{"id":803,"date":"2021-09-08T17:07:07","date_gmt":"2021-09-08T09:07:07","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=803"},"modified":"2021-09-08T17:07:07","modified_gmt":"2021-09-08T09:07:07","slug":"%e6%b7%b1%e5%85%a5%e6%8e%a2%e8%ae%a8%ef%bc%9a%e4%b8%ba%e4%bb%80%e4%b9%88%e8%a6%81%e5%81%9a%e7%89%b9%e5%be%81%e5%bd%92%e4%b8%80%e5%8c%96-%e6%a0%87%e5%87%86%e5%8c%96%ef%bc%9f","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2021\/09\/08\/803\/","title":{"rendered":"\u6df1\u5165\u63a2\u8ba8\uff1a\u4e3a\u4ec0\u4e48\u8981\u505a\u7279\u5f81\u5f52\u4e00\u5316\/\u6807\u51c6\u5316\uff1f"},"content":{"rendered":"
\u4f5c\u8005\u4e28shine-lee<\/section>\n
\u6765\u6e90\u4e28https:\/\/blog.csdn.net\/blogshinelee\/article\/details\/102875044<\/section>\n
\u7f16\u8f91\u4e28\u6781\u5e02\u5e73\u53f0<\/section>\n
\n
\n
\u6781\u5e02\u5bfc\u8bfb<\/strong><\/p>\n<\/section>\n
\n
\u672c\u6587\u89e3\u8bfb\u4e86\u4e00\u9879\u6570\u636e\u9884\u5904\u7406\u4e2d\u7684\u91cd\u8981\u6280\u672f\u2014\u2014\u7279\u5f81\u5f52\u4e00\u5316\uff0c\u63d0\u51fa\u5e76\u89e3\u7b54\u4e865\u4e2a\u76f8\u5173\u95ee\u9898\uff0c\u540c\u65f6\u5206\u6790\u4e86\u76f8\u5173\u65b9\u6cd5\u548c\u9002\u7528\u573a\u666f\u3002<\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u5199\u5728\u524d\u9762<\/h1>\n
Feature scaling<\/strong>\uff0c\u5e38\u89c1\u7684\u63d0\u6cd5\u6709\u201c\u7279\u5f81\u5f52\u4e00\u5316\u201d\u3001\u201c\u6807\u51c6\u5316\u201d\uff0c\u662f\u6570\u636e\u9884\u5904\u7406\u4e2d\u7684\u91cd\u8981\u6280\u672f\uff0c\u6709\u65f6\u751a\u81f3\u51b3\u5b9a\u4e86\u7b97\u6cd5\u80fd\u4e0d\u80fdwork\u4ee5\u53cawork\u5f97\u597d\u4e0d\u597d\u3002\u8c08\u5230feature scaling\u7684\u5fc5\u8981\u6027\uff0c\u6700\u5e38\u7528\u76842\u4e2a\u4f8b\u5b50\u53ef\u80fd\u662f\uff1a<\/section>\n
    \n
  • \n
    \u7279\u5f81\u95f4\u7684\u5355\u4f4d\uff08\u5c3a\u5ea6\uff09\u53ef\u80fd\u4e0d\u540c<\/strong>\uff0c\u6bd4\u5982\u8eab\u9ad8\u548c\u4f53\u91cd\uff0c\u6bd4\u5982\u6444\u6c0f\u5ea6\u548c\u534e\u6c0f\u5ea6\uff0c\u6bd4\u5982\u623f\u5c4b\u9762\u79ef\u548c\u623f\u95f4\u6570\uff0c\u4e00\u4e2a\u7279\u5f81\u7684\u53d8\u5316\u8303\u56f4\u53ef\u80fd\u662f[1000, 10000]\uff0c\u53e6\u4e00\u4e2a\u7279\u5f81\u7684\u53d8\u5316\u8303\u56f4\u53ef\u80fd\u662f[\u22120.1,0.2]\uff0c\u5728\u8fdb\u884c\u8ddd\u79bb\u6709\u5173\u7684\u8ba1\u7b97\u65f6\uff0c\u5355\u4f4d\u7684\u4e0d\u540c\u4f1a\u5bfc\u81f4\u8ba1\u7b97\u7ed3\u679c\u7684\u4e0d\u540c\uff0c\u5c3a\u5ea6\u5927\u7684\u7279\u5f81\u4f1a\u8d77\u51b3\u5b9a\u6027\u4f5c\u7528\uff0c\u800c\u5c3a\u5ea6\u5c0f\u7684\u7279\u5f81\u5176\u4f5c\u7528\u53ef\u80fd\u4f1a\u88ab\u5ffd\u7565\uff0c\u4e3a\u4e86\u6d88\u9664\u7279\u5f81\u95f4\u5355\u4f4d\u548c\u5c3a\u5ea6\u5dee\u5f02\u7684\u5f71\u54cd\uff0c\u4ee5\u5bf9\u6bcf\u7ef4\u7279\u5f81\u540c\u7b49\u770b\u5f85\uff0c\u9700\u8981\u5bf9\u7279\u5f81\u8fdb\u884c\u5f52\u4e00\u5316<\/strong>\u3002<\/p>\n<\/section>\n<\/li>\n
  • \n
    \u539f\u59cb\u7279\u5f81\u4e0b\uff0c\u56e0\u5c3a\u5ea6\u5dee\u5f02\uff0c\u5176\u635f\u5931\u51fd\u6570\u7684\u7b49\u9ad8\u7ebf\u56fe\u53ef\u80fd\u662f\u692d\u5706\u5f62<\/strong>\uff0c\u68af\u5ea6\u65b9\u5411\u5782\u76f4\u4e8e\u7b49\u9ad8\u7ebf\uff0c\u4e0b\u964d\u4f1a\u8d70zigzag\u8def\u7ebf\uff0c\u800c\u4e0d\u662f\u6307\u5411local minimum\u3002\u901a\u8fc7\u5bf9\u7279\u5f81\u8fdb\u884czero-mean and unit-variance\u53d8\u6362\u540e\uff0c\u5176\u635f\u5931\u51fd\u6570\u7684\u7b49\u9ad8\u7ebf\u56fe\u66f4\u63a5\u8fd1\u5706\u5f62\uff0c\u68af\u5ea6\u4e0b\u964d\u7684\u65b9\u5411\u9707\u8361\u66f4\u5c0f\uff0c\u6536\u655b\u66f4\u5feb\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u56fe\u7247\u6765\u81eaAndrew Ng\u3002<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
    \"\u56fe\u7247\"<\/section>\n
    \n
    Feature Scaling from Andrew Ng<\/figcaption><\/figure>\n<\/section>\n
    <\/section>\n
    \u5bf9\u4e8efeature scaling\u4e2d\u6700\u5e38\u4f7f\u7528\u7684Standardization\uff0c\u4f3c\u4e4e\u201c\u65e0\u8111\u4e0a\u201d\u5c31\u884c\u4e86\uff0c\u672c\u6587\u60f3\u591a\u63a2\u7a76\u4e00\u4e9b\u4e3a\u4ec0\u4e48\uff0c<\/section>\n
      \n
    • \n
      \u5e38\u7528\u7684feature scaling\u65b9\u6cd5\u90fd\u6709\u54ea\u4e9b\uff1f<\/section>\n<\/li>\n
    • \n
      \u4ec0\u4e48\u60c5\u51b5\u4e0b\u8be5\u4f7f\u7528\u4ec0\u4e48feature scaling\u65b9\u6cd5\uff1f\u6709\u6ca1\u6709\u4e00\u4e9b\u6307\u5bfc\u601d\u60f3\uff1f<\/section>\n<\/li>\n
    • \n
      \u6240\u6709\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u90fd\u9700\u8981feature scaling\u5417\uff1f\u6709\u6ca1\u6709\u4f8b\u5916\uff1f<\/section>\n<\/li>\n
    • \n
      \u635f\u5931\u51fd\u6570\u7684\u7b49\u9ad8\u7ebf\u56fe\u90fd\u662f\u692d\u5706\u6216\u540c\u5fc3\u5706\u5417\uff1f\u80fd\u7528\u692d\u5706\u548c\u5706\u6765\u7b80\u5355\u89e3\u91cafeature scaling\u7684\u4f5c\u7528\u5417\uff1f<\/section>\n<\/li>\n
    • \n
      \u5982\u679c\u635f\u5931\u51fd\u6570\u7684\u7b49\u9ad8\u7ebf\u56fe\u5f88\u590d\u6742\uff0cfeature scaling\u8fd8\u6709\u5176\u4ed6\u76f4\u89c2\u89e3\u91ca\u5417\uff1f<\/section>\n<\/li>\n<\/ul>\n
      \u6839\u636e\u67e5\u9605\u5230\u7684\u8d44\u6599\uff0c\u672c\u6587\u5c06\u5c1d\u8bd5\u56de\u7b54\u4e0a\u9762\u7684\u95ee\u9898\u3002\u4f46\u7b14\u8005\u80fd\u529b\u6709\u9650\uff0c\u7a7a\u6709\u56f0\u60d1\uff0c\u80fd\u8bb2\u5230\u54ea\u7b97\u54ea\u5427\uff08\u5fae\u7b11\uff09\u3002<\/section>\n

