{"id":3381,"date":"2020-03-27T15:03:57","date_gmt":"2020-03-27T07:03:57","guid":{"rendered":"http:\/\/www.sniper97.cn\/?p=3381"},"modified":"2020-03-27T15:03:57","modified_gmt":"2020-03-27T07:03:57","slug":"%e3%80%90%e5%90%b4%e6%81%a9%e8%be%be%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e3%80%91%e5%ba%8f%e5%88%97%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"http:\/\/www.sniper97.cn\/index.php\/note\/deep-learning\/3381\/","title":{"rendered":"\u3010\u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u3011\u5e8f\u5217\u6a21\u578b"},"content":{"rendered":"\n

\u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u7b2c\u4e94\u8bfe\u7b2c\u4e00\u5468\uff1a\u5e8f\u5217\u6a21\u578b<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

1.\u4e3a\u4ec0\u4e48\u9009\u62e9\u5e8f\u5217\u6a21\u578b<\/h2>\n\n\n

\u5f88\u591a\u4efb\u52a1\u4e2d\u6570\u636e\u5bf9\u5148\u540e\u654f\u611f\uff0c\u4e3e\u4e2a\u7b80\u5355\u7684\u4f8b\u5b50\uff0cI love u \u548cu love i\uff0c\u610f\u601d\u660e\u663e\u4e0d\u540c\uff0c\u56e0\u6b64\u9700\u8981\u5bf9\u987a\u5e8f\u654f\u611f\u7684\u6a21\u578b\u6765\u5904\u7406\u7c7b\u4f3c\u6570\u636e\u3002<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

2.\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08RNN\uff09<\/h2>\n\n\n

\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\u5c31\u662f\u5c06\u4e0a\u4e00\u4e2a\u8f93\u5165\u4f5c\u4e3a\u4e0b\u4e00\u4e2a\u7684\u8f93\u5165<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

\u7f51\u7edc\u524d\u5411\u4f20\u64ad\u793a\u610f\u56fe\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

3.\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\u7684\u53cd\u5411\u4f20\u64ad<\/h2>\n\n\n
\"\"<\/figure><\/div>\n\n\n

4.\u4e0d\u540c\u7c7b\u578b\u7684\u5faa\u73af\u795e\u7ecf\u7f51\u7edc<\/h2>\n\n\n

\u6839\u636e\u8f93\u5165\u8f93\u51fa\u4e2a\u6570\u7684\u4e0d\u540c\uff0c\u5927\u6982\u5206\u4e3a\u51e0\u7c7b\uff1a<\/p>\n\n\n

many-to-many\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n
\"\"<\/figure><\/div>\n\n\n

many-to-one\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

one-to-one\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

one-to-many\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

5.\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\u7684\u68af\u5ea6\u6d88\u5931<\/h2>\n\n\n

\u7531\u4e8e\u6211\u4eec\u7684\u65f6\u95f4\u5e8f\u5217\u5e38\u5e38\u5bfc\u81f4\u795e\u7ecf\u7f51\u7edc\u5f88\u6df1\uff0c\u56e0\u6b64\u4e5f\u5c31\u4f1a\u4ea7\u751f\u68af\u5ea6\u6d88\u5931\u548c\u68af\u5ea6\u7206\u70b8\u7684\u95ee\u9898\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u622a\u53d6\u53c2\u6570\u6765\u907f\u514d\u68af\u5ea6\u7206\u70b8\uff0c\u90a3\u4e48\u68af\u5ea6\u6d88\u5931\u600e\u4e48\u5904\u7406\u5462\u3002<\/p>\n\n\n

\u7279\u522b\u662f\u5bf9\u4e8e <\/p>\n\n\n

The cat, which already ate \u2026\u2026, was full.<\/strong> <\/p>\n\n\n

The cats, which ate \u2026\u2026, were full.<\/strong><\/p>\n\n\n

\u6211\u4eec\u5c31\u9700\u8981\u8bb0\u5f55\u4e00\u4e2a\u503cc\uff0c\u6765\u544a\u8bc9\u540e\u9762\uff0c\u524d\u9762\u7684\u540d\u8bcd\u662f\u5355\u6570\u3002<\/p>\n\n\n

\u8fd9\u79cd\u7c7b\u4f3c\u7684\u95ee\u9898\uff0c\u8fc7\u957f\u7684\u8ddd\u79bb\u53ef\u80fd\u5bfc\u81f4\u5dee\u5f02\u6839\u672c\u5f71\u54cd\u4e0d\u5230\uff0c\u56e0\u6b64\u5f15\u5165\u95e8\u7684\u6982\u5ff5\u3002<\/p>\n\n\n

6.GRU\u5355\u5143<\/h2>\n\n\n

GRU\u5355\u5143\u5c31\u662f\u901a\u8fc7\u95e8\u63a7\u5faa\u73af\u5355\u5143\u6539\u53d8RNN\u9690\u85cf\u5c42\u7ed3\u6784\uff0c\u4f7f\u5979\u53ef\u4ee5\u66f4\u597d\u5730\u6355\u6349\u6df1\u5c42\u8fde\u63a5\uff0c\u6539\u5584\u4e86\u68af\u5ea6\u6d88\u5931\u3002<\/p>\n\n\n

GRU\u5355\u5143\u5b9e\u9645\u4e0a\u5c31\u662f\u901a\u8fc7\u4e00\u4e2a\u503c\uff0c\u6765\u51b3\u5b9a\u5f53\u524d\u5f53\u524d\u7ed3\u70b9\u8ba1\u7b97\u7684\u503c\u548c\u524d\u4e00\u4e2a\u9636\u6bb5\u7684\u7ed3\u679c\u7684\u5360\u6bd4\u3002<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

