{"id":204592,"date":"2025-05-29T10:18:10","date_gmt":"2025-05-29T02:18:10","guid":{"rendered":"https:\/\/server.hk\/cnblog\/204592\/"},"modified":"2025-05-29T10:18:10","modified_gmt":"2025-05-29T02:18:10","slug":"pytorch-%e4%b8%ad%e7%9a%84-isclose","status":"publish","type":"post","link":"https:\/\/server.hk\/cnblog\/204592\/","title":{"rendered":"PyTorch \u4e2d\u7684 isclose"},"content":{"rendered":"<p><b><\/b>     <\/p>\n<h1>PyTorch \u4e2d\u7684 isclose<\/h1>\n<p>\u672c\u7bc7\u6587\u7ae0\u5411\u5927\u5bb6\u4ecb\u7ecd\u300aPyTorch \u4e2d\u7684 isclose\u300b\uff0c\u4e3b\u8981\u5305\u62ec\uff0c\u5177\u6709\u4e00\u5b9a\u7684\u53c2\u8003\u4ef7\u503c\uff0c\u9700\u8981\u7684\u670b\u53cb\u53ef\u4ee5\u53c2\u8003\u4e00\u4e0b\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.17golang.com\/uploads\/20241111\/1731301585673190d1bf29f.jpg\" class=\"aligncenter\"><\/p>\n<p>\u8bf7\u6211\u559d\u676f\u5496\u5561<\/p>\n<p>*\u5907\u5fd8\u5f55\uff1a<\/p>\n<ul>\n<li> \u6211\u7684\u5e16\u5b50\u89e3\u91ca\u4e86 equal()\u3001eq() \u548c ne()\u3002<\/li>\n<li> \u6211\u7684\u5e16\u5b50\u89e3\u91ca\u4e86 gt() \u548c lt()\u3002<\/li>\n<li> \u6211\u7684\u5e16\u5b50\u89e3\u91ca\u4e86 ge() \u548c le()\u3002<\/li>\n<\/ul>\n<p>isclose() \u53ef\u4ee5\u68c0\u67e5\u7b2c\u4e00\u4e2a 0d \u6216\u66f4\u591a d \u5f20\u91cf\u7684\u96f6\u4e2a\u6216\u591a\u4e2a\u5143\u7d20\u662f\u5426\u7b49\u4e8e\u6216\u63a5\u8fd1\u7b49\u4e8e\u7b2c\u4e8c\u4e2a 0d \u6216\u66f4\u591a d \u5f20\u91cf\u7684\u96f6\u4e2a\u6216\u591a\u4e2a\u5143\u7d20\uff0c\u5f97\u5230 0d \u6216\u66f4\u591a\u96f6\u4e2a\u6216\u591a\u4e2a\u5143\u7d20\u7684 d \u5f20\u91cf\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<p>*\u5907\u5fd8\u5f55\uff1a<\/p>\n<ul>\n<li> isclose() \u53ef\u4ee5\u4e0e torch \u6216\u5f20\u91cf\u4e00\u8d77\u4f7f\u7528\u3002<\/li>\n<li>\u7b2c\u4e00\u4e2a\u53c2\u6570\uff08\u8f93\u5165\uff09\u4f7f\u7528 torch \u6216\u4f7f\u7528\u5f20\u91cf\uff08\u5fc5\u9700\u7c7b\u578b\uff1aint\u3001float\u3001complex \u6216 bool \u7684\u5f20\u91cf\uff09\u3002<\/li>\n<li>\u5e26\u6709 torch \u7684\u7b2c\u4e8c\u4e2a\u53c2\u6570\u6216\u5e26\u6709\u5f20\u91cf\u7684\u7b2c\u4e00\u4e2a\u53c2\u6570\u662f\u5176\u4ed6\uff08\u5fc5\u9700\u7c7b\u578b\uff1aint\u3001float\u3001complex \u6216 bool \u7684\u5f20\u91cf\uff09\u3002<\/li>\n<li>\u5e26\u6709 torch \u7684\u7b2c\u4e09\u4e2a\u53c2\u6570\u6216\u5e26\u6709\u5f20\u91cf\u7684\u7b2c\u4e8c\u4e2a\u53c2\u6570\u662f rtol(optional-default:1e-05-type:float)\u3002<\/li>\n<li>\u5e26\u6709 torch \u7684\u7b2c\u56db\u4e2a\u53c2\u6570\u6216\u5e26\u6709\u5f20\u91cf\u7684\u7b2c\u4e09\u4e2a\u53c2\u6570\u662f atol(optional-default:1e-08-type:float)\u3002<\/li>\n<li>\u5e26\u6709 