      \u5e38\u7528feature scaling\u65b9\u6cd5<\/h1>\n
      \u5728\u95ee\u4e3a\u4ec0\u4e48\u524d\uff0c\u5148\u770b\u662f\u4ec0\u4e48\u3002<\/section>\n
      \u7ed9\u5b9a\u6570\u636e\u96c6\uff0c\u4ee4\u7279\u5f81\u5411\u91cf\u4e3ax\uff0c\u7ef4\u6570\u4e3aD\uff0c\u6837\u672c\u6570\u91cf\u4e3aR\uff0c\u53ef\u6784\u6210D\u00d7R\u7684\u77e9\u9635\uff0c\u4e00\u5217\u4e3a\u4e00\u4e2a\u6837\u672c\uff0c\u4e00\u884c\u4e3a\u4e00\u7ef4\u7279\u5f81\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u56fe\u7247\u6765\u81eaHung-yi Lee pdf-Gradient Descent\uff1a<\/section>\n
      \"\u56fe\u7247\"<\/section>\n
      \n
      feature matrix<\/figcaption><\/figure>\n<\/section>\n
      feature scaling\u7684\u65b9\u6cd5\u53ef\u4ee5\u5206\u62102\u7c7b\uff0c\u9010\u884c\u8fdb\u884c\u548c\u9010\u5217\u8fdb\u884c\u3002\u9010\u884c\u662f\u5bf9\u6bcf\u4e00\u7ef4\u7279\u5f81\u64cd\u4f5c\uff0c\u9010\u5217\u662f\u5bf9\u6bcf\u4e2a\u6837\u672c\u64cd\u4f5c\uff0c\u4e0a\u56fe\u4e3a\u9010\u884c\u64cd\u4f5c\u4e2d\u7279\u5f81\u6807\u51c6\u5316\u7684\u793a\u4f8b\u3002<\/section>\n
      \u5177\u4f53\u5730\uff0c\u5e38\u7528feature scaling\u65b9\u6cd5\u5982\u4e0b\uff0c\u6765\u81eawiki\uff0c<\/section>\n
        \n
      • \n
        Rescaling (min-max normalization\u3001range scaling)<\/strong>\uff1a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
        \"\u56fe\u7247\"<\/section>\n
        \u5c06\u6bcf\u4e00\u7ef4\u7279\u5f81\u7ebf\u6027\u6620\u5c04\u5230\u76ee\u6807\u8303\u56f4[a,b]\uff0c\u5373\u5c06\u6700\u5c0f\u503c\u6620\u5c04\u4e3aa\uff0c\u6700\u5927\u503c\u6620\u5c04\u4e3ab\uff0c\u5e38\u7528\u76ee\u6807\u8303\u56f4\u4e3a[0,1]\u548c[\u22121,1]\uff0c\u7279\u522b\u5730\uff0c\u6620\u5c04\u5230[0,1]\u8ba1\u7b97\u65b9\u5f0f\u4e3a\uff1a<\/section>\n
        \"\u56fe\u7247\"<\/section>\n
          \n
        • \n
          Mean normalization<\/strong>\uff1a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
          \"\u56fe\u7247\"<\/section>\n
          \u5c06\u5747\u503c\u6620\u5c04\u4e3a0<\/strong>\uff0c\u540c\u65f6\u7528\u6700\u5927\u503c\u6700\u5c0f\u503c\u7684\u5dee\u5bf9\u7279\u5f81\u8fdb\u884c\u5f52\u4e00\u5316\uff0c\u4e00\u79cd\u66f4\u5e38\u89c1\u7684\u505a\u6cd5\u662f\u7528\u6807\u51c6\u5dee\u8fdb\u884c\u5f52\u4e00\u5316\uff0c\u5982\u4e0b\u3002<\/section>\n
            \n
          • \n
            Standardization (Z-score Normalization)<\/strong>\uff1a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
            \"\u56fe\u7247\"<\/section>\n
            \u6bcf\u7ef4\u7279\u5f810\u5747\u503c1\u65b9\u5dee\uff08zero-mean and unit-variance\uff09<\/strong>\u3002<\/section>\n
              \n
            • \n
              Scaling to unit length<\/strong>\uff1a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
              \"\u56fe\u7247\"<\/section>\n
              \u5c06\u6bcf\u4e2a\u6837\u672c\u7684\u7279\u5f81\u5411\u91cf\u9664\u4ee5\u5176\u957f\u5ea6\uff0c\u5373\u5bf9\u6837\u672c\u7279\u5f81\u5411\u91cf\u7684\u957f\u5ea6\u8fdb\u884c\u5f52\u4e00\u5316\uff0c\u957f\u5ea6\u7684\u5ea6\u91cf\u5e38\u4f7f\u7528\u7684\u662fL2 norm\uff08\u6b27\u6c0f\u8ddd\u79bb\uff09\uff0c\u6709\u65f6\u4e5f\u4f1a\u91c7\u7528L1 norm\uff0c\u4e0d\u540c\u5ea6\u91cf\u65b9\u5f0f\u7684\u4e00\u79cd\u5bf9\u6bd4\u53ef\u4ee5\u53c2\u89c1\u8bba\u6587\u201cCVPR2005-Histograms of Oriented Gradients for Human Detection\u201d\u3002<\/section>\n
              \u4e0a\u8ff04\u79cdfeature scaling\u65b9\u5f0f\uff0c\u524d3\u79cd\u4e3a\u9010\u884c\u64cd\u4f5c\uff0c\u6700\u540e1\u79cd\u4e3a\u9010\u5217\u64cd\u4f5c\u3002<\/section>\n
              \u5bb9\u6613\u8ba9\u4eba\u56f0\u60d1\u7684\u4e00\u70b9\u662f\u6307\u4ee3\u6df7\u6dc6\uff0cStandardization\u6307\u4ee3\u6bd4\u8f83\u6e05\u6670\uff0c\u4f46\u662f\u5355\u8bf4Normalization\u6709\u65f6\u4f1a\u6307\u4ee3min-max normalization\uff0c\u6709\u65f6\u4f1a\u6307\u4ee3Standardization\uff0c\u6709\u65f6\u4f1a\u6307\u4ee3Scaling to unit length<\/strong>\u3002<\/section>\n

