{"id":9409,"date":"2025-06-10T15:54:05","date_gmt":"2025-06-10T07:54:05","guid":{"rendered":"\/?p=9409"},"modified":"2025-06-10T15:54:05","modified_gmt":"2025-06-10T07:54:05","slug":"%e5%bc%a0%e9%87%8f%e7%ae%80%e8%bf%b0","status":"publish","type":"post","link":"\/?p=9409","title":{"rendered":"\u5f20\u91cf\u7b80\u8ff0"},"content":{"rendered":"

\u5f20\u91cf<\/h3>\n

\u5c31\u50cf\u5411\u91cf\u662f\u6807\u91cf\u7684\u63a8\u5e7f\uff0c\u77e9\u9635\u662f\u5411\u91cf\u7684\u63a8\u5e7f\u4e00\u6837\uff0c\u6211\u4eec\u53ef\u4ee5\u6784\u5efa\u5177\u6709\u66f4\u591a\u8f74\u7684\u6570\u636e\u7ed3\u6784\u3002 \u5f20\u91cf\u662f\u63cf\u8ff0\u5177\u6709\u4efb\u610f\u6570\u91cf\u8f74\u7684n\u7ef4\u6570\u7ec4\u7684\u901a\u7528\u65b9\u6cd5\u3002 \u4f8b\u5982\uff0c\u5411\u91cf\u662f\u4e00\u9636\u5f20\u91cf\uff0c\u77e9\u9635\u662f\u4e8c\u9636\u5f20\u91cf\u3002 \u5f20\u91cf\u7528\u7279\u6b8a\u5b57\u4f53\u7684\u5927\u5199\u5b57\u6bcd\u8868\u793a\uff08\u4f8b\u5982\uff0cX\u3001Y\u548cZ\uff09\uff0c \u5b83\u4eec\u7684\u7d22\u5f15\u673a\u5236\u4e0e\u77e9\u9635\u7c7b\u4f3c\u3002<\/p>\n

\u5f53\u6211\u4eec\u5f00\u59cb\u5904\u7406\u56fe\u50cf\u65f6\uff0c\u5f20\u91cf\u5c06\u53d8\u5f97\u66f4\u52a0\u91cd\u8981\uff0c\u56fe\u50cf\u4ee5n\u7ef4\u6570\u7ec4\u5f62\u5f0f\u51fa\u73b0\uff0c \u5176\u4e2d3\u4e2a\u8f74\u5bf9\u5e94\u4e8e\u9ad8\u5ea6\u3001\u5bbd\u5ea6\uff0c\u4ee5\u53ca\u4e00\u4e2a\u901a\u9053<\/em>\uff08channel\uff09\u8f74\uff0c \u7528\u4e8e\u8868\u793a\u989c\u8272\u901a\u9053\uff08\u7ea2\u8272\u3001\u7eff\u8272\u548c\u84dd\u8272\uff09\u3002 <\/p>\n

\u5728Python\u4e2d\uff0c\u5f20\u91cf\uff08Tensor\uff09<\/strong> \u662f\u591a\u7ef4\u6570\u7ec4\u7684\u6269\u5c55\u5f62\u5f0f\uff0c\u662f\u673a\u5668\u5b66\u4e60\u548c\u6df1\u5ea6\u5b66\u4e60\u7684\u6838\u5fc3\u6570\u636e\u7ed3\u6784\u3002\u5b83\u4e0d\u4ec5\u662f\u6570\u503c\u7684\u5bb9\u5668\uff0c\u66f4\u627f\u8f7d\u4e86\u8ba1\u7b97\u56fe\u3001\u786c\u4ef6\u52a0\u901f\u548c\u81ea\u52a8\u5fae\u5206\u7b49\u9ad8\u7ea7\u7279\u6027\u3002\u4ee5\u4e0b\u4ece\u591a\u4e2a\u7ef4\u5ea6\u5168\u9762\u89e3\u6790\u5176\u6982\u5ff5\u4e0e\u5e94\u7528\uff1a<\/p>\n

\n

\ud83d\udd22 \u4e00\u3001\u5f20\u91cf\u7684\u672c\u8d28\uff1a\u591a\u7ef4\u6570\u7ec4<\/strong><\/h3>\n

\u5f20\u91cf\u53ef\u89c6\u4e3a\u6807\u91cf\u3001\u5411\u91cf\u548c\u77e9\u9635\u7684\u9ad8\u7ef4\u63a8\u5e7f<\/strong>\uff0c\u5176\u7ef4\u5ea6\uff08\u79e9\uff09\u51b3\u5b9a\u4e86\u6570\u636e\u7684\u7ed3\u6784\u590d\u6742\u5ea6\uff1a<\/p>\n

