</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>B &gt; 0.495 &amp; 时间 &gt; 0.877</td>\n      <td>331</td>\n      <td>0.0473</td>\n      <td>146</td>\n      <td>0.0412</td>\n      <td>185</td>\n      <td>0.0535</td>\n      <td>0.5589</td>\n      <td>0.4941</td>\n      <td>1.1311</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>B &lt;= 0.495  &amp; A &lt;= 8.5</td>\n      <td>3605</td>\n      <td>0.5150</td>\n      <td>1889</td>\n      <td>0.5335</td>\n      <td>1716</td>\n      <td>0.4961</td>\n      <td>0.4760</td>\n      <td>0.4941</td>\n      <td>0.9633</td>\n    </tr>\n    <tr>\n      <th>5</th>\n      <td>B &gt; 0.495 &amp; A &gt; 4.5</td>\n      <td>1496</td>\n      <td>0.2137</td>\n      <td>749</td>\n      <td>0.2115</td>\n      <td>747</td>\n      <td>0.2160</td>\n      <td>0.4993</td>\n      <td>0.4941</td>\n      <td>1.0105</td>\n    </tr>\n    <tr>\n      <th>6</th>\n      <td>B &gt; 0.495 &amp; A &lt;= 4.5</td>\n      <td>1503</td>\n      <td>0.2147</td>\n      <td>717</td>\n      <td>0.2025</td>\n      <td>786</td>\n      <td>0.2272</td>\n      <td>0.5230</td>\n      <td>0.4941</td>\n      <td>1.0583</td>\n    </tr>\n    <tr>\n      <th>7</th>\n      <td>B &lt;= 0.495  &amp; A &gt; 8.5</td>\n      <td>396</td>\n      <td>0.0566</td>\n      <td>186</td>\n      <td>0.0525</td>\n      <td>210</td>\n      <td>0.0607</td>\n      <td>0.5303</td>\n      <td>0.4941</td>\n      <td>1.0732</td>\n    </tr>\n    <tr>\n      <th>8</th>\n      <td>A &lt;= 8.5  &amp; A &lt;= 3.5</td>\n      <td>2772</td>\n      <td>0.3960</td>\n      <td>1432</td>\n      <td>0.4044</td>\n      <td>1340</td>\n      <td>0.3874</td>\n      <td>0.4834</td>\n      <td>0.4941</td>\n      <td>0.9783</td>\n    </tr>\n    <tr>\n      <th>9</th>\n      <td>A &lt;= 8.5  &amp; A &gt; 3.5</td>\n      <td>3526</td>\n      <td>0.5037</td>\n      <td>1775</td>\n      <td>0.5013</td>\n      <td>1751</td>\n      <td>0.5062</td>\n      <td>0.4966</td>\n      <td>0.4941</td>\n      <td>1.0050</td>\n    </tr>\n    <tr>\n      <th>10</th>\n      <td>A &gt; 8.5 &amp; C &lt;= 0.494</td>\n      <td>358</td>\n      <td>0.0511</td>\n      <td>174</td>\n      <td>0.0491</td>\n      <td>184</td>\n      <td>0.0532</td>\n      <td>0.5140</td>\n      <td>0.4941</td>\n      <td>1.0401</td>\n    </tr>\n    <tr>\n      <th>11</th>\n      <td>A &gt; 8.5 &amp; C &gt; 0.494</td>\n      <td>344</td>\n      <td>0.0491</td>\n      <td>160</td>\n      <td>0.0452</td>\n      <td>184</td>\n      <td>0.0532</td>\n      <td>0.5349</td>\n      <td>0.4941</td>\n      <td>1.0824</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
+     },
+     "execution_count": 4,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "report[2]"
+   ],
+   "metadata": {
+    "collapsed": false
+   }
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "outputs": [],
+   "source": [],
+   "metadata": {
+    "collapsed": false
+   }
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}
diff --git a/scorecardpipeline/rule_extraction.py b/scorecardpipeline/rule_extraction.py
new file mode 100644
index 0000000..1a68b50
--- /dev/null
+++ b/scorecardpipeline/rule_extraction.py
@@ -0,0 +1,358 @@
+# -*- coding: utf-8 -*-
+"""
+@Time    : 2024/2/29 13:29
+@Author  : itlubber
+@Site    : itlubber.art
+"""
+import warnings
+import os
+import re
+import graphviz
+import dtreeviz
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from matplotlib import font_manager
+from openpyxl.worksheet.worksheet import Worksheet
+
+import category_encoders as ce
+from optbinning import OptimalBinning
+from sklearn.tree import DecisionTreeClassifier
+
+from .utils import init_setting
+from .excel_writer import ExcelWriter, dataframe2excel
+
+
+class DecisionTreeRuleExtractor:
+    def __init__(self, target="target", labels=["positive", "negative"], feature_map={}, nan=-1., max_iter=128, writer=None, combiner=None, seed=None, theme_color="2639E9"):
+        """决策树自动规则挖掘工具包
+
+        :param target: 数据集中好坏样本标签列名称，默认 target