              \u8ba1\u7b97\u65b9\u5f0f\u4e0a\u5bf9\u6bd4\u5206\u6790<\/h1>\n
              \u524d3\u79cdfeature scaling\u7684\u8ba1\u7b97\u65b9\u5f0f\u4e3a\u51cf\u4e00\u4e2a\u7edf\u8ba1\u91cf\u518d\u9664\u4ee5\u4e00\u4e2a\u7edf\u8ba1\u91cf<\/strong>\uff0c\u6700\u540e1\u79cd\u4e3a\u9664\u4ee5\u5411\u91cf\u81ea\u8eab\u7684\u957f\u5ea6<\/strong>\u3002<\/section>\n
                \n
              • \n
                \u51cf\u4e00\u4e2a\u7edf\u8ba1\u91cf<\/strong>\u53ef\u4ee5\u770b\u6210\u9009\u54ea\u4e2a\u503c\u4f5c\u4e3a\u539f\u70b9\uff0c\u662f\u6700\u5c0f\u503c\u8fd8\u662f\u5747\u503c\uff0c\u5e76\u5c06\u6574\u4e2a\u6570\u636e\u96c6\u5e73\u79fb\u5230\u8fd9\u4e2a\u65b0\u7684\u539f\u70b9\u4f4d\u7f6e<\/strong>\u3002\u5982\u679c\u7279\u5f81\u95f4\u504f\u7f6e\u4e0d\u540c\u5bf9\u540e\u7eed\u8fc7\u7a0b\u6709\u8d1f\u9762\u5f71\u54cd\uff0c\u5219\u8be5\u64cd\u4f5c\u662f\u6709\u76ca\u7684\uff0c\u53ef\u4ee5\u770b\u6210\u662f\u67d0\u79cd\u504f\u7f6e\u65e0\u5173\u64cd\u4f5c<\/strong>\uff1b\u5982\u679c\u539f\u59cb\u7279\u5f81\u503c\u6709\u7279\u6b8a\u610f\u4e49\uff0c\u6bd4\u5982\u7a00\u758f\u6027\uff0c\u8be5\u64cd\u4f5c\u53ef\u80fd\u4f1a\u7834\u574f\u5176\u7a00\u758f\u6027\u3002<\/section>\n<\/li>\n
              • \n
                \u9664\u4ee5\u4e00\u4e2a\u7edf\u8ba1\u91cf<\/strong>\u53ef\u4ee5\u770b\u6210\u5728\u5750\u6807\u8f74\u65b9\u5411\u4e0a\u5bf9\u7279\u5f81\u8fdb\u884c\u7f29\u653e<\/strong>\uff0c\u7528\u4e8e\u964d\u4f4e\u7279\u5f81\u5c3a\u5ea6\u7684\u5f71\u54cd\uff0c\u53ef\u4ee5\u770b\u6210\u662f\u67d0\u79cd\u5c3a\u5ea6\u65e0\u5173\u64cd\u4f5c<\/strong>\u3002\u7f29\u653e\u53ef\u4ee5\u4f7f\u7528\u6700\u5927\u503c\u6700\u5c0f\u503c\u95f4\u7684\u8de8\u5ea6\uff0c\u4e5f\u53ef\u4ee5\u4f7f\u7528\u6807\u51c6\u5dee\uff08\u5230\u4e2d\u5fc3\u70b9\u7684\u5e73\u5747\u8ddd\u79bb\uff09\uff0c\u524d\u8005\u5bf9outliers\u654f\u611f\uff0coutliers\u5bf9\u540e\u8005\u5f71\u54cd\u4e0eoutliers\u6570\u91cf\u548c\u6570\u636e\u96c6\u5927\u5c0f\u6709\u5173\uff0coutliers\u8d8a\u5c11\u6570\u636e\u96c6\u8d8a\u5927\u5f71\u54cd\u8d8a\u5c0f\u3002<\/section>\n<\/li>\n
              • \n
                \u9664\u4ee5\u957f\u5ea6<\/strong>\u76f8\u5f53\u4e8e\u628a\u957f\u5ea6\u5f52\u4e00\u5316\uff0c\u628a\u6240\u6709\u6837\u672c\u6620\u5c04\u5230\u5355\u4f4d\u7403\u4e0a<\/strong>\uff0c\u53ef\u4ee5\u770b\u6210\u662f\u67d0\u79cd\u957f\u5ea6\u65e0\u5173\u64cd\u4f5c<\/strong>\uff0c\u6bd4\u5982\uff0c\u8bcd\u9891\u7279\u5f81\u8981\u79fb\u9664\u6587\u7ae0\u957f\u5ea6\u7684\u5f71\u54cd\uff0c\u56fe\u50cf\u5904\u7406\u4e2d\u67d0\u4e9b\u7279\u5f81\u8981\u79fb\u9664\u5149\u7167\u5f3a\u5ea6\u7684\u5f71\u54cd\uff0c\u4ee5\u53ca\u65b9\u4fbf\u8ba1\u7b97\u4f59\u5f26\u8ddd\u79bb\u6216\u5185\u79ef\u76f8\u4f3c\u5ea6\u7b49\u3002<\/section>\n<\/li>\n<\/ul>\n