    \n
  • 0\u9636\u5f20\u91cf\uff08\u6807\u91cf\uff09<\/strong>\uff1a\u5355\u4e00\u6570\u503c\uff08\u5982 3.14<\/code>\uff09<\/li>\n
  • 1\u9636\u5f20\u91cf\uff08\u5411\u91cf\uff09<\/strong>\uff1a\u4e00\u7ef4\u6570\u7ec4\uff08\u5982 [1, 2, 3]<\/code>\uff09<\/li>\n
  • 2\u9636\u5f20\u91cf\uff08\u77e9\u9635\uff09<\/strong>\uff1a\u4e8c\u7ef4\u6570\u7ec4\uff08\u5982 [[1, 2], [3, 4]]<\/code>\uff09<\/li>\n
  • 3\u9636\u5f20\u91cf\uff08\u7acb\u65b9\u4f53\uff09<\/strong>\uff1a\u4e09\u7ef4\u6570\u7ec4\uff08\u5982RGB\u56fe\u50cf\uff1a\u9ad8\u5ea6\u00d7\u5bbd\u5ea6\u00d7\u989c\u8272\u901a\u9053\uff09<\/li>\n
  • \u66f4\u9ad8\u9636\u5f20\u91cf<\/strong>\uff1a\u5982\u89c6\u9891\u6570\u636e\uff08\u65f6\u95f4\u00d7\u9ad8\u5ea6\u00d7\u5bbd\u5ea6\u00d7\u901a\u9053\uff09<\/li>\n<\/ul>\n
    \n

    \u4f8b\u5982\uff1a\u4e00\u5f20256\u00d7256\u7684\u5f69\u8272\u56fe\u50cf\u662f\u5f62\u72b6\u4e3a (256, 256, 3)<\/code> \u76843\u9636\u5f20\u91cf\u3002<\/p>\n<\/blockquote>\n

    \n

    \u26a1 \u4e8c\u3001\u6838\u5fc3\u7279\u6027\uff1a\u8d85\u8d8a\u4f20\u7edf\u6570\u7ec4\u7684\u4f18\u52bf<\/strong><\/h3>\n

    \u4e0eNumPy\u6570\u7ec4\u76f8\u6bd4\uff0c\u5f20\u91cf\u5728\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u4e2d\u5177\u5907\u72ec\u7279\u80fd\u529b\uff1a<\/p>\n\n\n\n\n\n\n\n\n
    \u7279\u6027<\/strong><\/th>\nNumPy\u6570\u7ec4<\/strong><\/th>\n\u5f20\u91cf<\/strong><\/th>\n<\/tr>\n<\/thead>\n
    \u8ba1\u7b97\u8bbe\u5907<\/strong><\/td>\n\u4ec5\u9650CPU<\/td>\n\u652f\u6301GPU\/TPU\u52a0\u901f\uff08\u8ba1\u7b97\u901f\u5ea6\u63d0\u534710-100\u500d\uff09<\/td>\n<\/tr>\n
    \u81ea\u52a8\u5fae\u5206<\/strong><\/td>\n\u274c \u4e0d\u652f\u6301<\/td>\n\u2705 \u652f\u6301\u68af\u5ea6\u8ba1\u7b97\uff08\u795e\u7ecf\u7f51\u7edc\u8bad\u7ec3\u6838\u5fc3\uff09<\/td>\n<\/tr>\n
    \u6846\u67b6\u96c6\u6210<\/strong><\/td>\n\u901a\u7528\u79d1\u5b66\u8ba1\u7b97<\/td>\n\u6df1\u5ea6\u96c6\u6210\u4e8ePyTorch\/TensorFlow<\/td>\n<\/tr>\n
    \u52a8\u6001\u8ba1\u7b97\u56fe<\/strong><\/td>\n\u274c \u4e0d\u652f\u6301<\/td>\n\u2705 \u652f\u6301\u64cd\u4f5c\u8ffd\u8e2a\u4e0e\u4f18\u5316\uff08\u5982PyTorch\u52a8\u6001\u56fe\uff09<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
    \n

    \u4f8b\uff1a\u5f20\u91cf\u53ef\u901a\u8fc7 .backward()<\/code> \u81ea\u52a8\u8ba1\u7b97\u68af\u5ea6\uff0c\u800cNumPy\u9700\u624b\u52a8\u5b9e\u73b0\u53cd\u5411\u4f20\u64ad\u3002<\/p>\n<\/blockquote>\n