+        :param labels: 好坏样本标签名称，传入一个长度为2的列表，第0个元素为好样本标签，第1个元素为坏样本标签，默认 ["positive", "negative"]
+        :param feature_map: 变量名称及其含义，在后续输出报告和策略信息时增加可读性，默认 {}
+        :param nan: 在决策树策略挖掘时，默认空值填充的值，默认 -1
+        :param max_iter: 最多支持在数据集上训练多少颗树模型，每次生成一棵树后，会剔除特征重要性最高的特征后，再生成树，默认 128
+        :param writer: 在之前程序运行时生成的 ExcelWriter，可以支持传入一个已有的writer，后续所有内容将保存至该workbook中，默认 None
+        """
+        self.seed = seed
+        self.nan = nan
+        self.target = target
+        self.labels = labels
+        self.theme_color = theme_color
+        self.feature_map = feature_map
+        self.decision_trees = []
+        self.max_iter = max_iter
+        self.target_enc = None
+        self.feature_names = None
+        self.dt_rules = pd.DataFrame()
+        self.end_row = 2
+        self.start_col = 2
+        self.describe_columns = ["组合策略", "命中数", "命中率", "好样本数", "好样本占比", "坏样本数", "坏样本占比", "坏率", "样本整体坏率", "LIFT值"]
+
+        init_setting()
+
+        if writer:
+            self.writer = writer
+        else:
+            self.writer = ExcelWriter(theme_color=self.theme_color)
+
+    def encode_cat_features(self, X, y):
+        cat_features = list(set(X.select_dtypes(include=[object, pd.CategoricalDtype]).columns))
+        cat_features_index = [i for i, f in enumerate(X.columns) if f in cat_features]
+
+        if len(cat_features) > 0:
+            if self.target_enc is None:
+                self.target_enc = ce.TargetEncoder(cols=cat_features)
+                self.target_enc.fit(X[cat_features], y)
+                self.target_enc.target_mapping = {}
+                X_TE = X.join(self.target_enc.transform(X[cat_features]).add_suffix('_target'))
+                for col in cat_features:
+                    mapping = X_TE[[col, f"{col}_target"]].drop_duplicates()
+                    self.target_enc.target_mapping[col] = dict(zip(mapping[col], mapping[f"{col}_target"]))
+            else:
+                X_TE = X.join(self.target_enc.transform(X[cat_features]).add_suffix('_target'))
+
+            X_TE = X_TE.drop(columns=cat_features)
+            return X_TE.rename(columns={f"{c}_target": c for c in cat_features})
+        else:
+            return X
+
+    def get_dt_rules(self, tree, feature_names, total_bad_rate, total_count):
+        tree_ = tree.tree_
+        left = tree.tree_.children_left
+        right = tree.tree_.children_right
+        feature_name = [feature_names[i] if i != -2 else "undefined!" for i in tree_.feature]
+        rules = dict()
+
+        result_dataframe = pd.DataFrame()
+
+        def recurse(node, depth, parent):  # 搜每个节点的规则
+            nonlocal result_dataframe
+
+            if tree_.feature[node] != -2:  # 非叶子节点,搜索每个节点的规则
+                name = feature_name[node]
+                thd = np.round(tree_.threshold[node], 3)
+                s = "{} <= {} ".format(name, thd, node)
+                # 左子
+                if node == 0:
+                    rules[node] = s
+                else:
+                    rules[node] = rules[parent] + ' & ' + s
+                recurse(left[node], depth + 1, node)
+                s = "{} > {}".format(name, thd)
+                # 右子
+                if node == 0:
+                    rules[node] = s
+                else:
+                    rules[node] = rules[parent] + ' & ' + s
+                recurse(right[node], depth + 1, node)
+            else:
+                result = pd.DataFrame()
+                result['组合策略'] = rules[parent],
+                result['好样本数'] = tree_.value[node][0][0].astype(int)
+                result['好样本占比'] = result['好样本数'] / (total_count * (1 - total_bad_rate))
+                result['坏样本数'] = tree_.value[node][0][1].astype(int)
+                result['坏样本占比'] = result['坏样本数'] / (total_count * total_bad_rate)
+                result['命中数'] = result['好样本数'] + result['坏样本数']
+                result['命中率'] = result['命中数'] / total_count
+                result['坏率'] = result['坏样本数'] / result['命中数']
+                result['样本整体坏率'] = total_bad_rate