                \u7a00\u758f\u6570\u636e\u3001outliers\u76f8\u5173\u7684\u66f4\u591a\u6570\u636e\u9884\u5904\u7406\u5185\u5bb9\u53ef\u4ee5\u53c2\u89c1scikit learn-5.3. Preprocessing data\u3002<\/section>\n
                \u4ece\u51e0\u4f55\u4e0a\u89c2\u5bdf\u4e0a\u8ff0\u65b9\u6cd5\u7684\u4f5c\u7528\uff0c\u56fe\u7247\u6765\u81eaCS231n-Neural Networks Part 2: Setting up the Data and the Loss\uff0czero-mean\u5c06\u6570\u636e\u96c6\u5e73\u79fb\u5230\u539f\u70b9\uff0cunit-variance\u4f7f\u6bcf\u7ef4\u7279\u5f81\u4e0a\u7684\u8de8\u5ea6\u76f8\u5f53\uff0c\u56fe\u4e2d\u53ef\u4ee5\u660e\u663e\u770b\u51fa\u4e24\u7ef4\u7279\u5f81\u95f4\u5b58\u5728\u7ebf\u6027\u76f8\u5173\u6027\uff0cStandardization\u64cd\u4f5c\u5e76\u6ca1\u6709\u6d88\u9664\u8fd9\u79cd\u76f8\u5173\u6027\u3002<\/section>\n
                \n
                \"\u56fe\u7247\"
                Standardization<\/figcaption><\/figure>\n<\/section>\n
                \u53ef\u901a\u8fc7PCA\u65b9\u6cd5\u79fb\u9664\u7ebf\u6027\u76f8\u5173\u6027\uff08decorrelation\uff09\uff0c\u5373\u5f15\u5165\u65cb\u8f6c\uff0c\u627e\u5230\u65b0\u7684\u5750\u6807\u8f74\u65b9\u5411\uff0c\u5728\u65b0\u5750\u6807\u8f74\u65b9\u5411\u4e0a\u7528\u201c\u6807\u51c6\u5dee\u201d\u8fdb\u884c\u7f29\u653e\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u56fe\u7247\u6765\u81ea\u94fe\u63a5\uff0c\u56fe\u4e2d\u540c\u65f6\u63cf\u8ff0\u4e86unit length\u7684\u4f5c\u7528\u2014\u2014\u5c06\u6240\u6709\u6837\u672c\u6620\u5c04\u5230\u5355\u4f4d\u7403\u4e0a\u3002<\/section>\n
                \n
                \"\u56fe\u7247\"
                Effect of the operations of standardization and length normalization<\/figcaption><\/figure>\n<\/section>\n
                \u5f53\u7279\u5f81\u7ef4\u6570\u66f4\u591a\u65f6\uff0c\u5bf9\u6bd4\u5982\u4e0b\uff0c\u56fe\u7247\u6765\u81eayoutube\uff0c<\/section>\n
                \"\u56fe\u7247\"<\/section>\n
                \n
                feature scaling comparison<\/figcaption><\/figure>\n<\/section>\n
                \u603b\u7684\u6765\u8bf4\uff0c\u5f52\u4e00\u5316\/\u6807\u51c6\u5316\u7684\u76ee\u7684\u662f\u4e3a\u4e86\u83b7\u5f97\u67d0\u79cd\u201c\u65e0\u5173\u6027\u201d\u2014\u2014\u504f\u7f6e\u65e0\u5173\u3001\u5c3a\u5ea6\u65e0\u5173\u3001\u957f\u5ea6\u65e0\u5173\u2026\u2026\u5f53\u5f52\u4e00\u5316\/\u6807\u51c6\u5316\u65b9\u6cd5\u80cc\u540e\u7684\u7269\u7406\u610f\u4e49\u548c\u51e0\u4f55\u542b\u4e49\u4e0e\u5f53\u524d\u95ee\u9898\u7684\u9700\u8981\u76f8\u5951\u5408\u65f6\uff0c\u5176\u5bf9\u89e3\u51b3\u8be5\u95ee\u9898\u5c31\u6709\u6b63\u5411\u4f5c\u7528\uff0c\u53cd\u4e4b\uff0c\u5c31\u4f1a\u8d77\u53cd\u4f5c\u7528\u3002\u6240\u4ee5\uff0c\u201c\u4f55\u65f6\u9009\u62e9\u4f55\u79cd\u65b9\u6cd5\u201d\u53d6\u51b3\u4e8e\u5f85\u89e3\u51b3\u7684\u95ee\u9898\uff0c\u5373problem-dependent\u3002<\/strong><\/section>\n