\u5b9e\u9645\u4e0a\u5b8c\u6574\u7684GRU\u5355\u5143\u8fd8\u6709\u4e00\u4e2a\u95e8\u63a7\uff0c\u5341\u5206\u7c7b\u4f3c\u4e8e\u4e0a\u9762\u8bf4\u7684\u95e8\uff0c\u4e0a\u9762\u7684\u95e8\u8bb0\u5f55\u4e86\u5f53\u524d\u7ed3\u70b9\u548c\u4e0a\u4e00\u4e2a\u7ed3\u70b9\u7684\u5173\u7cfb\uff0c\u56e0\u6b64\u8fd8\u6709\u4e00\u4e2a\u8bb0\u5f55\u4e0b\u4e00\u4e2a\u7ed3\u70b9\u548c\u5f53\u524d\u7ed3\u70b9\u7684\u5173\u7cfb\u3002\u4e5f\u5c31\u662f\uff1a<\/p>\n\n\n

\"\"<\/figure>\n\n\n

\u56e0\u6b64\u6709\u5b8c\u6574\u516c\u5f0f\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

7.\u957f\u77ed\u65f6\u8bb0\u5fc6\u7f51\u7edc\uff08LSTM\uff09<\/h2>\n\n\n

LSTM\u662f\u4e00\u4e2a\u6bd4GRU\u66f4\u5f3a\u5927\u548c\u901a\u7528\u7684\u7248\u672c<\/p>\n\n\n

\u76f8\u6bd4GRU\uff0cLSTM\u6709\u66f4\u65b0\u95e8update\u3001\u9057\u5fd8\u95e8forget\u548c\u8f93\u51fa\u95e8output\u3002<\/p>\n\n\n

\u800c\u5bf9\u4e8eupdate\u548cforget\u4ece\u516c\u5f0f\u4e2d\u6211\u4eec\u53ef\u4ee5\u770b\u5230\uff0c\u5b9e\u9645\u4e0a\u5c31\u662fGRU\u7684\uff081- \u0393 \uff09\u53d8\u6210\u4e86\u9057\u5fd8\u95e8\u3002<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

\u6211\u4eec\u5f88\u5bb9\u6613\u7684\u53d1\u73b0\uff0cLSTM\u5bf9\u4e8e\u7279\u5f81c\uff08\u5355\u590d\u6570\uff09\u6709\u4e00\u4e2a\u6a2a\u5411\u7684\u4f20\u64ad<\/p>\n\n\n

8.\u53cc\u5411\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08BRNN\uff09<\/h2>\n\n\n

\u6211\u4eec\u73b0\u5728\u5df2\u7ecf\u53ef\u4ee5\u4f7f\u7528RNN\u83b7\u5f97\u4e4b\u524d\u5e8f\u5217\u7684\u72b6\u6001\uff0c\u90a3\u4e48\u4e4b\u540e\u5462\uff1f\u6211\u4eec\u4f7f\u7528\u4e24\u4e2aRNN\u7ec4\u5408\u800c\u6210\u7684\u53cc\u5411\u5faa\u73af\u795e\u7ecf\u7f51\u7edcBRNN\u6765\u89e3\u51b3\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u3002<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

\u53ef\u4ee5\u770b\u5230\uff0c\u8fd9\u4e2a\u7f51\u7edc\u4e2d\u6709\u4e24\u7ec4RNN\uff0c\u4e00\u4e2a\u5411\u524d\uff0c\u4e00\u4e2a\u5411\u540e\uff0c\u56e0\u6b64\u8fd9\u4e2a\u7f51\u7edc\u5c31\u53ef\u4ee5\u5b8c\u6210\u5bf9\u4e8e\u4e0a\u4e0b\u6587\u7684\u5173\u7cfb\u83b7\u53d6\u3002<\/p>\n\n\n

9.\u6df1\u5c42\u5faa\u73af\u795e\u7ecf\u7f51\u7edc<\/h2>\n\n\n

\u6211\u4eec\u5728\u4e4b\u524d\u4ecb\u7ecd\u4e86\u5f88\u6df1\u7684CNN\u6a21\u578b\u6216\u8005DNN\u6a21\u578b\uff0c\u5bf9\u4e8eRNN\uff0c\u7531\u4e8e\u65f6\u95f4\u5e8f\u5217\u8f83\u957f\uff0c\u6211\u4eec\u4e00\u822c\u4e0d\u505a\u5f88\u6df1\u7684\u7f51\u7edc\uff0c\u4e09\u5c42\u57fa\u672c\u5c31\u6700\u591a\u4e86\uff1a<\/p>\n\n\n

\"\"<\/figure><\/div>\n\n\n

\u4f46\u662f\u5bf9\u4e8e\u8f93\u51fa\uff0c\u6211\u4eec\u53ef\u4ee5\u7ee7\u7eed\u9009\u62e9\u5d4c\u5957DNN\u3002<\/p>\n\n\n

\u6d4b\u9a8c<\/h2>\n\n\n

1. \u5047\u8bbe\u4f60\u7684\u8bad\u7ec3\u6837\u672c\u662f\u53e5\u5b50(\u5355\u8bcd\u5e8f\u5217)\uff0c\u4e0b\u9762\u54ea\u4e2a\u9009\u9879\u6307\u7684\u662f\u7b2c<\/strong>i<\/strong><\/em>\u4e2a\u8bad\u7ec3\u6837\u672c\u4e2d\u7684\u7b2c<\/strong>j<\/strong><\/em>\u4e2a\u8bcd? <\/strong><\/p>\n\n\n

x<\/em>^{(i<\/em>)<j<\/em>>} <\/p>\n\n\n

2. \u770b\u4e00\u4e0b\u4e0b\u9762\u7684\u8fd9\u4e2a\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff1a <\/strong> <\/p>\n\n\n

\"\"<\/figure>\n\n\n

\u5728\u4e0b\u9762\u7684\u6761\u4ef6\u4e2d\uff0c\u6ee1\u8db3\u4e0a\u56fe\u4e2d\u7684\u7f51\u7edc\u7ed3\u6784\u7684\u53c2\u6570\u662f\uff1a <\/strong><\/p>\n\n\n

  1. Tx<\/em>=Ty<\/em><\/li>
  2. Tx<\/em><Ty<\/em><\/li>
  3. Tx<\/em>>Ty<\/em><\/li>
  4. Tx<\/em> = 1<\/li><\/ol>\n\n\n

    1\u3002\uff08\u6bcf\u4e00\u4e2a\u8f93\u5165\u90fd\u6709\u4e00\u4e2a\u8f93\u51fa\uff09<\/p>\n\n\n

