From 8af8d4638605f2404a7aae8d82cbc67c9d513d63 Mon Sep 17 00:00:00 2001
From: shataowei <taowei.sha@ymm56.com>
Date: Wed, 20 Nov 2019 19:40:03 +0800
Subject: [PATCH] =?UTF-8?q?xgboost=E9=83=A8=E5=88=86?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .idea/workspace.xml                           | 51 +++++--------------
 README.md                                     |  2 +
 .../Xgboost.md"                               | 13 ++++-
 3 files changed, 28 insertions(+), 38 deletions(-)

diff --git a/.idea/workspace.xml b/.idea/workspace.xml
index 1b832af..fefe9f3 100644
--- a/.idea/workspace.xml
+++ b/.idea/workspace.xml
@@ -3,6 +3,7 @@
   <component name="ChangeListManager">
     <list default="true" id="be5bf74e-ad1c-4197-8a1f-ac93566d8aa7" name="Default" comment="init">
       <change type="MODIFICATION" beforePath="$PROJECT_DIR$/README.md" afterPath="$PROJECT_DIR$/README.md" />
+      <change type="MODIFICATION" beforePath="$PROJECT_DIR$/机器学习/集成学习/Xgboost.md" afterPath="$PROJECT_DIR$/机器学习/集成学习/Xgboost.md" />
     </list>
     <ignored path="Reflection_Summary.iws" />
     <ignored path=".idea/workspace.xml" />
@@ -36,32 +37,8 @@
           </provider>
           <provider selected="true" editor-type-id="split-provider[text-editor;MarkdownPreviewEditor]">
             <state split_layout="SPLIT">
-              <first_editor relative-caret-position="415">
-                <caret line="239" column="47" selection-start-line="239" selection-start-column="47" selection-end-line="239" selection-end-column="47" />
-                <folding />
-              </first_editor>
-              <second_editor>
-                <js_state />
-              </second_editor>
-            </state>
-          </provider>
-        </entry>
-      </file>
-      <file leaf-file-name="GBDT.md" pinned="false" current-in-tab="false">
-        <entry file="file://$PROJECT_DIR$/机器学习/集成学习/GBDT.md">
-          <provider selected="true" editor-type-id="split-provider[text-editor;markdown-preview-editor]">
-            <state split_layout="SPLIT">
-              <first_editor relative-caret-position="720">
-                <caret line="62" column="10" selection-start-line="62" selection-start-column="10" selection-end-line="62" selection-end-column="10" />
-                <folding />
-              </first_editor>
-              <second_editor />
-            </state>
-          </provider>
-          <provider editor-type-id="split-provider[text-editor;MarkdownPreviewEditor]">
-            <state split_layout="SPLIT">
-              <first_editor relative-caret-position="0">
-                <caret line="0" column="0" selection-start-line="0" selection-start-column="0" selection-end-line="0" selection-end-column="0" />
+              <first_editor relative-caret-position="400">
+                <caret line="241" column="25" selection-start-line="241" selection-start-column="25" selection-end-line="241" selection-end-column="25" />
                 <folding />
               </first_editor>
               <second_editor>
@@ -75,8 +52,8 @@
         <entry file="file://$PROJECT_DIR$/机器学习/集成学习/Xgboost.md">
           <provider selected="true" editor-type-id="split-provider[text-editor;markdown-preview-editor]">
             <state split_layout="SPLIT">
-              <first_editor relative-caret-position="495">
-                <caret line="38" column="23" selection-start-line="38" selection-start-column="2" selection-end-line="38" selection-end-column="23" />
+              <first_editor relative-caret-position="445">
+                <caret line="47" column="22" selection-start-line="47" selection-start-column="22" selection-end-line="47" selection-end-column="22" />
                 <folding />
               </first_editor>
               <second_editor />
@@ -137,8 +114,8 @@
         <option value="$PROJECT_DIR$/机器学习/逻辑回归/lr.md" />
         <option value="$PROJECT_DIR$/机器学习/支持向量机/支持向量机.md" />
         <option value="$PROJECT_DIR$/数据预处理/特征提取/数据变换.md" />
-        <option value="$PROJECT_DIR$/机器学习/集成学习/Xgboost.md" />
         <option value="$PROJECT_DIR$/机器学习/集成学习/GBDT.md" />
+        <option value="$PROJECT_DIR$/机器学习/集成学习/Xgboost.md" />
         <option value="$PROJECT_DIR$/README.md" />
       </list>
     </option>
@@ -1409,11 +1386,11 @@
         </state>
       </provider>
     </entry>
-    <entry file="file://$PROJECT_DIR$/机器学习/集成学习/Xgboost.md">