                feature scaling \u9700\u8981\u8fd8\u662f\u4e0d\u9700\u8981<\/h1>\n
                \u4e0b\u56fe\u6765\u81eadata school-Comparing supervised learning algorithms\uff0c\u5bf9\u6bd4\u4e86\u51e0\u4e2a\u76d1\u7763\u5b66\u4e60\u7b97\u6cd5\uff0c\u6700\u53f3\u4fa7\u4e24\u5217\u4e3a\u662f\u5426\u9700\u8981feature scaling\u3002<\/section>\n
                \n
                \"\u56fe\u7247\"
                Comparing supervised learning algorithms<\/figcaption><\/figure>\n<\/section>\n
                \u4e0b\u9762\u5177\u4f53\u5206\u6790\u4e00\u4e0b\u3002<\/section>\n

                \u4ec0\u4e48\u65f6\u5019\u9700\u8981feature scaling\uff1f<\/h2>\n
                  \n
                • \n
                  \u6d89\u53ca\u6216\u9690\u542b\u8ddd\u79bb\u8ba1\u7b97<\/strong>\u7684\u7b97\u6cd5\uff0c\u6bd4\u5982K-means\u3001KNN\u3001PCA\u3001SVM\u7b49\uff0c\u4e00\u822c\u9700\u8981feature scaling\uff0c\u56e0\u4e3a\uff1a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
                  zero-mean\u4e00\u822c\u53ef\u4ee5\u589e\u52a0\u6837\u672c\u95f4\u4f59\u5f26\u8ddd\u79bb\u6216\u8005\u5185\u79ef\u7ed3\u679c\u7684\u5dee\u5f02<\/strong>\uff0c\u533a\u5206\u529b\u66f4\u5f3a\uff0c\u5047\u8bbe\u6570\u636e\u96c6\u96c6\u4e2d\u5206\u5e03\u5728\u7b2c\u4e00\u8c61\u9650\u9065\u8fdc\u7684\u53f3\u4e0a\u89d2\uff0c\u5c06\u5176\u5e73\u79fb\u5230\u539f\u70b9\u5904\uff0c\u53ef\u4ee5\u60f3\u8c61\u6837\u672c\u95f4\u4f59\u5f26\u8ddd\u79bb\u7684\u5dee\u5f02\u88ab\u653e\u5927\u4e86\u3002\u5728\u6a21\u7248\u5339\u914d\u4e2d\uff0czero-mean\u53ef\u4ee5\u660e\u663e\u63d0\u9ad8\u54cd\u5e94\u7ed3\u679c\u7684\u533a\u5206\u5ea6\u3002<\/section>\n
                  \u5c31\u6b27\u5f0f\u8ddd\u79bb\u800c\u8a00\uff0c\u589e\u5927\u67d0\u4e2a\u7279\u5f81\u7684\u5c3a\u5ea6\uff0c\u76f8\u5f53\u4e8e\u589e\u52a0\u4e86\u5176\u5728\u8ddd\u79bb\u8ba1\u7b97\u4e2d\u7684\u6743\u91cd\uff0c\u5982\u679c\u6709\u660e\u786e\u7684\u5148\u9a8c\u77e5\u8bc6\u8868\u660e\u67d0\u4e2a\u7279\u5f81\u5f88\u91cd\u8981\uff0c\u90a3\u4e48\u9002\u5f53\u589e\u52a0\u5176\u6743\u91cd\u53ef\u80fd\u6709\u6b63\u5411\u6548\u679c\uff0c\u4f46\u5982\u679c\u6ca1\u6709\u8fd9\u6837\u7684\u5148\u9a8c\uff0c\u6216\u8005\u76ee\u7684\u5c31\u662f\u60f3\u77e5\u9053\u54ea\u4e9b\u7279\u5f81\u66f4\u91cd\u8981\uff0c\u90a3\u4e48\u5c31\u9700\u8981\u5148feature scaling\uff0c\u5bf9\u5404\u7ef4\u7279\u5f81\u7b49\u800c\u89c6\u4e4b<\/strong>\u3002<\/section>\n
                  \u589e\u5927\u5c3a\u5ea6\u7684\u540c\u65f6\u4e5f\u589e\u5927\u4e86\u8be5\u7279\u5f81\u7ef4\u5ea6\u4e0a\u7684\u65b9\u5dee\uff0cPCA\u7b97\u6cd5\u503e\u5411\u4e8e\u5173\u6ce8\u65b9\u5dee\u8f83\u5927\u7684\u7279\u5f81\u6240\u5728\u7684\u5750\u6807\u8f74\u65b9\u5411\uff0c\u5176\u4ed6\u7279\u5f81\u53ef\u80fd\u4f1a\u88ab\u5ffd\u89c6\uff0c\u56e0\u6b64\uff0c\u5728PCA\u524d\u505aStandardization\u6548\u679c\u53ef\u80fd\u66f4\u597d\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u56fe\u7247\u6765\u81eascikit learn-Importance of Feature Scaling\uff0c<\/section>\n
                  \n
                  \"\u56fe\u7247\"
                  PCA and Standardization<\/figcaption><\/figure>\n<\/section>\n
                    \n
                  • \n
                    \u635f\u5931\u51fd\u6570\u4e2d\u542b\u6709\u6b63\u5219\u9879<\/strong>\u65f6\uff0c\u4e00\u822c\u9700\u8981feature scaling\uff1a\u5bf9\u4e8e\u7ebf\u6027\u6a21\u578by=wx+b\u800c\u8a00\uff0cx\u7684\u4efb\u4f55\u7ebf\u6027\u53d8\u6362\uff08\u5e73\u79fb\u3001\u653e\u7f29\uff09\uff0c\u90fd\u53ef\u4ee5\u88abw\u548cb\u201c\u5438\u6536\u201d\u6389\uff0c\u7406\u8bba\u4e0a\uff0c\u4e0d\u4f1a\u5f71\u54cd\u6a21\u578b\u7684\u62df\u5408\u80fd\u529b\u3002\u4f46\u662f\uff0c\u5982\u679c\u635f\u5931\u51fd\u6570\u4e2d\u542b\u6709\u6b63\u5219\u9879\uff0c\u5982\u03bb\u2223\u2223w\u2223\u2223^2\uff0c\u03bb\u4e3a\u8d85\u53c2\u6570\uff0c\u5176\u5bf9w\u7684\u6bcf\u4e00\u4e2a\u53c2\u6570\u65bd\u52a0\u540c\u6837\u7684\u60e9\u7f5a\uff0c\u4f46\u5bf9\u4e8e\u67d0\u4e00\u7ef4\u7279\u5f81xi\u800c\u8a00\uff0c\u5176scale\u8d8a\u5927\uff0c\u7cfb\u6570wi\u8d8a\u5c0f\uff0c\u5176\u5728\u6b63\u5219\u9879\u4e2d\u7684\u6bd4\u91cd\u5c31\u4f1a\u53d8\u5c0f\uff0c\u76f8\u5f53\u4e8e\u5bf9wi\u60e9\u7f5a\u53d8\u5c0f\uff0c\u5373\u635f\u5931\u51fd\u6570\u4f1a\u76f8\u5bf9\u5ffd\u89c6\u90a3\u4e9bscale\u589e\u5927\u7684\u7279\u5f81\uff0c\u8fd9\u5e76\u4e0d\u5408\u7406\uff0c\u6240\u4ee5\u9700\u8981feature scaling\uff0c\u4f7f\u635f\u5931\u51fd\u6570\u5e73\u7b49\u770b\u5f85\u6bcf\u4e00\u7ef4\u7279\u5f81\u3002<\/p>\n<\/section>\n<\/li>\n