    \n

    \ud83d\udee0\ufe0f \u4e09\u3001\u521b\u5efa\u4e0e\u64cd\u4f5c\uff1a\u4ee3\u7801\u5b9e\u8df5<\/strong><\/h3>\n

    1. \u521b\u5efa\u5f20\u91cf<\/strong><\/h4>\n
      \n
    • \n

      PyTorch\uff1a<\/p>\n

      import torch\nscalar = torch.tensor(3)          # \u6807\u91cf\nmatrix = torch.tensor([[1, 2], [3, 4]])  # \u77e9\u9635\nzeros_3d = torch.zeros((2, 3, 4)) # \u5168\u96f6\u4e09\u7ef4\u5f20\u91cf<\/code><\/pre>\n<\/li>\n
    • \n
      TensorFlow\uff1a<\/p>\n
      import tensorflow as tf\ntensor = tf.constant([[1.0, 2.0], [3.0, 4.0]])  # \u5e38\u91cf\u5f20\u91cf<\/code><\/pre>\n<\/li>\n<\/ul>\n
      2. \u5173\u952e\u64cd\u4f5c<\/strong><\/h4>\n
        \n
      • \n
        \u5f62\u72b6\u53d8\u6362\uff1a<\/p>\n
        # \u5c0612\u5143\u7d20\u5411\u91cf\u8f6c\u4e3a3\u00d74\u77e9\u9635\uff08-1\u8868\u793a\u81ea\u52a8\u8ba1\u7b97\u7ef4\u5ea6\uff09\nreshaped = torch.arange(12).view(3, -1)  # \u7ed3\u679c\uff1a[[0,1,2,3], [4,5,6,7], [8,9,10,11]]<\/code><\/pre>\n<\/li>\n
      • \n
        \u6570\u5b66\u8fd0\u7b97\uff1a<\/p>\n
        a = torch.tensor([[1, 2], [3, 4]])\nb = torch.tensor([[5, 6], [7, 8]])\nc = a + b  # \u9010\u5143\u7d20\u52a0\u6cd5\nd = torch.matmul(a, b)  # \u77e9\u9635\u4e58\u6cd5<\/code><\/pre>\n<\/li>\n
      • \n
        \u81ea\u52a8\u5fae\u5206\uff08PyTorch\u793a\u4f8b\uff09\uff1a<\/p>\n
        x = torch.tensor(2.0, requires_grad=True)\ny = x**2 + 3*x\ny.backward()  # \u8ba1\u7b97\u68af\u5ea6\nprint(x.grad)  # \u8f93\u51fa\uff1a7.0\uff08\u5bfc\u6570 dy\/dx = 2x+3\uff09<\/code><\/pre>\n<\/li>\n<\/ul>\n
        \n
        \ud83c\udf10 \u56db\u3001\u5e94\u7528\u573a\u666f\uff1a\u4ece\u6570\u636e\u5230\u667a\u80fd\u7cfb\u7edf<\/strong><\/h3>\n
          \n
        1. \u6570\u636e\u8868\u793a<\/strong>\n
            \n
          • \u56fe\u50cf\uff1a3\u9636\u5f20\u91cf\uff08\u9ad8\u5ea6\u00d7\u5bbd\u5ea6\u00d7\u901a\u9053\uff09<\/li>\n
          • \u6587\u672c\uff1a\u8bcd\u5d4c\u5165\u77e9\u9635\uff08\u8bcd\u6570\u00d7\u5d4c\u5165\u7ef4\u5ea6\uff09<\/li>\n
          • \u89c6\u9891\uff1a4\u9636\u5f20\u91cf\uff08\u5e27\u6570\u00d7\u9ad8\u5ea6\u00d7\u5bbd\u5ea6\u00d7\u901a\u9053\uff09<\/li>\n<\/ul>\n<\/li>\n
          • \u6df1\u5ea6\u5b66\u4e60\u6a21\u578b<\/strong>\n
              \n
            • \u8f93\u5165\/\u8f93\u51fa<\/strong>\uff1a\u56fe\u50cf\u5206\u7c7b\u4e2d\uff0c\u8f93\u5165\u4e3a\u56fe\u7247\u5f20\u91cf\uff0c\u8f93\u51fa\u4e3a\u9884\u6d4b\u6982\u7387\u5411\u91cf<\/li>\n
            • \u53c2\u6570\u5b58\u50a8<\/strong>\uff1a\u795e\u7ecf\u7f51\u7edc\u6743\u91cd\u4ee5\u5f20\u91cf\u5f62\u5f0f\u4fdd\u5b58\uff08\u5982\u5168\u8fde\u63a5\u5c42\u6743\u91cd\u77e9\u9635\uff09<\/li>\n
            • \u635f\u5931\u8ba1\u7b97<\/strong>\uff1a\u5747\u65b9\u8bef\u5dee\uff08MSE\uff09\u7b49\u635f\u5931\u51fd\u6570\u76f4\u63a5\u64cd\u4f5c\u5f20\u91cf<\/li>\n<\/ul>\n<\/li>\n