+    <entry file="file://$PROJECT_DIR$/机器学习/集成学习/GBDT.md">
       <provider selected="true" editor-type-id="split-provider[text-editor;markdown-preview-editor]">
         <state split_layout="SPLIT">
-          <first_editor relative-caret-position="495">
-            <caret line="38" column="23" selection-start-line="38" selection-start-column="2" selection-end-line="38" selection-end-column="23" />
+          <first_editor relative-caret-position="-155">
+            <caret line="62" column="10" selection-start-line="62" selection-start-column="10" selection-end-line="62" selection-end-column="10" />
             <folding />
           </first_editor>
           <second_editor />
@@ -1431,11 +1408,11 @@
         </state>
       </provider>
     </entry>
-    <entry file="file://$PROJECT_DIR$/机器学习/集成学习/GBDT.md">
+    <entry file="file://$PROJECT_DIR$/机器学习/集成学习/Xgboost.md">
       <provider selected="true" editor-type-id="split-provider[text-editor;markdown-preview-editor]">
         <state split_layout="SPLIT">
-          <first_editor relative-caret-position="720">
-            <caret line="62" column="10" selection-start-line="62" selection-start-column="10" selection-end-line="62" selection-end-column="10" />
+          <first_editor relative-caret-position="445">
+            <caret line="47" column="22" selection-start-line="47" selection-start-column="22" selection-end-line="47" selection-end-column="22" />
             <folding />
           </first_editor>
           <second_editor />
@@ -1465,8 +1442,8 @@
       </provider>
       <provider selected="true" editor-type-id="split-provider[text-editor;MarkdownPreviewEditor]">
         <state split_layout="SPLIT">
-          <first_editor relative-caret-position="415">
-            <caret line="239" column="47" selection-start-line="239" selection-start-column="47" selection-end-line="239" selection-end-column="47" />
+          <first_editor relative-caret-position="400">
+            <caret line="241" column="25" selection-start-line="241" selection-start-column="25" selection-end-line="241" selection-end-column="25" />
             <folding />
           </first_editor>
           <second_editor>
diff --git a/README.md b/README.md
index 4572273..391382a 100644
--- a/README.md
+++ b/README.md
@@ -233,11 +233,13 @@
     - [gbdt和randomforest区别](机器学习/集成学习/GBDT.md#L164)
     - [GBDT和LR的差异](机器学习/集成学习/GBDT.md#L164)
     - [xgboost对比gbdt/boosting Tree有了哪些方向上的优化](机器学习/集成学习/Xgboost.md#L164)
+    - [xgboost和gbdt的区别](机器学习/集成学习/Xgboost.md#L164)
     - [xgboost优化目标/损失函数改变成什么样](机器学习/集成学习/Xgboost.md#L164)
     - [xgboost如何使用MAE或MAPE作为目标函数](机器学习/集成学习/Xgboost.md#L164)
     - [xgboost如何寻找分裂节点的候选集](机器学习/集成学习/Xgboost.md#L164)
     - [xgboost如何处理缺失值](机器学习/集成学习/Xgboost.md#L164)
     - [xgboost在计算速度上有了哪些点上提升](机器学习/集成学习/Xgboost.md#L164)
+    - [xgboost特征重要性是如何得到的](机器学习/集成学习/Xgboost.md#L164)
 - FM/FFM
 - SVM
     - [简单介绍SVM](机器学习/支持向量机/支持向量机.md#L164)
diff --git "a/\346\234\272\345\231\250\345\255\246\344\271\240/\351\233\206\346\210\220\345\255\246\344\271\240/Xgboost.md" "b/\346\234\272\345\231\250\345\255\246\344\271\240/\351\233\206\346\210\220\345\255\246\344\271\240/Xgboost.md"
index 7d0dbde..109bc35 100644
--- "a/\346\234\272\345\231\250\345\255\246\344\271\240/\351\233\206\346\210\220\345\255\246\344\271\240/Xgboost.md"
+++ "b/\346\234\272\345\231\250\345\255\246\344\271\240/\351\233\206\346\210\220\345\255\246\344\271\240/Xgboost.md"
@@ -8,6 +8,12 @@
     - 数据事先排序并且以block形式存储，有利于并行计算
     - 基于分布式通信框架rabit，可以运行在MPI和yarn上
     - 实现做了面向体系结构的优化，针对cache和内存做了性能优化
+
+# xgboost和gbdt的区别？
+- gbdt用的是cart回归树作为基模型，xgboost还可以用线性模型，加上天生的正则项，就是带L1和L2逻辑回归(分类)和线性回归(回归)
+- gbdt对loss是泰勒一阶展开，xgboost是泰勒二阶展开
+- gbdt没有在loss中带入结点个数和预测值的正则项
+- xgboost在对特征进行了分block预排序，使得在做特征分裂的时候，最终选增益最大的那个特征去做分裂，那么各个特征的增益计算就可以开多线程进行    
     
 # xgboost优化目标/损失函数改变成什么样？
 - 原始：![](https://tva1.sinaimg.cn/large/006y8mN6gy1g94mjezeisj307401fmx0.jpg)
@@ -39,4 +45,9 @@ MAPE:![](https://tva1.sinaimg.cn/large/006y8mN6gy1g94mx7uyuej303f0170sj.jpg)
 # xgboost在计算速度上有了哪些点上提升？
 - 特征预排序
     - 按特征进行存储，每一个block代表一个特征的值，样本在该block中按照它在该特征的值排好序。这些block只需要在程序开始的时候计算一次，后续排序只需要线性扫描这些block即可
-    - block可以仅存放样本的索引，而不是样本本身，这样节省了大量的存储空间
\ No newline at end of file
+    - block可以仅存放样本的索引，而不是样本本身，这样节省了大量的存储空间
+
+# xgboost特征重要性是如何得到的？
+- ’weight‘：代表着某个特征被选作分裂结点的次数；
+- ’gain‘：使用该特征作为分类结点的信息增益；
+- ’cover‘：某特征作为划分结点，覆盖样本总数的平均值；
\ No newline at end of file