                  • \n
                    \u68af\u5ea6\u4e0b\u964d\u7b97\u6cd5\uff0c\u9700\u8981feature scaling<\/strong>\u3002\u68af\u5ea6\u4e0b\u964d\u7684\u53c2\u6570\u66f4\u65b0\u516c\u5f0f\u5982\u4e0b\uff0c<\/p>\n<\/section>\n<\/li>\n<\/ul>\n
                    \"\u56fe\u7247\"<\/section>\n
                    E(W)\u4e3a\u635f\u5931\u51fd\u6570\uff0c\u6536\u655b\u901f\u5ea6\u53d6\u51b3\u4e8e\uff1a\u53c2\u6570\u7684\u521d\u59cb\u4f4d\u7f6e\u5230local minima\u7684\u8ddd\u79bb\uff0c\u4ee5\u53ca\u5b66\u4e60\u7387\u03b7\u7684\u5927\u5c0f<\/strong>\u3002\u4e00\u7ef4\u60c5\u51b5\u4e0b\uff0c\u5728local minima\u9644\u8fd1<\/strong>\uff0c\u4e0d\u540c\u5b66\u4e60\u7387\u5bf9\u68af\u5ea6\u4e0b\u964d\u7684\u5f71\u54cd\u5982\u4e0b\u56fe\u6240\u793a\uff1a<\/section>\n
                    \n
                    \"\u56fe\u7247\"
                    Gradient descent for different learning rates<\/figcaption><\/figure>\n<\/section>\n
                    \u591a\u7ef4\u60c5\u51b5\u4e0b\u53ef\u4ee5\u5206\u89e3\u6210\u591a\u4e2a\u4e0a\u56fe\uff0c\u6bcf\u4e2a\u7ef4\u5ea6\u4e0a\u5206\u522b\u4e0b\u964d\uff0c\u53c2\u6570W\u4e3a\u5411\u91cf\uff0c\u4f46\u5b66\u4e60\u7387\u53ea\u67091\u4e2a\uff0c\u5373\u6240\u6709\u53c2\u6570\u7ef4\u5ea6\u5171\u7528\u540c\u4e00\u4e2a\u5b66\u4e60\u7387\uff08\u6682\u4e0d\u8003\u8651\u4e3a\u6bcf\u4e2a\u7ef4\u5ea6\u90fd\u5206\u914d\u5355\u72ec\u5b66\u4e60\u7387\u7684\u7b97\u6cd5\uff09\u3002\u6536\u655b\u610f\u5473\u7740\u5728\u6bcf\u4e2a\u53c2\u6570\u7ef4\u5ea6\u4e0a\u90fd\u53d6\u5f97\u6781\u5c0f\u503c\uff0c\u6bcf\u4e2a\u53c2\u6570\u7ef4\u5ea6\u4e0a\u7684\u504f\u5bfc\u6570\u90fd\u4e3a0\uff0c\u4f46\u662f\u6bcf\u4e2a\u53c2\u6570\u7ef4\u5ea6\u4e0a\u7684\u4e0b\u964d\u901f\u5ea6\u662f\u4e0d\u540c\u7684\uff0c\u4e3a\u4e86\u6bcf\u4e2a\u7ef4\u5ea6\u4e0a\u90fd\u80fd\u6536\u655b\uff0c\u5b66\u4e60\u7387\u5e94\u53d6\u6240\u6709\u7ef4\u5ea6\u5728\u5f53\u524d\u4f4d\u7f6e\u5408\u9002\u6b65\u957f\u4e2d\u6700\u5c0f\u7684\u90a3\u4e2a\u3002\u4e0b\u9762\u8ba8\u8bbafeature scaling\u5bf9gradient descent\u7684\u4f5c\u7528\uff0c<\/section>\n
                      \n
                    • \n
                        \n
                      • \n
                          \n
                        • \n
                          zero center\u4e0e\u53c2\u6570\u521d\u59cb\u5316\u76f8\u914d\u5408\uff0c\u7f29\u77ed\u521d\u59cb\u53c2\u6570\u4f4d\u7f6e\u4e0elocal minimum\u95f4\u7684\u8ddd\u79bb\uff0c\u52a0\u5feb\u6536\u655b<\/strong>\u3002\u6a21\u578b\u7684\u6700\u7ec8\u53c2\u6570\u662f\u672a\u77e5\u7684\uff0c\u6240\u4ee5\u4e00\u822c\u968f\u673a\u521d\u59cb\u5316\uff0c\u6bd4\u5982\u4ece0\u5747\u503c\u7684\u5747\u5300\u5206\u5e03\u6216\u9ad8\u65af\u5206\u5e03\u4e2d\u91c7\u6837\u5f97\u5230\uff0c\u5bf9\u7ebf\u6027\u6a21\u578b\u800c\u8a00\uff0c\u5176\u5206\u754c\u9762\u521d\u59cb\u4f4d\u7f6e\u5927\u81f4\u5728\u539f\u70b9\u9644\u8fd1\uff0cbias\u7ecf\u5e38\u521d\u59cb\u5316\u4e3a0\uff0c\u5219\u5206\u754c\u9762\u76f4\u63a5\u901a\u8fc7\u539f\u70b9\u3002\u540c\u65f6\uff0c\u4e3a\u4e86\u6536\u655b\uff0c\u5b66\u4e60\u7387\u4e0d\u4f1a\u5f88\u5927\u3002\u800c\u6bcf\u4e2a\u6570\u636e\u96c6\u7684\u7279\u5f81\u5206\u5e03\u662f\u4e0d\u4e00\u6837\u7684\uff0c\u5982\u679c\u5176\u5206\u5e03\u96c6\u4e2d\u4e14\u8ddd\u79bb\u539f\u70b9\u8f83\u8fdc\uff0c\u6bd4\u5982\u4f4d\u4e8e\u7b2c\u4e00\u8c61\u9650\u9065\u8fdc\u7684\u53f3\u4e0a\u89d2\uff0c\u5206\u754c\u9762\u53ef\u80fd\u9700\u8981\u82b1\u8d39\u5f88\u591a\u6b65\u9aa4\u624d\u80fd\u201c\u722c\u5230\u201d\u6570\u636e\u96c6\u6240\u5728\u7684\u4f4d\u7f6e\u3002\u6240\u4ee5\uff0c\u65e0\u8bba\u4ec0\u4e48\u6570\u636e\u96c6\uff0c\u5148\u5e73\u79fb\u5230\u539f\u70b9\uff0c\u518d\u914d\u5408\u53c2\u6570\u521d\u59cb\u5316\uff0c\u53ef\u4ee5\u4fdd\u8bc1\u5206\u754c\u9762\u4e00\u5b9a\u4f1a\u7a7f\u8fc7\u6570\u636e\u96c6\u3002\u6b64\u5916\uff0coutliers\u5e38\u5206\u5e03\u5728\u6570\u636e\u96c6\u7684\u5916\u56f4<\/strong>\uff0c\u4e0e\u5206\u754c\u9762\u4ece\u5916\u90e8\u5411\u5185\u632a\u52a8\u76f8\u6bd4\uff0c\u4ece\u4e2d\u5fc3\u533a\u57df\u5f00\u59cb\u632a\u52a8\u53ef\u80fd\u53d7outliers\u7684\u5f71\u54cd\u66f4\u5c0f\u3002<\/p>\n<\/section>\n<\/li>\n
                        • \n
                          \u5bf9\u4e8e\u91c7\u7528\u5747\u65b9\u8bef\u5dee\u635f\u5931LMS\u7684\u7ebf\u6027\u6a21\u578b\uff0c\u635f\u5931\u51fd\u6570\u6070\u4e3a\u4e8c\u9636\uff0c\u5982\u4e0b\u56fe\u6240\u793a<\/p>\n<\/section>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n