    3. \u8fd9\u4e9b\u4efb\u52a1\u4e2d\u7684\u54ea\u4e00\u4e2a\u4f1a\u4f7f\u7528\u591a\u5bf9\u4e00\u7684RNN\u4f53\u7cfb\u7ed3\u6784\uff1f <\/strong><\/p>\n\n\n

    1. \u8bed\u97f3\u8bc6\u522b\uff08\u8f93\u5165\u8bed\u97f3\uff0c\u8f93\u51fa\u6587\u672c\uff09 <\/li>
    2. \u60c5\u611f\u5206\u7c7b\uff08\u8f93\u5165\u4e00\u6bb5\u6587\u5b57\uff0c\u8f93\u51fa0\u62161\u8868\u793a\u6b63\u9762\u6216\u8005\u8d1f\u9762\u7684\u60c5\u7eea<\/li>
    3. \u56fe\u50cf\u5206\u7c7b\uff08\u8f93\u5165\u4e00\u5f20\u56fe\u7247\uff0c\u8f93\u51fa\u5bf9\u5e94\u7684\u6807\u7b7e\uff09\u3002 <\/li>
    4. \u4eba\u58f0\u6027\u522b\u8bc6\u522b\uff08\u8f93\u5165\u8bed\u97f3\uff0c\u8f93\u51fa\u8bf4\u8bdd\u4eba\u7684\u6027\u522b\uff09\u3002 <\/li><\/ol>\n\n\n

      2\uff0c4\u3002<\/p>\n\n\n

      4. \u5047\u8bbe\u4f60\u73b0\u5728\u6b63\u5728\u8bad\u7ec3\u4e0b\u9762\u8fd9\u4e2aRNN\u7684\u8bed\u8a00\u6a21\u578b\uff1a \u5728<\/strong>t<\/strong><\/em>\u65f6\uff0c\u8fd9\u4e2aRNN\u5728\u505a\u4ec0\u4e48\uff1f <\/strong><\/p>\n\n\n

      \"\"<\/figure><\/div>\n\n\n
      \"\"<\/figure>\n\n\n

      5. \u4f60\u5df2\u7ecf\u5b8c\u6210\u4e86\u4e00\u4e2a\u8bed\u8a00\u6a21\u578bRNN\u7684\u8bad\u7ec3\uff0c\u5e76\u7528\u5b83\u6765\u5bf9\u53e5\u5b50\u8fdb\u884c\u968f\u673a\u53d6\u6837\uff0c\u5982\u4e0b\u56fe\uff1a \u5728\u6bcf\u4e2a\u65f6\u95f4\u6b65<\/strong>t<\/strong><\/em>\u90fd\u5728\u505a\u4ec0\u4e48\uff1f <\/strong><\/p>\n\n\n

      \"\"<\/figure><\/div>\n\n\n
      1. (1)\u4f7f\u7528RNN\u8f93\u51fa\u7684\u6982\u7387\uff0c\u9009\u62e9\u8be5\u65f6\u95f4\u6b65\u7684\u6700\u9ad8\u6982\u7387\u5355\u8bcd\u4f5c\u4e3ay<\/em>^<t<\/em>>\uff0c(2)\u7136\u540e\u5c06\u8bad\u7ec3\u96c6\u4e2d\u7684\u6b63\u786e\u7684\u5355\u8bcd\u4f20\u9012\u5230\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u3002 <\/li>
      2. (i)\u4f7f\u7528\u7531RNN\u8f93\u51fa\u7684\u6982\u7387\u5c06\u8be5\u65f6\u95f4\u6b65\u7684\u6240\u9009\u5355\u8bcd\u8fdb\u884c\u968f\u673a\u91c7\u6837\u4f5c\u4e3a y<\/em>^<t<\/em>>\uff0c(2)\u7136\u540e\u5c06\u8bad\u7ec3\u96c6\u4e2d\u7684\u5b9e\u9645\u5355\u8bcd\u4f20\u9012\u5230\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u3002 <\/li>
      3. (1)\u4f7f\u7528\u7531RNN\u8f93\u51fa\u7684\u6982\u7387\u6765\u9009\u62e9\u8be5\u65f6\u95f4\u6b65\u7684\u6700\u9ad8\u6982\u7387\u8bcd\u4f5c\u4e3a y<\/em>^<t<\/em>>\uff0c(2)\u7136\u540e\u5c06\u8be5\u9009\u62e9\u7684\u8bcd\u4f20\u9012\u7ed9\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u3002 <\/li>
      4. (1)\u4f7f\u7528RNN\u8be5\u65f6\u95f4\u6b65\u8f93\u51fa\u7684\u6982\u7387\u5bf9\u5355\u8bcd\u968f\u673a\u62bd\u6837\u7684\u7ed3\u679c\u4f5c\u4e3ay<\/em>^<t<\/em>>\uff0c(2)\u7136\u540e\u5c06\u6b64\u9009\u5b9a\u5355\u8bcd\u4f20\u9012\u7ed9\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u3002 <\/li><\/ol>\n\n\n

        4\u3002\uff08\u56e0\u4e3a\u9898\u76ee\u4e2d\u8bf4\u4e86\u5bf9\u53e5\u5b50\u8fdb\u884c\u968f\u673a\u53d6\u6837\uff09<\/p>\n\n\n

        6. \u4f60\u6b63\u5728\u8bad\u7ec3\u4e00\u4e2aRNN\u7f51\u7edc\uff0c\u4f60\u53d1\u73b0\u4f60\u7684\u6743\u91cd\u4e0e\u6fc0\u6d3b\u503c\u90fd\u662f\u201cNaN\u201d\uff0c\u4e0b\u5217\u9009\u9879\u4e2d\uff0c\u54ea\u4e00\u4e2a\u662f\u5bfc\u81f4\u8fd9\u4e2a\u95ee\u9898\u7684\u6700\u6709\u53ef\u80fd\u7684\u539f\u56e0\uff1f <\/strong><\/p>\n\n\n

