{"id":1057577,"date":"2024-12-31T15:09:15","date_gmt":"2024-12-31T07:09:15","guid":{"rendered":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/1057577.html"},"modified":"2024-12-31T15:09:17","modified_gmt":"2024-12-31T07:09:17","slug":"python%e6%89%8b%e7%bb%98%e4%b8%ad%e6%a2%af%e5%ba%a6%e5%a6%82%e4%bd%95%e5%bd%92%e4%b8%80%e5%8c%96","status":"publish","type":"post","link":"https:\/\/docs.pingcode.com\/ask\/1057577.html","title":{"rendered":"python\u624b\u7ed8\u4e2d\u68af\u5ea6\u5982\u4f55\u5f52\u4e00\u5316"},"content":{"rendered":"

\"python\u624b\u7ed8\u4e2d\u68af\u5ea6\u5982\u4f55\u5f52\u4e00\u5316\"<\/p>\n

\u68af\u5ea6\u5f52\u4e00\u5316\u7684\u4e3b\u8981\u65b9\u6cd5\u6709\uff1a\u4f7f\u7528\u6700\u5c0f-\u6700\u5927\u5f52\u4e00\u5316\u3001\u6807\u51c6\u5316\u3001L2\u6b63\u5219\u5316<\/strong>\u3002\u5176\u4e2d\u6700\u5c0f-\u6700\u5927\u5f52\u4e00\u5316\u662f\u5c06\u68af\u5ea6\u503c\u7f29\u653e\u5230\u4e00\u4e2a\u6307\u5b9a\u7684\u8303\u56f4\uff08\u901a\u5e38\u662f0\u52301\uff09\uff0c\u901a\u8fc7\u7ebf\u6027\u53d8\u6362\u7684\u65b9\u6cd5\u5b9e\u73b0\u3002\u6807\u51c6\u5316\u5219\u662f\u901a\u8fc7\u51cf\u53bb\u5747\u503c\u5e76\u9664\u4ee5\u6807\u51c6\u5dee\u6765\u4f7f\u6570\u636e\u7b26\u5408\u6807\u51c6\u6b63\u6001\u5206\u5e03\uff08\u5747\u503c\u4e3a0\uff0c\u6807\u51c6\u5dee\u4e3a1\uff09\u3002L2\u6b63\u5219\u5316\u662f\u5728\u68af\u5ea6\u66f4\u65b0\u8fc7\u7a0b\u4e2d\u52a0\u5165\u4e00\u4e2a\u6b63\u5219\u9879\uff0c\u4ece\u800c\u9632\u6b62\u6a21\u578b\u8fc7\u62df\u5408\u3002\u4e0b\u9762\u6211\u4eec\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u548c\u8ba8\u8bba\u8fd9\u4e9b\u65b9\u6cd5\u3002<\/p>\n<\/p>\n

\u4e00\u3001\u6700\u5c0f-\u6700\u5927\u5f52\u4e00\u5316<\/h3>\n<\/p>\n

\u6700\u5c0f-\u6700\u5927\u5f52\u4e00\u5316\u901a\u8fc7\u7ebf\u6027\u53d8\u6362\u5c06\u68af\u5ea6\u503c\u7f29\u653e\u5230\u4e00\u4e2a\u6307\u5b9a\u7684\u8303\u56f4\uff08\u901a\u5e38\u662f0\u52301\uff09\u3002\u8fd9\u79cd\u65b9\u6cd5\u7684\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n

[ X_{norm} = \\frac{X – X_{min}}{X_{max} – X_{min}} ]<\/p>\n<\/p>\n

\u8fd9\u79cd\u65b9\u6cd5\u7684\u4f18\u70b9\u662f\u7b80\u5355\u76f4\u89c2\uff0c\u4e14\u80fd\u4fdd\u7559\u539f\u59cb\u6570\u636e\u7684\u5206\u5e03\u7279\u6027\u3002<\/p>\n<\/p>\n

1\u3001\u8ba1\u7b97\u6700\u5c0f\u503c\u548c\u6700\u5927\u503c<\/h4>\n<\/p>\n

\u9996\u5148\uff0c\u8ba1\u7b97\u68af\u5ea6\u77e9\u9635\u7684\u6700\u5c0f\u503c\u548c\u6700\u5927\u503c\uff1a<\/p>\n<\/p>\n