                          \"\u56fe\u7247\"<\/p>\n

                          \u4e0d\u540c\u65b9\u5411\u4e0a\u7684\u4e0b\u964d\u901f\u5ea6\u53d8\u5316\u4e0d\u540c\uff08\u4e8c\u9636\u5bfc\u4e0d\u540c\uff0c\u66f2\u7387\u4e0d\u540c\uff09<\/strong>\uff0c\u6070\u7531\u8f93\u5165\u7684\u534f\u65b9\u5dee\u77e9\u9635\u51b3\u5b9a\uff0c\u901a\u8fc7scaling\u6539\u53d8\u4e86\u635f\u5931\u51fd\u6570\u7684\u5f62\u72b6\uff0c\u51cf\u5c0f\u4e0d\u540c\u65b9\u5411\u4e0a\u7684\u66f2\u7387\u5dee\u5f02<\/strong>\u3002\u5c06\u6bcf\u4e2a\u7ef4\u5ea6\u4e0a\u7684\u4e0b\u964d\u5206\u89e3\u6765\u770b\uff0c\u7ed9\u5b9a\u4e00\u4e2a\u4e0b\u964d\u6b65\u957f\uff0c\u5982\u679c\u4e0d\u591f\u5c0f\uff0c\u6709\u7684\u7ef4\u5ea6\u4e0b\u964d\u7684\u591a\uff0c\u6709\u7684\u4e0b\u964d\u7684\u5c11\uff0c\u6709\u7684\u8fd8\u53ef\u80fd\u5728\u4e0a\u5347\uff0c\u635f\u5931\u51fd\u6570\u7684\u6574\u4f53\u8868\u73b0\u53ef\u80fd\u662f\u4e0a\u5347\u4e5f\u53ef\u80fd\u662f\u4e0b\u964d\uff0c\u5c31\u4f1a\u4e0d\u7a33\u5b9a\u3002scaling\u540e\u4e0d\u540c\u65b9\u5411\u4e0a\u7684\u66f2\u7387\u76f8\u5bf9\u66f4\u63a5\u8fd1\uff0c\u66f4\u5bb9\u6613\u9009\u62e9\u5230\u5408\u9002\u7684\u5b66\u4e60\u7387\uff0c\u4f7f\u4e0b\u964d\u8fc7\u7a0b\u76f8\u5bf9\u66f4\u7a33\u5b9a\u3002<\/strong><\/p>\n

                            \n
                          • \n
                              \n
                            • \n
                              \u53e6\u6709\u4eceHessian\u77e9\u9635\u7279\u5f81\u503c\u4ee5\u53cacondition number\u89d2\u5ea6\u7684\u7406\u89e3\uff0c\u8be6\u89c1Lecun paper-Efficient BackProp\u4e2d\u7684Convergence of Gradient Descent\u4e00\u8282\uff0c\u6709\u6e05\u6670\u7684\u6570\u5b66\u63cf\u8ff0\uff0c\u540c\u65f6\u8fd8\u4ecb\u7ecd\u4e86\u767d\u5316\u7684\u4f5c\u7528\u2014\u2014\u89e3\u9664\u7279\u5f81\u95f4\u7684\u7ebf\u6027\u76f8\u5173\u6027\uff0c\u4f7f\u6bcf\u4e2a\u7ef4\u5ea6\u4e0a\u7684\u68af\u5ea6\u4e0b\u964d\u53ef\u72ec\u7acb\u770b\u5f85\u3002<\/p>\n<\/section>\n<\/li>\n
                            • \n
                              \u6587\u7ae0\u5f00\u7bc7\u7684\u692d\u5706\u5f62\u548c\u5706\u5f62\u7b49\u9ad8\u7ebf\u56fe\uff0c\u4ec5\u5728\u91c7\u7528\u5747\u65b9\u8bef\u5dee\u7684\u7ebf\u6027\u6a21\u578b\u4e0a\u9002\u7528\uff0c\u5176\u4ed6\u635f\u5931\u51fd\u6570\u6216\u66f4\u590d\u6742\u7684\u6a21\u578b\uff0c\u5982\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\uff0c\u635f\u5931\u51fd\u6570\u7684error surface\u53ef\u80fd\u5f88\u590d\u6742\uff0c\u5e76\u4e0d\u80fd\u7b80\u5355\u5730\u7528\u692d\u5706\u548c\u5706\u6765\u523b\u753b<\/strong>\uff0c\u6240\u4ee5\u7528\u5b83\u6765\u89e3\u91cafeature scaling\u5bf9\u6240\u6709\u635f\u5931\u51fd\u6570\u7684\u68af\u5ea6\u4e0b\u964d\u7684\u4f5c\u7528\uff0c\u4f3c\u4e4e\u8fc7\u4e8e\u7b80\u5316\uff0c\u89c1Hinton vedio-3.2 The error surface for a linear neuron\u3002<\/p>\n<\/section>\n<\/li>\n
                            • \n
                              \u5bf9\u4e8e\u635f\u5931\u51fd\u6570\u4e0d\u662f\u5747\u65b9\u8bef\u5dee\u7684\u60c5\u51b5\uff0c\u53ea\u8981\u6743\u91cdw\u4e0e\u8f93\u5165\u7279\u5f81x\u95f4\u662f\u76f8\u4e58\u5173\u7cfb\uff0c\u635f\u5931\u51fd\u6570\u5bf9w\u7684\u504f\u5bfc\u5fc5\u7136\u542b\u6709\u56e0\u5b50x\uff0cw\u7684\u68af\u5ea6\u4e0b\u964d\u901f\u5ea6\u5c31\u4f1a\u53d7\u5230\u7279\u5f81x\u5c3a\u5ea6\u7684\u5f71\u54cd\u3002\u7406\u8bba\u4e0a\u4e3a\u6bcf\u4e2a\u53c2\u6570\u90fd\u8bbe\u7f6e\u4e0a\u81ea\u9002\u5e94\u7684\u5b66\u4e60\u7387\uff0c\u53ef\u4ee5\u5438\u6536\u6389x\u5c3a\u5ea6\u7684\u5f71\u54cd\uff0c\u4f46\u5728\u5b9e\u8df5\u4e2d\u51fa\u4e8e\u8ba1\u7b97\u91cf\u7684\u8003\u8651\uff0c\u5f80\u5f80\u8fd8\u662f\u6240\u6709\u53c2\u6570\u5171\u7528\u4e00\u4e2a\u5b66\u4e60\u7387\uff0c\u6b64\u65f6x\u5c3a\u5ea6\u4e0d\u540c\u53ef\u80fd\u4f1a\u5bfc\u81f4\u4e0d\u540c\u65b9\u5411\u4e0a\u7684\u4e0b\u964d\u901f\u5ea6\u60ac\u6b8a\u8f83\u5927\uff0c\u5b66\u4e60\u7387\u4e0d\u5bb9\u6613\u9009\u62e9\uff0c\u4e0b\u964d\u8fc7\u7a0b\u4e5f\u53ef\u80fd\u4e0d\u7a33\u5b9a\uff0c\u901a\u8fc7scaling\u53ef\u5bf9\u4e0d\u540c\u65b9\u5411\u4e0a\u7684\u4e0b\u964d\u901f\u5ea6\u6709\u6240\u63a7\u5236\uff0c\u4f7f\u4e0b\u964d\u8fc7\u7a0b\u76f8\u5bf9\u66f4\u7a33\u5b9a\u3002<\/p>\n<\/section>\n<\/li>\n<\/ul>\n<\/li>\n
                            • \n
                              \u5bf9\u4e8e\u4f20\u7edf\u7684\u795e\u7ecf\u7f51\u7edc\uff0c\u5bf9\u8f93\u5165\u505afeature scaling\u4e5f\u5f88\u91cd\u8981\uff0c\u56e0\u4e3a\u91c7\u7528sigmoid\u7b49\u6709\u9971\u548c\u533a\u7684\u6fc0\u6d3b\u51fd\u6570\uff0c\u5982\u679c\u8f93\u5165\u5206\u5e03\u8303\u56f4\u5f88\u5e7f\uff0c\u53c2\u6570\u521d\u59cb\u5316\u65f6\u6ca1\u6709\u9002\u914d\u597d\uff0c\u5f88\u5bb9\u6613\u76f4\u63a5\u9677\u5165\u9971\u548c\u533a\uff0c\u5bfc\u81f4\u68af\u5ea6\u6d88\u5931<\/strong>\uff0c\u6240\u4ee5\uff0c\u9700\u8981\u5bf9\u8f93\u5165\u505aStandardization\u6216\u6620\u5c04\u5230[0,1]\u3001[\u22121,1]\uff0c\u914d\u5408\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u53c2\u6570\u521d\u59cb\u5316\u65b9\u6cd5\uff0c\u5bf9\u503c\u57df\u8fdb\u884c\u63a7\u5236\u3002\u4f46\u81ea\u4ece\u6709\u4e86Batch Normalization\uff0c\u6bcf\u6b21\u7ebf\u6027\u53d8\u6362\u6539\u53d8\u7279\u5f81\u5206\u5e03\u540e\uff0c\u90fd\u4f1a\u91cd\u65b0\u8fdb\u884cNormalization\uff0c\u4f3c\u4e4e\u53ef\u4ee5\u4e0d\u592a\u9700\u8981\u5bf9\u7f51\u7edc\u7684\u8f93\u5165\u8fdb\u884cfeature scaling\u4e86\uff1f\u4f46\u4e60\u60ef\u4e0a\u8fd8\u662f\u4f1a\u505afeature scaling\u3002<\/p>\n<\/section>\n<\/li>\n<\/ul>\n