            • \u9ad8\u9636\u5e94\u7528<\/strong>\n
                \n
              • \u63a8\u8350\u7cfb\u7edf<\/strong>\uff1a\u7528\u6237-\u7269\u54c1\u4ea4\u4e92\u6570\u636e\u75283\u9636\u5f20\u91cf\u8868\u793a\uff08\u7528\u6237\u00d7\u7269\u54c1\u00d7\u65f6\u95f4\uff09\uff0c\u901a\u8fc7\u5f20\u91cf\u5206\u89e3\u9884\u6d4b\u504f\u597d<\/li>\n
              • \u533b\u5b66\u6210\u50cf<\/strong>\uff1aMRI\u626b\u63cf\u6570\u636e\u4ee53D\/4D\u5f20\u91cf\u5206\u6790\u75c5\u53d8\u7279\u5f81<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n
                \n
                \ud83d\udc8e \u603b\u7ed3<\/strong><\/h3>\n
                Python\u4e2d\u7684\u5f20\u91cf\u662f\u8fde\u63a5\u6570\u636e\u4e0e\u667a\u80fd\u7b97\u6cd5\u7684\u6865\u6881<\/strong>\uff1a<\/p>\n
                  \n
                • \u57fa\u7840<\/strong>\uff1a\u591a\u7ef4\u6570\u7ec4\u7684\u7edf\u4e00\u8868\u793a\uff0c\u652f\u6301\u6807\u91cf\u2192\u9ad8\u7ef4\u6570\u636e\uff1b<\/li>\n
                • \u8fdb\u9636<\/strong>\uff1aGPU\u52a0\u901f\u4e0e\u81ea\u52a8\u5fae\u5206\u8d4b\u80fd\u590d\u6742\u6a21\u578b\u8bad\u7ec3\uff1b<\/li>\n
                • \u5b9e\u8df5<\/strong>\uff1a\u901a\u8fc7PyTorch\/TensorFlow\u64cd\u4f5c\uff0c\u5b9e\u73b0\u4ece\u6570\u636e\u9884\u5904\u7406\u5230\u6a21\u578b\u90e8\u7f72\u7684\u5168\u6d41\u7a0b\u3002<\/li>\n<\/ul>\n
                  \n
                  \u63d0\u793a\uff1a\u521d\u5b66\u8005\u5efa\u8bae\u4ecePyTorch\u5f20\u91cf\u64cd\u4f5c<\/strong>\u5165\u624b\uff0c\u7ed3\u5408\u5b9e\u9645\u9879\u76ee<\/strong>\uff08\u5982\u624b\u5199\u6570\u5b57\u8bc6\u522b\uff09\u6df1\u5316\u7406\u89e3\u3002<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"
                  \u5f20\u91cf \u5c31\u50cf\u5411\u91cf\u662f\u6807\u91cf\u7684\u63a8\u5e7f\uff0c\u77e9\u9635\u662f\u5411\u91cf\u7684\u63a8\u5e7f\u4e00\u6837\uff0c\u6211\u4eec\u53ef\u4ee5\u6784\u5efa\u5177\u6709\u66f4\u591a\u8f74\u7684\u6570\u636e\u7ed3\u6784\u3002 \u5f20\u91cf\u662f\u63cf\u8ff0\u5177\u6709\u4efb\u610f\u6570\u91cf\u8f74\u7684n\u7ef4\u6570\u7ec4\u7684\u901a\u7528\u65b9\u6cd5\u3002 \u4f8b   \u2026 …<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ngg_post_thumbnail":0},"categories":[224],"tags":[],"_links":{"self":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9409"}],"collection":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9409"}],"version-history":[{"count":1,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9409\/revisions"}],"predecessor-version":[{"id":9410,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/9409\/revisions\/9410"}],"wp:attachment":[{"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9409"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9409"},{"taxonomy":"post_tag","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9409"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}