import numpy as np<\/p>\n
\u793a\u4f8b\u68af\u5ea6\u77e9\u9635<\/strong><\/h2>\n
gradient_matrix = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]])<\/p>\n
X_min = np.min(gradient_matrix)<\/p>\n
X_max = np.max(gradient_matrix)<\/p>\n
<\/code><\/pre>\n<\/p>\n
2\u3001\u5e94\u7528\u6700\u5c0f-\u6700\u5927\u5f52\u4e00\u5316<\/h4>\n<\/p>\n
\u4f7f\u7528\u4e0a\u8ff0\u516c\u5f0f\u5c06\u68af\u5ea6\u77e9\u9635\u5f52\u4e00\u5316\uff1a<\/p>\n<\/p>\n
X_norm = (gradient_matrix - X_min) \/ (X_max - X_min)<\/p>\n
print(X_norm)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u8fd9\u6837\uff0c\u6211\u4eec\u5c31\u5c06\u68af\u5ea6\u77e9\u9635\u7684\u503c\u7f29\u653e\u5230\u4e860\u52301\u4e4b\u95f4\u3002<\/p>\n<\/p>\n
\u4e8c\u3001\u6807\u51c6\u5316<\/h3>\n<\/p>\n
\u6807\u51c6\u5316\u7684\u76ee\u7684\u662f\u4f7f\u6570\u636e\u7684\u5747\u503c\u4e3a0\uff0c\u6807\u51c6\u5dee\u4e3a1\uff0c\u4f7f\u5f97\u6570\u636e\u7b26\u5408\u6807\u51c6\u6b63\u6001\u5206\u5e03\u3002\u5176\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n
[ X_{norm} = \\frac{X – \\mu}{\\sigma} ]<\/p>\n<\/p>\n
\u5176\u4e2d\uff0c(\\mu)\u662f\u6570\u636e\u7684\u5747\u503c\uff0c(\\sigma)\u662f\u6570\u636e\u7684\u6807\u51c6\u5dee\u3002<\/p>\n<\/p>\n
1\u3001\u8ba1\u7b97\u5747\u503c\u548c\u6807\u51c6\u5dee<\/h4>\n<\/p>\n
\u9996\u5148\uff0c\u8ba1\u7b97\u68af\u5ea6\u77e9\u9635\u7684\u5747\u503c\u548c\u6807\u51c6\u5dee\uff1a<\/p>\n<\/p>\n
X_mean = np.mean(gradient_matrix)<\/p>\n
X_std = np.std(gradient_matrix)<\/p>\n
<\/code><\/pre>\n<\/p>\n
2\u3001\u5e94\u7528\u6807\u51c6\u5316<\/h4>\n<\/p>\n
\u4f7f\u7528\u4e0a\u8ff0\u516c\u5f0f\u5c06\u68af\u5ea6\u77e9\u9635\u6807\u51c6\u5316\uff1a<\/p>\n<\/p>\n
X_norm = (gradient_matrix - X_mean) \/ X_std<\/p>\n
print(X_norm)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e09\u3001L2\u6b63\u5219\u5316<\/h3>\n<\/p>\n
L2\u6b63\u5219\u5316\u901a\u8fc7\u5728\u68af\u5ea6\u66f4\u65b0\u8fc7\u7a0b\u4e2d\u52a0\u5165\u4e00\u4e2a\u6b63\u5219\u9879\uff0c\u4ee5\u9632\u6b62\u6a21\u578b\u8fc7\u62df\u5408\u3002\u5176\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n
[ X_{norm} = \\frac{X}{|X|_2} ]<\/p>\n<\/p>\n
\u5176\u4e2d\uff0c(|X|_2)\u662f\u68af\u5ea6\u77e9\u9635\u7684L2\u8303\u6570\u3002<\/p>\n<\/p>\n
1\u3001\u8ba1\u7b97L2\u8303\u6570<\/h4>\n<\/p>\n
\u9996\u5148\uff0c\u8ba1\u7b97\u68af\u5ea6\u77e9\u9635\u7684L2\u8303\u6570\uff1a<\/p>\n<\/p>\n
L2_norm = np.linalg.norm(gradient_matrix)<\/p>\n
<\/code><\/pre>\n<\/p>\n
2\u3001\u5e94\u7528L2\u6b63\u5219\u5316<\/h4>\n<\/p>\n
\u4f7f\u7528\u4e0a\u8ff0\u516c\u5f0f\u5c06\u68af\u5ea6\u77e9\u9635\u8fdb\u884cL2\u5f52\u4e00\u5316\uff1a<\/p>\n<\/p>\n
X_norm = gradient_matrix \/ L2_norm<\/p>\n
print(X_norm)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u56db\u3001\u9009\u62e9\u5f52\u4e00\u5316\u65b9\u6cd5\u7684\u8003\u8651\u56e0\u7d20<\/h3>\n<\/p>\n
\u5728\u9009\u62e9\u68af\u5ea6\u5f52\u4e00\u5316\u65b9\u6cd5\u65f6\uff0c\u9700\u8981\u8003\u8651\u4ee5\u4e0b\u56e0\u7d20\uff1a<\/p>\n<\/p>\n