                              \u4ec0\u4e48\u65f6\u5019\u4e0d\u9700\u8981Feature Scaling\uff1f<\/h2>\n
                              \u4e0e\u8ddd\u79bb\u8ba1\u7b97\u65e0\u5173\u7684\u6982\u7387\u6a21\u578b\uff0c\u4e0d\u9700\u8981feature scaling\uff0c\u6bd4\u5982Naive Bayes\uff1b<\/section>\n
                              \u4e0e\u8ddd\u79bb\u8ba1\u7b97\u65e0\u5173\u7684\u57fa\u4e8e\u6811\u7684\u6a21\u578b\uff0c\u4e0d\u9700\u8981feature scaling\uff0c\u6bd4\u5982\u51b3\u7b56\u6811\u3001\u968f\u673a\u68ee\u6797\u7b49\uff0c\u6811\u4e2d\u8282\u70b9\u7684\u9009\u62e9\u53ea\u5173\u6ce8\u5f53\u524d\u7279\u5f81\u5728\u54ea\u91cc\u5207\u5206\u5bf9\u5206\u7c7b\u66f4\u597d\uff0c\u5373\u53ea\u5728\u610f\u7279\u5f81\u5185\u90e8\u7684\u76f8\u5bf9\u5927\u5c0f\uff0c\u800c\u4e0e\u7279\u5f81\u95f4\u7684\u76f8\u5bf9\u5927\u5c0f\u65e0\u5173\u3002<\/section>\n

                              \u5c0f\u7ed3<\/h1>\n
                              \u8fd9\u7bc7\u6587\u7ae0\u5199\u5f97\u5341\u5206\u8270\u96be\uff0c\u4e00\u5f00\u59cb\u4ee5\u4e3a\u86ee\u7b80\u5355\u76f4\u63a5\uff0c\u4f46\u968f\u7740\u63a2\u7d22\u7684\u6df1\u5165\uff0c\u5192\u51fa\u7684\u95ee\u53f7\u8d8a\u6765\u8d8a\u591a\uff0c\u6253\u7834\u4e86\u5f88\u591a\u539f\u6765\u7684\u201c\u7406\u6240\u5f53\u7136\u201d\uff0c\u6240\u4ee5\uff0c\u5728\u5199\u7684\u8fc7\u7a0b\u4e2d\u4e0d\u505c\u5730\u505a\u52a0\u6cd5\uff0c\u5f88\u591a\u5730\u65b9\u60f3\u89e3\u91ca\u5f97\u5c3d\u91cf\u76f4\u89c2\uff0c\u53c8\u4e0d\u60f3\u7167\u642c\u592a\u591a\u516c\u5f0f\uff0c\u4f46\u81ea\u5df1\u7684\u7406\u89e3\u53c8\u4e0d\u591f\u6df1\u523b\uff0c\u5bfc\u81f4\u73b0\u5728\u53d9\u8ff0\u8fd9\u4e48\u5197\u957f\uff0c\u5e0c\u671b\u4ee5\u540e\u5728\u5199\u6587\u65f6\u80fd\u66f4\u4e13\u6ce8\u66f4\u7cbe\u70bc\u3002<\/section>\n
                              <\/div>\n

                              <\/g><\/svg><\/i> \"Loading\"<\/p>\n

                              <\/div>\n","protected":false},"excerpt":{"rendered":"

                              \u4f5c\u8005\u4e28shine-lee \u6765\u6e90\u4e28https:\/\/blog.csdn.net\/blogshinelee\/arti […]<\/p>\n

                              <\/div>\n

                              <\/g><\/svg><\/i> \"Loading\"<\/p>\n

                              <\/div>\n","protected":false},"author":1,"featured_media":804,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-803","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/803","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=803"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/803\/revisions"}],"predecessor-version":[{"id":818,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/803\/revisions\/818"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/804"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=803"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=803"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=803"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}