+                result['LIFT值'] = result['坏率'] / result['样本整体坏率']
+
+                result_dataframe = pd.concat([result_dataframe, result], axis=0)
+
+        recurse(0, 1, 0)
+
+        return result_dataframe.sort_values("LIFT值", ascending=True)[self.describe_columns].reset_index(drop=True)
+
+    def select_dt_rules(self, decision_tree, x, y, lift=0., max_samples=1., save=None, verbose=False, drop=False):
+        rules = self.get_dt_rules(decision_tree, x.columns, sum(y) / len(y), len(y))
+        total_rules = len(rules)
+
+        try:
+            viz_model = dtreeviz.model(decision_tree,
+                                       X_train=x,
+                                       y_train=y,
+                                       feature_names=x.columns,
+                                       target_name=self.target,
+                                       class_names=self.labels,
+                                       )
+        except AttributeError:
+            raise "请检查 dtreeviz 版本"
+
+        rules = rules.query(f"LIFT值 >= {lift} & 命中率 <= {max_samples}").reset_index(drop=True)
+
+        if len(rules) > 0:
+            # font_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'matplot_chinese.ttf')
+            # font_manager.fontManager.addfont(font_path)
+            # plt.rcParams['font.family'] = font_manager.FontProperties(fname=font_path).get_name()
+            # plt.rcParams['axes.unicode_minus'] = False
+
+            decision_tree_viz = viz_model.view(
+                scale=1.5,
+                orientation='LR',
+                colors={
+                    "classes": [None, None, ["#2639E9", "#F76E6C"], ["#2639E9", "#F76E6C", "#FE7715", "#FFFFFF"]],
+                    "arrow": "#2639E9",
+                    'text_wedge': "#F76E6C",
+                    "pie": "#2639E9",
+                    "tile_alpha": 1,
+                    "legend_edge": "#FFFFFF",
+                },
+                ticks_fontsize=10,
+                label_fontsize=10,
+                fontname=plt.rcParams['font.family'],
+            )
+            if verbose:
+                from IPython.core.display_functions import display
+                if self.feature_map is not None and len(self.feature_map) > 0:
+                    display(rules.replace(self.feature_map, regex=True))
+                else:
+                    display(rules)
+                display(decision_tree_viz)
+            if save:
+                if os.path.dirname(save) and not os.path.exists(os.path.dirname(save)):
+                    os.makedirs(os.path.dirname(save))
+
+                try:
+                    decision_tree_viz.save("combine_rules_cache.svg")
+                except graphviz.backend.execute.ExecutableNotFound:
+                    print("请确保您已安装 graphviz 程序并且正确配置了 PATH 路径。可参考: https://stackoverflow.com/questions/35064304/runtimeerror-make-sure-the-graphviz-executables-are-on-your-systems-path-aft")
+
+                try:
+                    import cairosvg
+                    cairosvg.svg2png(url="combine_rules_cache.svg", write_to=save, dpi=240)
+                except:
+                    from reportlab.graphics import renderPDF
+                    from svglib.svglib import svg2rlg
+                    drawing = svg2rlg("combine_rules_cache.svg")
+                    renderPDF.drawToFile(drawing, save, dpi=240, fmt="PNG")
+
+        if os.path.isfile("combine_rules_cache.svg"):
+            os.remove("combine_rules_cache.svg")
+
+        if os.path.isfile("combine_rules_cache"):
+            os.remove("combine_rules_cache")
+
+        if drop:
+            if len(rules) > 0:
+                return rules, decision_tree.feature_names_in_[list(decision_tree.feature_importances_).index(max(decision_tree.feature_importances_))], total_rules
+            else:
+                return rules, decision_tree.feature_names_in_[list(decision_tree.feature_importances_).index(min(decision_tree.feature_importances_))], total_rules
+        else:
+            return rules, total_rules
+