        \u68af\u5ea6\u7206\u70b8\u3002<\/p>\n\n\n

        7. \u5047\u8bbe\u4f60\u6b63\u5728\u8bad\u7ec3\u4e00\u4e2aLSTM\u7f51\u7edc\uff0c\u4f60\u6709\u4e00\u4e2a10,000\u8bcd\u7684\u8bcd\u6c47\u8868\uff0c\u5e76\u4e14\u4f7f\u7528\u4e00\u4e2a\u6fc0\u6d3b\u503c\u7ef4\u5ea6\u4e3a100\u7684LSTM\u5757\uff0c\u5728\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65\u4e2d\uff0c\u0393<\/strong>u<\/strong><\/em>\u7684\u7ef4\u5ea6\u662f\u591a\u5c11\uff1f<\/strong> <\/p>\n\n\n

        100\u3002\uff08\u6fc0\u6d3b\u503c\u7ef4\u5ea6=LSTM\u957f\u5ea6=\u95e8\u63a7\u4e2a\u6570\uff09<\/p>\n\n\n

        8. \u8fd9\u91cc\u6709\u4e00\u4e9bGRU\u7684\u66f4\u65b0\u65b9\u7a0b\uff1a \u7231\u4e3d\u4e1d\u5efa\u8bae\u901a\u8fc7\u79fb\u9664 \u0393<\/strong>u<\/strong><\/em>\u6765\u7b80\u5316GRU\uff0c\u5373\u8bbe\u7f6e\u0393<\/strong>u<\/strong><\/em>\uff1d1\u3002\u8d1d\u8482\u63d0\u51fa\u901a\u8fc7\u79fb\u9664\u0393<\/strong>r<\/strong><\/em>\u6765\u7b80\u5316GRU\uff0c\u5373\u8bbe\u7f6e\u0393<\/strong>r<\/strong><\/em>\uff1d1\u3002\u54ea\u79cd\u6a21\u578b\u66f4\u5bb9\u6613\u5728\u68af\u5ea6\u4e0d\u6d88\u5931\u95ee\u9898\u7684\u60c5\u51b5\u4e0b\u8bad\u7ec3\uff0c\u5373\u4f7f\u5728\u5f88\u957f\u7684\u8f93\u5165\u5e8f\u5217\u4e0a\u4e5f\u53ef\u4ee5\u8fdb\u884c\u8bad\u7ec3\uff1f <\/strong><\/p>\n\n\n

        \"\"<\/figure><\/div>\n\n\n
        1. \u7231\u4e3d\u4e1d\u7684\u6a21\u578b\uff08\u5373\u79fb\u9664\u0393u<\/em>\uff09\uff0c\u56e0\u4e3a\u5bf9\u4e8e\u4e00\u4e2a\u65f6\u95f4\u6b65\u800c\u8a00\uff0c\u5982\u679c\u0393r<\/em>\u22480\uff0c\u68af\u5ea6\u53ef\u4ee5\u901a\u8fc7\u65f6\u95f4\u6b65\u53cd\u5411\u4f20\u64ad\u800c\u4e0d\u4f1a\u8870\u51cf\u3002<\/li>
        2. \u7231\u4e3d\u4e1d\u7684\u6a21\u578b\uff08\u5373\u79fb\u9664\u0393u<\/em>\uff09\uff0c\u56e0\u4e3a\u5bf9\u4e8e\u4e00\u4e2a\u65f6\u95f4\u6b65\u800c\u8a00\uff0c\u5982\u679c\u0393r<\/em>\u22481\uff0c\u68af\u5ea6\u53ef\u4ee5\u901a\u8fc7\u65f6\u95f4\u6b65\u53cd\u5411\u4f20\u64ad\u800c\u4e0d\u4f1a\u8870\u51cf\u3002 <\/li>
        3. \u8d1d\u8482\u7684\u6a21\u578b\uff08\u5373\u79fb\u9664\u0393r<\/em>\uff09\uff0c\u56e0\u4e3a\u5bf9\u4e8e\u4e00\u4e2a\u65f6\u95f4\u6b65\u800c\u8a00\uff0c\u5982\u679c\u0393u<\/em>\u22480\uff0c\u68af\u5ea6\u53ef\u4ee5\u901a\u8fc7\u65f6\u95f4\u6b65\u53cd\u5411\u4f20\u64ad\u800c\u4e0d\u4f1a\u8870\u51cf\u3002 <\/li>
        4. \u8d1d\u8482\u7684\u6a21\u578b\uff08\u5373\u79fb\u9664\u0393r<\/em>\uff09\uff0c\u56e0\u4e3a\u5bf9\u4e8e\u4e00\u4e2a\u65f6\u95f4\u6b65\u800c\u8a00\uff0c\u5982\u679c\u0393u<\/em>\u22481\uff0c\u68af\u5ea6\u53ef\u4ee5\u901a\u8fc7\u65f6\u95f4\u6b65\u53cd\u5411\u4f20\u64ad\u800c\u4e0d\u4f1a\u8870\u51cf\u3002 <\/li><\/ol>\n\n\n

          3\u3002\uff08 \u8981\u4f7f\u4fe1\u53f7\u53cd\u5411\u4f20\u64ad\u800c\u4e0d\u6d88\u5931\uff0c\u6211\u4eec\u9700\u8981 c<\/em><t<\/em>>\u9ad8\u5ea6\u4f9d\u8d56\u4e8ec<\/em><t<\/em>\u22121> \uff09<\/p>\n\n\n

          9. \u8fd9\u91cc\u6709\u4e00\u4e9bGRU\u548cLSTM\u7684\u65b9\u7a0b: \u4ece\u8fd9\u4e9b\u6211\u4eec\u53ef\u4ee5\u770b\u5230\uff0c\u5728LSTM\u4e2d\u7684\u66f4\u65b0\u95e8\u548c\u9057\u5fd8\u95e8\u5728GRU\u4e2d\u626e\u6f14\u7c7b\u4f3c \u2013\u2013\u2013\u2013\u2013\u4e0e\u2013\u2013\u2013\u2013\u2013\u7684\u89d2\u8272\uff0c\u7a7a\u767d\u5904\u5e94\u8be5\u586b\u4ec0\u4e48\uff1f <\/strong><\/p>\n\n\n

