Add. Use Regularization

机器学习竞赛实战_优胜解决方案/信用卡欺诈检测/逻辑回归-信用卡欺诈检测.ipynb

@@ -659,6 +659,76 @@
     "print(\"下采样样本总数: \", len(X_train_undersample)+len(X_test_undersample))"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 模型的评估方法——召回率\n",
+    "由于目前正负样本极度不平衡，如果用准确率，那么和上面说的一样，模型把全部评定为正常样本，准确率就达到99.99%。这里用召回率，即异常样本找到多少个。\n",
+    "\n",
+    "Recall = TP/(TP+FN)\n",
+    "\n",
+    "* TP，即 True Positive =正确地判断成正例\n",
+    "* TN，即 True negative=正确地判断成负例\n",
+    "* FP，即False Positive =错误地判断成正例\n",
+    "* FN ，即False negative =错误地判断成负例"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# 在sklearn库的线性模块中，调取“逻辑回归”\n",
+    "from sklearn.linear_model import LogisticRegression  \n",
+    "\n",
+    "#交叉验证模块中，选用“K折交叉验证”\n",
+    "#cross_val_score函数返回的是一个使用交叉验证以后的评分标准。\n",
+    "from sklearn.model_selection import KFold, cross_val_score  \n",
+    "\n",
+    "# 混淆矩阵、召回率\n",
+    "# sklearn中的classification_report函数用于显示主要分类指标的文本报告，在报告中显示每个类的精确度，召回率，F1值等信息。\n",
+    "from sklearn.metrics import confusion_matrix,recall_score,classification_report \n",
+    "\n",
+    "# cross_val_predict 和 cross_val_score的使用方法是一样的，但是它返回的是一个使用交叉验证以后的输出值，而不是评分标准。\n",
+    "from sklearn.model_selection import cross_val_predict"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 正则化惩罚——提高模型泛化能力\n",
+    "模型过拟合通常出现在，数据量少的同时特征又多，也就是当下的情况，异常样本非常少，特征维度有28个。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# 编写Kflod函数——printing_Kfold_scores，实际中我们可以直接调用\n",
+    "def printing_Kfold_scores(x_train_data,y_train_data):\n",
+    "    fold = KFold(len(y_train_data),5,shuffle=False)      #shuffle=False是指数据集不用洗牌\n",
+    " \n",
+    " # 定义不同力度的正则化惩罚力度，越大惩罚力度越大\n",
+    "    c_param_range = [0.01,0.1,1,10,100]                  \n",
+    " # 展示结果用的表格\n",
+    "    results_table = pd.DataFrame(index = range(len(c_param_range),2), columns = ['C_parameter','Mean recall score'])\n",
+    "    results_table['C_parameter'] = c_param_range\n",
+    " \n",
+    " # k-fold 表示K折的交叉验证，这里会得到两个索引集合: 训练集 = indices[0], 验证集 = indices[1]\n",
+    "    j = 0\n",
+    " # 循环遍历不同的参数（这里的c_param_rang是5个——5折交叉验证）\n",
+    "    for c_param in c_param_range:\n",
+    "        print('-------------------------------------------')\n",
+    "        print('正则化惩罚力度: ', c_param)\n",
+    "        print('-------------------------------------------')\n",
+    "        print('\\n')"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,