+    def query_dt_rules(self, x, y, parsed_rules=None):
+        total_count = len(y)
+        total_bad_rate = y.sum() / len(y)
+
+        rules = pd.DataFrame()
+
+        if isinstance(parsed_rules, pd.DataFrame):
+            parsed_rules = parsed_rules["组合策略"].unique()
+
+        for rule in parsed_rules:
+            select_index = x.query(rule).index
+            if len(select_index) > 0:
+                y_select = y[select_index]
+                df = pd.Series()
+                df['组合策略'] = rule
+                df['好样本数'] = len(y_select) - y_select.sum()
+                df['好样本占比'] = df['好样本数'] / (total_count * (1 - total_bad_rate))
+                df['坏样本数'] = y_select.sum()
+                df['坏样本占比'] = df['坏样本数'] / (total_count * total_bad_rate)
+                df['命中数'] = df['好样本数'] + df['坏样本数']
+                df['命中率'] = df['命中数'] / total_count
+                df['坏率'] = df['坏样本数'] / df['命中数']
+                df['样本整体坏率'] = total_bad_rate
+                df['LIFT值'] = df['坏率'] / df['样本整体坏率']
+            else:
+                df = pd.Series({'组合策略': rule, '好样本数': 0, '好样本占比': 0., '坏样本数': 0, '坏样本占比': 0., '命中数': 0, '命中率': 0., '坏率': 0., '样本整体坏率': total_bad_rate, 'LIFT值': 0., })
+
+            rules = pd.concat([rules, pd.DataFrame(df).T]).reset_index(drop=True)
+
+        return rules[self.describe_columns]
+
+    def insert_dt_rules(self, parsed_rules, end_row, start_col, save=None, sheet=None, figsize=(500, 350)):
+        if isinstance(sheet, Worksheet):
+            worksheet = sheet
+        else:
+            worksheet = self.writer.get_sheet_by_name(sheet or "决策树组合策略挖掘")
+
+        end_row, end_col = dataframe2excel(parsed_rules, self.writer, sheet_name=worksheet, start_row=end_row + 1, start_col=start_col, percent_cols=['好样本占比', '坏样本占比', '命中率', '坏率', '样本整体坏率', 'LIFT值'], condition_cols=["坏率", "LIFT值"])
+
+        if save is not None:
+            end_row, end_col = self.writer.insert_pic2sheet(worksheet, save, (end_row + 1, start_col), figsize=figsize)
+
+        return end_row, end_col
+
+    def fit(self, x, y=None, max_depth=2, lift=0., max_samples=1., min_score=None, verbose=False, *args, **kwargs):
+        """组合策略挖掘
+
+        :param x: 包含标签的数据集
+        :param max_depth: 决策树最大深度，即最多组合的特征个数，默认 2
+        :param lift: 组合策略最小的lift值，默认 0.，即全部组合策略
+        :param max_samples: 每条组合策略的最大样本占比，默认 1.0，即全部组合策略
+        :param min_score: 决策树拟合时最小的auc，如果不满足则停止后续生成决策树
+        :param verbose: 是否调试模式，仅在 jupyter 环境有效
+        :param kwargs: DecisionTreeClassifier 参数
+        """
+        worksheet = self.writer.get_sheet_by_name("策略详情")
+
+        y = x[self.target]
+        X_TE = self.encode_cat_features(x.drop(columns=[self.target]), y)
+        X_TE = X_TE.fillna(self.nan)
+
+        self.feature_names = list(X_TE.columns)
+
+        for i in range(self.max_iter):
+            decision_tree = DecisionTreeClassifier(max_depth=max_depth, *args, **kwargs)
+            decision_tree = decision_tree.fit(X_TE, y)
+
+            if (min_score is not None and decision_tree.score(X_TE, y) < min_score) or len(X_TE.columns) < max_depth:
+                break
+
+            try:
+                parsed_rules, remove, total_rules = self.select_dt_rules(decision_tree, X_TE, y, lift=lift, max_samples=max_samples, verbose=verbose, save=f"model_report/auto_mining_rules/combiner_rules_{i}.png", drop=True)
+
+                if len(parsed_rules) > 0:
+                    self.dt_rules = pd.concat([self.dt_rules, parsed_rules]).reset_index(drop=True)
+
+                    if self.writer is not None:
+                        if self.feature_map is not None and len(self.feature_map) > 0:
+                            parsed_rules["组合策略"] = parsed_rules["组合策略"].replace(self.feature_map, regex=True)
+                        self.end_row, _ = self.insert_dt_rules(parsed_rules, self.end_row, self.start_col, save=f"model_report/auto_mining_rules/combiner_rules_{i}.png", figsize=(500, 100 * total_rules), sheet=worksheet)
+
+                X_TE = X_TE.drop(columns=remove)