          \"\"<\/figure><\/div>\n\n\n

          \u0393u<\/em> \u4e0e 1\u2212\u0393u<\/em><\/p>\n\n\n

          10. \u4f60\u6709\u4e00\u53ea\u5ba0\u7269\u72d7\uff0c\u5b83\u7684\u5fc3\u60c5\u5f88\u5927\u7a0b\u5ea6\u4e0a\u53d6\u51b3\u4e8e\u5f53\u524d\u548c\u8fc7\u53bb\u51e0\u5929\u7684\u5929\u6c14\u3002\u4f60\u5df2\u7ecf\u6536\u96c6\u4e86\u8fc7\u53bb365\u5929\u7684\u5929\u6c14\u6570\u636e<\/strong>x<\/strong><\/em><1>,\u2026,<\/strong>x<\/strong><\/em><365>\uff0c\u8fd9\u4e9b\u6570\u636e\u662f\u4e00\u4e2a\u5e8f\u5217\uff0c\u4f60\u8fd8\u6536\u96c6\u4e86\u4f60\u7684\u72d7\u5fc3\u60c5\u7684\u6570\u636e<\/strong>y<\/strong><\/em><1>,\u2026,<\/strong>y<\/strong><\/em><365>\uff0c\u4f60\u60f3\u5efa\u7acb\u4e00\u4e2a\u6a21\u578b\u6765\u4ecex\u5230y\u8fdb\u884c\u6620\u5c04\uff0c\u4f60\u5e94\u8be5\u4f7f\u7528\u5355\u5411RNN\u8fd8\u662f\u53cc\u5411RNN\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff1f <\/strong><\/p>\n\n\n

          \u5355\u5411RNN\uff0c\u56e0\u4e3a\u72d7\u4e5f\u4e0d\u80fd\u9884\u77e5\u672a\u6765x\u3002<\/p>\n\n\n

          \u7f16\u7a0b\u4f5c\u4e1a<\/h2>\n\n\n

          \u9996\u5148\u7b80\u5355\u5b9e\u73b0\u4ee5\u4e0bRNN\u548cLSTM\u7136\u540e\u4f7f\u7528RNN\u9884\u6d4b\u6050\u9f99\u540d\u5b57\u3002<\/p>\n\n\n

          RNN\u6a21\u578b\u548c\u516c\u5f0f\u5982\u4e0b\uff1a <\/p>\n\n\n

          \"\"<\/figure><\/div>\n\n\n

          \u9996\u5148\u5bfc\u5305\uff1a<\/p>\n\n\n

          import numpy as np
          from course_5_week_1 import rnn_utils<\/pre>\n\n\n
          \u7136\u540e\u662f\u4e00\u4e2acell\u7684\u524d\u5411\u4f20\u64ad\uff0c\u5176\u4e2d\u6700\u4e3b\u8981\u7684\u5c31\u662f\u54ea\u4e24\u4e2a\u516c\u5f0f\uff1a<\/p>\n\n\n
          def rnn_cell_forward(xt, a_prev, parameters):
              \"\"\"
          <\/em>    rnn cell<\/em>\u524d\u5411\u4f20\u64ad
          <\/em>    <\/em>:param<\/em><\/strong> xt:
          <\/em>    <\/em>:param<\/em><\/strong> a_prev:
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em># \u4eceparameters\u201d\u83b7\u53d6\u53c2\u6570
              Wax = parameters[\"Wax\"]
              Waa = parameters[\"Waa\"]
              Wya = parameters[\"Wya\"]
              ba = parameters[\"ba\"]
              by = parameters[\"by\"]

              a_next = np.tanh(np.dot(Waa, a_prev) + np.dot(Wax, xt) + ba)
              # \u4f7f\u7528\u4e0a\u9762\u7684\u516c\u5f0f\u8ba1\u7b97\u5f53\u524d\u5355\u5143\u7684\u8f93\u51fa
              yt_pred = rnn_utils.softmax(np.dot(Wya, a_next) + by)

              # \u4fdd\u5b58\u53cd\u5411\u4f20\u64ad\u9700\u8981\u7684\u503c
              cache = (a_next, a_prev, xt, parameters)

              return a_next, yt_pred, cache<\/pre>\n\n\n
          \u7136\u540e\u6211\u4eec\u505a\u5168\u90e8\u7684\u524d\u5411\u4f20\u64ad\uff1a<\/p>\n\n\n
          def rnn_forward(X, a0, parameters):
              \"\"\"
          <\/em>    rnn <\/em>\u524d\u5411\u4f20\u64ad
          <\/em>    <\/em>:param<\/em><\/strong> X:
          <\/em>    <\/em>:param<\/em><\/strong> a0:
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em>caches = []

              # \u83b7\u53d6x\u548cWya\u7684\u7ef4\u5ea6\u4fe1\u606f
              n_x, m, T_x = X.shape
              n_y, n_a = parameters['Wya'].shape

              # \u4f7f\u75280\u6765\u521d\u59cb\u5316a\u548cy
              a = np.zeros([n_a, m, T_x])
              y_pred = np.zeros([n_y, m, T_x])

              # \u521d\u59cb\u5316next
              a_next = a0

              # \u904d\u5386\u6240\u6709\u7684\u5e8f\u5217
              for i in range(T_x):
                  # cell\u524d\u5411\u4f20\u64ad
                  a_next, yt_pred, cache = rnn_cell_forward(X[:, :, i], a_next, parameters)
                  # \u4f7f\u7528 a \u6765\u4fdd\u5b58\u201cnext\u201d\u9690\u85cf\u72b6\u6001\uff08\u7b2c i\uff09\u4e2a\u4f4d\u7f6e\u3002
                  a[:, :, i] = a_next
                  # \u4fdd\u5b58\u9884\u6d4b\u503c
                  y_pred[:, :, i] = yt_pred

                  caches.append(cache)
              caches = (caches, X)

              return a, y_pred, caches
          <\/pre>\n\n\n
          \u540c\u6837\uff0c\u5bf9\u4e8eLSTM\uff0c\u6a21\u578b\u548c\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n\n\n
          \"\"<\/figure><\/div>\n\n\n
          lstm\u7684cell\u5982\u4e0b\uff0c\u6838\u5fc3\u8fd8\u662f\u90a36\u4e2a\u516c\u5f0f\uff1a<\/p>\n\n\n
          def lstm_cell_forward(xt, a_prev, c_prev, parameters):
              \"\"\"
          <\/em>    lstm cell <\/em>\u524d\u5411\u4f20\u64ad
          <\/em>    <\/em>:param<\/em><\/strong> xt:
          <\/em>    <\/em>:param<\/em><\/strong> a_prev:
          <\/em>    <\/em>:param<\/em><\/strong> c_prev:
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em># \u4ece\u201cparameters\u201d\u4e2d\u83b7\u53d6\u76f8\u5173\u503c
              Wf = parameters[\"Wf\"]
              bf = parameters[\"bf\"]
              Wi = parameters[\"Wi\"]
              bi = parameters[\"bi\"]
              Wc = parameters[\"Wc\"]
              bc = parameters[\"bc\"]
              Wo = parameters[\"Wo\"]
              bo = parameters[\"bo\"]
              Wy = parameters[\"Wy\"]
              by = parameters[\"by\"]

              # \u83b7\u53d6x_t\u3001Wy\u7ef4\u5ea6\u4fe1\u606f
              n_x, m = xt.shape
              n_y, n_a = Wy.shape

              # \u94fe\u63a5a_prev\u4e0ext
              contact = np.zeros([n_a + n_x, m])
              contact[:n_a, :] = a_prev
              contact[n_a:, :] = xt

              # \u6839\u636e\u516c\u5f0f\u8ba1\u7b97
              # \u9057\u5fd8\u95e8\uff0c\u516c\u5f0f1
              ft = rnn_utils.sigmoid(np.dot(Wf, contact) + bf)

              # \u66f4\u65b0\u95e8\uff0c\u516c\u5f0f2
              ut = rnn_utils.sigmoid(np.dot(Wi, contact) + bi)

              # \u66f4\u65b0\u5355\u5143\uff0c\u516c\u5f0f3
              cct = np.tanh(np.dot(Wc, contact) + bc)

              # \u66f4\u65b0\u5355\u5143\uff0c\u516c\u5f0f4
              c_next = ft * c_prev + ut * cct
              # \u8f93\u51fa\u95e8\uff0c\u516c\u5f0f5
              ot = rnn_utils.sigmoid(np.dot(Wo, contact) + bo)

              # \u8f93\u51fa\u95e8\uff0c\u516c\u5f0f6
              # a_next = np.multiply(ot, np.tan(c_next))
              a_next = ot * np.tanh(c_next)
              # 3.\u8ba1\u7b97LSTM\u5355\u5143\u7684\u9884\u6d4b\u503c
              yt_pred = rnn_utils.softmax(np.dot(Wy, a_next) + by)

              # \u4fdd\u5b58\u5305\u542b\u4e86\u53cd\u5411\u4f20\u64ad\u6240\u9700\u8981\u7684\u53c2\u6570
              cache = (a_next, c_next, a_prev, c_prev, ft, ut, cct, ot, xt, parameters)

              return a_next, c_next, yt_pred, cache<\/pre>\n\n\n
          \u56e0\u6b64\u6574\u4e2alstm\u7684\u524d\u5411\u4f20\u64ad\u5982\u4e0b\uff1a<\/p>\n\n\n
          def lstm_forward(x, a0, parameters):
              \"\"\"
          <\/em>    lstm<\/em>\u524d\u5411\u4f20\u64ad
          <\/em>    <\/em>:param<\/em><\/strong> x:
          <\/em>    <\/em>:param<\/em><\/strong> a0:
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em>caches = []

              n_x, m, T_x = x.shape
              n_y, n_a = parameters['Wy'].shape

              # \u4f7f\u75280\u6765\u521d\u59cb\u5316a,c,y
              a = np.zeros([n_a, m, T_x])
              c = np.zeros([n_a, m, T_x])
              y = np.zeros([n_y, m, T_x])

              # \u521d\u59cb\u5316anext  cnext
              a_next = a0
              c_next = np.zeros([n_a, m])

              # \u904d\u5386\u6240\u6709\u7684\u65f6\u95f4\u6b65
              for t in range(T_x):
                  # \u66f4\u65b0\u4e0b\u4e00\u4e2a\u9690\u85cf\u72b6\u6001\uff0c\u4e0b\u4e00\u4e2a\u8bb0\u5fc6\u72b6\u6001\uff0c\u8ba1\u7b97\u9884\u6d4b\u503c\uff0c\u83b7\u53d6cache
                  a_next, c_next, yt_pred, cache = lstm_cell_forward(x[:, :, t], a_next, c_next, parameters)

                  # \u4fdd\u5b58\u65b0\u7684\u4e0b\u4e00\u4e2a\u9690\u85cf\u72b6\u6001\u5230\u53d8\u91cfa\u4e2d
                  a[:, :, t] = a_next