+                self.decision_trees.append(decision_tree)
+            except:
+                import traceback
+                traceback.print_exc()
+
+        if len(self.dt_rules) <= 0:
+            print(f"未挖掘到有效策略, 可以考虑适当调整预设的筛选参数, 降低 lift / 提高 max_samples, 当前筛选标准为: 提取 lift >= {lift} 且 max_samples <= {max_samples} 的策略")
+
+        return self
+
+    def transform(self, x, y=None):
+        y = x[self.target]
+        X_TE = self.encode_cat_features(x.drop(columns=[self.target]), y)
+        X_TE = X_TE.fillna(self.nan)
+        if self.dt_rules is not None and len(self.dt_rules) > 0:
+            parsed_rules = self.query_dt_rules(X_TE, y, parsed_rules=self.dt_rules)
+            if self.feature_map is not None and len(self.feature_map) > 0:
+                parsed_rules["组合策略"] = parsed_rules["组合策略"].replace(self.feature_map, regex=True)
+            return parsed_rules
+        else:
+            return pd.DataFrame(columns=self.describe_columns)
+
+    def report(self, valid=None, sheet="组合策略汇总", save=None):
+        """组合策略插入excel文档
+
+        :param valid: 验证数据集
+        :param sheet: 保存组合策略的表格sheet名称
+        :param save: 保存报告的文件路径
+
+        :return: 返回每个数据集组合策略命中情况
+        """
+        worksheet = self.writer.get_sheet_by_name(sheet or "决策树组合策略挖掘")
+
+        if sheet:
+            self.writer.workbook.move_sheet(sheet, -1)
+
+        parsed_rules_train = self.dt_rules.copy()
+
+        if self.feature_map is not None and len(self.feature_map) > 0:
+            parsed_rules_train["组合策略"] = parsed_rules_train["组合策略"].replace(self.feature_map, regex=True)
+
+        self.end_row, _ = self.writer.insert_value2sheet(worksheet, (2 if sheet else self.end_row + 2, self.start_col), value="组合策略: 训练集", style="header_middle", end_space=(2 if sheet else self.end_row + 2, self.start_col + len(parsed_rules_train.columns)))
+        self.end_row, _ = self.insert_dt_rules(parsed_rules_train, self.end_row, self.start_col, sheet=worksheet)
+        outputs = (parsed_rules_train,)
+
+        if valid is not None:
+            if isinstance(valid, pd.DataFrame) and len(valid) > 0:
+                parsed_rules_val = self.transform(valid)
+                self.end_row, _ = self.writer.insert_value2sheet(worksheet, (self.end_row + 2, self.start_col), value="组合策略: 验证集", style="header_middle", end_space=(self.end_row + 2, self.start_col + len(parsed_rules_val.columns)))
+                self.end_row, _ = self.insert_dt_rules(parsed_rules_val, self.end_row, self.start_col, sheet=worksheet)
+                outputs = outputs + (parsed_rules_val,)
+
+            elif isinstance(valid, (list, tuple)):
+                for i, dataset in enumerate(valid):
+                    if isinstance(dataset, pd.DataFrame) and len(dataset) > 0:
+                        parsed_rules_val = self.transform(dataset)
+                        self.end_row, _ = self.writer.insert_value2sheet(worksheet, (self.end_row + 2, self.start_col), value=f"组合策略: 验证集 {i + 1}", style="header_middle", end_space=(self.end_row + 2, self.start_col + len(parsed_rules_val.columns)))
+                        self.end_row, _ = self.insert_dt_rules(parsed_rules_val, self.end_row, self.start_col, sheet=worksheet)
+                        outputs = outputs + (parsed_rules_val,)
+
+            elif isinstance(valid, dict):
+                for k, dataset in valid.items():
+                    if isinstance(dataset, pd.DataFrame) and len(dataset) > 0:
+                        parsed_rules_val = self.transform(dataset)
+                        self.end_row, _ = self.writer.insert_value2sheet(worksheet, (self.end_row + 2, self.start_col), value=f"组合策略: {k}", style="header_middle", end_space=(self.end_row + 2, self.start_col + len(parsed_rules_val.columns)))
+                        self.end_row, _ = self.insert_dt_rules(parsed_rules_val, self.end_row, self.start_col, sheet=worksheet)
+                        outputs = outputs + (parsed_rules_val,)
+
+        if save:
+            self.writer.save(save)
+
+        return outputs