                  # \u4fdd\u5b58\u9884\u6d4b\u503c\u5230\u53d8\u91cfy\u4e2d
                  y[:, :, t] = yt_pred

                  # \u4fdd\u5b58\u4e0b\u4e00\u4e2a\u5355\u5143\u72b6\u6001\u5230\u53d8\u91cfc\u4e2d
                  c[:, :, t] = c_next

                  # \u628acache\u6dfb\u52a0\u5230caches\u4e2d
                  caches.append(cache)

              # \u4fdd\u5b58\u53cd\u5411\u4f20\u64ad\u9700\u8981\u7684\u53c2\u6570
              caches = (caches, x)

              return a, y, c, caches<\/pre>\n\n\n
          \u7136\u540e\u6211\u4eec\u505a\u4e2a\u7b80\u5355\u7684RNN\u7f51\u7edc\uff1a<\/p>\n\n\n
          \u9996\u5148\u5bfc\u5305\uff1a<\/p>\n\n\n
          import numpy as np
          import random
          import time

          from course_5_week_1 import cllm_utils as utils<\/pre>\n\n\n
          \u68af\u5ea6\u4fee\u526a\u7684\u65b9\u6cd5\uff1a<\/p>\n\n\n
          def clip(gradients, maxValue):
              \"\"\"
          <\/em>    <\/em>\u68af\u5ea6\u4fee\u526a
          <\/em>    <\/em>:param<\/em><\/strong> gradients:
          <\/em>    <\/em>:param<\/em><\/strong> maxValue:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em>dWaa, dWax, dWya, db, dby = gradients['dWaa'], gradients['dWax'], gradients['dWya'], gradients['db'], gradients[
                  'dby']
              # \u68af\u5ea6\u4fee\u526a
              for gradient in [dWaa, dWax, dWya, db, dby]:
                  np.clip(gradient, -maxValue, maxValue, out=gradient)

              gradients = {
                  'dWaa': dWaa,
                  'dWax': dWax,
                  'dWya': dWya,
                  'db': db,
                  'dby': dby
              }
              return gradients
          <\/pre>\n\n\n
          \u5728\u7f51\u7edc\u4e2d\u8fdb\u884c\u53d6\u6837\uff0c\u5c31\u662f\u5bf9\u6bcf\u4e00\u4e2aLSTM\u7684cell\u53d6\u51fa\u5355\u8bcd\uff0c\u6c47\u6210\u4e00\u4e2a\u53e5\u5b50\uff1a<\/p>\n\n\n
          def sample(parameters, char_to_ix, seed):
              \"\"\"
          <\/em>    <\/em>\u53d6\u6837
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:param<\/em><\/strong> char_to_is:
          <\/em>    <\/em>:param<\/em><\/strong> seed:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em>Waa, Wax, Wya, by, b = parameters['Waa'], parameters['Wax'], parameters['Wya'], parameters['by'], parameters['b']
              vocab_size = by.shape[0]
              n_a = Waa.shape[1]

              # \u521b\u5efaone hot \u5411\u91cf
              x = np.zeros((vocab_size, 1))
              # \u521d\u59cb\u5316a0
              a_prev = np.zeros((n_a, 1))
              # \u8f93\u51fa\u5217\u8868
              output = []
              # \u6362\u884c\u6807\u8bb0
              idx = -1
              counter = 0
              newline_character = char_to_ix['\\n']

              # \u5982\u679c\u4e0d\u662f\u6362\u884c\u7b26\u6216\u8005\u8f93\u51fa\u4e0d\u8d85\u8fc750\u4e2a
              while idx != newline_character and counter < 50:
                  # \u524d\u5411\u4f20\u64ad
                  a = np.tanh(np.dot(Wax, x) + np.dot(Waa, a_prev) + b)
                  z = np.dot(Wya, a) + by
                  y = utils.softmax(z)

                  # \u968f\u673a\u6570\u79cd\u5b50
                  np.random.seed(counter + seed)

                  # \u4ece\u968f\u673a\u6982\u7387\u5206\u5e03\u4e2d\u83b7\u53d6\u5b57\u7b26
                  idx = np.random.choice(list(range(vocab_size)), p=y.ravel())

                  # \u6dfb\u52a0\u7d22\u5f15
                  output.append(idx)

                  x = np.zeros((vocab_size, 1))
                  x[idx] = 1

                  # \u66f4\u65b0a_prev
                  a_prev = a

                  seed += 1
                  counter += 1

              if counter == 50:
                  output.append(char_to_ix['\\n'])

              return output<\/pre>\n\n\n
          \u7136\u540e\u662f\u4f18\u5316\u65b9\u6cd5\uff0c\u5305\u62ec\u524d\u5411\u3001\u53cd\u5411\u3001\u66f4\u65b0\u53c2\u6570\uff1a<\/p>\n\n\n
          def optimize(X, y, a_prev, parameters, learning_rate=0.01):
              \"\"\"
          <\/em>    rnn<\/em>\u7684\u5355\u6b65\u4f18\u5316
          <\/em>    <\/em>:param<\/em><\/strong> X:
          <\/em>    <\/em>:param<\/em><\/strong> y:
          <\/em>    <\/em>:param<\/em><\/strong> a_prev:
          <\/em>    <\/em>:param<\/em><\/strong> parameters:
          <\/em>    <\/em>:param<\/em><\/strong> learning_rate:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>    <\/em># \u524d\u5411\u4f20\u64ad
              loss, cache = utils.rnn_forward(X, y, a_prev, parameters)
              # \u53cd\u5411\u4f20\u64ad
              gradients, a = utils.rnn_backward(X, y, parameters, cache)
              # \u68af\u5ea6\u4fee\u526a
              gradients = clip(gradients, 5)
              # \u66f4\u65b0\u53c2\u6570
              parameters = utils.update_parameters(parameters, gradients, learning_rate)

              return loss, gradients, a[len(X) - 1]  # \u4ee3\u4ef7  \u53c2\u6570  \u9690\u85cf\u5c42<\/pre>\n\n\n
          \u6700\u540e\u5c06\u4ed6\u4eec\u5168\u90e8\u5408\u6210\u4e00\u4e2a\u6a21\u5757\u5e76\u8c03\u7528\uff1a<\/p>\n\n\n
          def model(data, ix_to_char, char_to_ix, num_iterations=3500,
                    n_a=50, dino_names=7, vocab_size=27):
              \"\"\"
          <\/em>    <\/em>\u6a21\u578b
          <\/em>    <\/em>:param<\/em><\/strong> data:
          <\/em>    <\/em>:param<\/em><\/strong> ix_to_char:
          <\/em>    <\/em>:param<\/em><\/strong> char_to_ix:
          <\/em>    <\/em>:param<\/em><\/strong> num_iterations:
          <\/em>    <\/em>:param<\/em><\/strong> n_a:
          <\/em>    <\/em>:param<\/em><\/strong> dino_names:
          <\/em>    <\/em>:param<\/em><\/strong> vocab_size:
          <\/em>    <\/em>:return<\/em><\/strong>:
          <\/em>    \"\"\"
          <\/em>
          <\/em>    <\/em># \u4ecevocab_size\u4e2d\u83b7\u53d6n_x\u3001n_y
              n_x, n_y = vocab_size, vocab_size

              # \u521d\u59cb\u5316\u53c2\u6570
              parameters = utils.initialize_parameters(n_a, n_x, n_y)

              # \u521d\u59cb\u5316\u635f\u5931
              loss = utils.get_initial_loss(vocab_size, dino_names)

              # \u6784\u5efa\u6050\u9f99\u540d\u79f0\u5217\u8868
              with open(\"dinos.txt\") as f:
                  examples = f.readlines()
              examples = [x.lower().strip() for x in examples]

              # \u6253\u4e71\u5168\u90e8\u7684\u6050\u9f99\u540d\u79f0
              np.random.seed(0)
              np.random.shuffle(examples)

              # \u521d\u59cb\u5316LSTM\u9690\u85cf\u72b6\u6001
              a_prev = np.zeros((n_a, 1))

              # \u5faa\u73af
              for j in range(num_iterations):
                  # \u5b9a\u4e49\u4e00\u4e2a\u8bad\u7ec3\u6837\u672c
                  index = j % len(examples)
                  X = [None] + [char_to_ix[ch] for ch in examples[index]]
                  Y = X[1:] + [char_to_ix[\"\\n\"]]

                  # \u6267\u884c\u5355\u6b65\u4f18\u5316\uff1a\u524d\u5411\u4f20\u64ad -> \u53cd\u5411\u4f20\u64ad -> \u68af\u5ea6\u4fee\u526a -> \u66f4\u65b0\u53c2\u6570
                  # \u9009\u62e9\u5b66\u4e60\u7387\u4e3a0.01
                  curr_loss, gradients, a_prev = optimize(X, Y, a_prev, parameters)

                  # \u4f7f\u7528\u5ef6\u8fdf\u6765\u4fdd\u6301\u635f\u5931\u5e73\u6ed1,\u8fd9\u662f\u4e3a\u4e86\u52a0\u901f\u8bad\u7ec3\u3002
                  loss = utils.smooth(loss, curr_loss)

                  # \u6bcf2000\u6b21\u8fed\u4ee3\uff0c\u901a\u8fc7sample()\u751f\u6210\u201c\\n\u201d\u5b57\u7b26\uff0c\u68c0\u67e5\u6a21\u578b\u662f\u5426\u5b66\u4e60\u6b63\u786e
                  if j % 2000 == 0:
                      print(\"\u7b2c\" + str(j + 1) + \"\u6b21\u8fed\u4ee3\uff0c\u635f\u5931\u503c\u4e3a\uff1a\" + str(loss))

                      seed = 0
                      for name in range(dino_names):
                          # \u91c7\u6837
                          sampled_indices = sample(parameters, char_to_ix, seed)
                          utils.print_sample(sampled_indices, ix_to_char)

                          # \u4e3a\u4e86\u5f97\u5230\u76f8\u540c\u7684\u6548\u679c\uff0c\u968f\u673a\u79cd\u5b50+1
                          seed += 1

                      print(\"\\n\")
              return parameters


          # \u5f00\u59cb\u65f6\u95f4
          start_time = time.clock()
          # \u5f00\u59cb\u8bad\u7ec3
          parameters = model(data, ix_to_char, char_to_ix, num_iterations=3500)
          # \u7ed3\u675f\u65f6\u95f4
          end_time = time.clock()
          # \u8ba1\u7b97\u65f6\u5dee
          minium = end_time - start_time
          print(\"\u6267\u884c\u4e86\uff1a\" + str(int(minium \/ 60)) + \"\u5206\" + str(int(minium % 60)) + \"\u79d2\")
          <\/pre>\n\n\n
          \u8f93\u51fa\uff1a<\/p>\n\n\n
          \u7b2c1\u6b21\u8fed\u4ee3\uff0c\u635f\u5931\u503c\u4e3a\uff1a23.0873360855\nNkzxwtdmfqoeyhsqwasjkjvu\nKneb\nKzxwtdmfqoeyhsqwasjkjvu\nNeb\nZxwtdmfqoeyhsqwasjkjvu\nEb\nXwtdmfqoeyhsqwasjkjvu\n\u7b2c2001\u6b21\u8fed\u4ee3\uff0c\u635f\u5931\u503c\u4e3a\uff1a27.8841604914\nLiusskeomnolxeros\nHmdaairus\nHytroligoraurus\nLecalosapaus\nXusicikoraurus\nAbalpsamantisaurus\nTpraneronxeros\n\u6267\u884c\u4e86\uff1a0\u52063\u79d2<\/code><\/pre>\n\n\n
          \u6211\u4eec\u53ef\u4ee5\u770b\u5230\u6548\u679c\u660e\u663e\u53d8\u597d\u4e86\uff0c\u66f4\u50cf\u4e00\u4e2a\u540d\u5b57\u4e86\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
          \u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u7b2c\u4e94\u8bfe\u7b2c\u4e00\u5468\uff1a\u5e8f\u5217\u6a21\u578b 1.\u4e3a\u4ec0\u4e48\u9009\u62e9\u5e8f\u5217\u6a21\u578b \u5f88\u591a\u4efb\u52a1\u4e2d\u6570\u636e\u5bf9\u5148\u540e\u654f\u611f\uff0c\u4e3e\u4e2a\u7b80\u5355\u7684\u4f8b […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_mi_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[7],"tags":[],"views":8297,"_links":{"self":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts\/3381"}],"collection":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/comments?post=3381"}],"version-history":[{"count":0,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts\/3381\/revisions"}],"wp:attachment":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/media?parent=3381"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/categories?post=3381"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/tags?post=3381"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}