torch \u7684\u7b2c\u4e94\u4e2a\u53c2\u6570\u6216\u5e26\u6709\u5f20\u91cf\u7684\u7b2c\u56db\u4e2a\u53c2\u6570\u662f equal_nan(optional-default:false-type:bool)\uff1a *\u5907\u6ce8\uff1a\n<ul>\n<li>\u5982\u679c\u4e3a true\uff0c\u5219 nan \u548c nan \u8fd4\u56de true\u3002<\/li>\n<li>\u57fa\u672c\u4e0a\uff0cnan \u548c nan \u8fd4\u56de false\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u516c\u5f0f\u4e3a |\u8f93\u5165 &#8211; \u5176\u4ed6| &lt;= rtol x |\u5176\u4ed6| + \u963f\u6258\u5c14\u3002 <\/li>\n<\/ul>\n<pre>import torch\n\ntensor1 = torch.tensor([1.00001001, 1.00000996, 1.00000995, torch.nan])\ntensor2 = torch.tensor([1., 1., 1., torch.nan])\n\ntorch.isclose(input=tensor1, other=tensor2)\ntorch.isclose(input=tensor1, other=tensor2,\n              rtol=1e-05, atol=1e-08, equal_nan=False)\n            # 0.00001   # 0.00000001\ntensor1.isclose(other=tensor2)\ntorch.isclose(input=tensor2, other=tensor1)\n# tensor([False, False, True, False])\n\ntorch.isclose(input=tensor1, other=tensor2, equal_nan=True)\n# tensor([False, False, True, True])\n\ntensor1 = torch.tensor([[1.00001001, 1.00000996],\n                        [1.00000995, torch.nan]])\ntensor2 = torch.tensor([[1., 1.],\n                        [1., torch.nan]])\ntorch.isclose(input=tensor1, other=tensor2)\ntorch.isclose(input=tensor2, other=tensor1)\n# tensor([[False, False],\n#         [True, False]])\n\ntensor1 = torch.tensor([[[1.00001001],\n                         [1.00000996]],\n                        [[1.00000995],\n                         [torch.nan]]])\ntensor2 = torch.tensor([[[1.], [1.]],\n                        [[1.], [torch.nan]]])\ntorch.isclose(input=tensor1, other=tensor2)\ntorch.isclose(input=tensor2, other=tensor1)\n# tensor([[[False], [False]],\n#         [[True], [False]]])\n\ntensor1 = torch.tensor([[1.00001001, 1.00000996],\n                        [1.00000995, torch.nan]])\ntensor2 = torch.tensor([1., 1.])\n\ntorch.isclose(input=tensor1, other=tensor2)\ntorch.isclose(input=tensor2, other=tensor1)\n# tensor([[False, False],\n#         [True, False]])\n\ntensor1 = torch.tensor([[1.00001001, 1.00000996],\n                        [1.00000995, torch.nan]])\ntensor2 = torch.tensor(1.)\n\ntorch.isclose(input=tensor1, other=tensor2)\ntorch.isclose(input=tensor2, other=tensor1)\n# tensor([[False, False],\n#         [True, False]])\n\ntensor1 = torch.tensor([0, 1, 2])\ntensor2 = torch.tensor(1)\n\ntorch.isclose(input=tensor1, other=tensor2)\n# tensor([False, True, False])\n\ntensor1 = torch.tensor([0.+0.j, 1.+0.j, 2.+0.j])\ntensor2 = torch.tensor(1.+0.j)\n\ntorch.isclose(input=tensor1, other=tensor2)\n# tensor([False, True, False])\n\ntensor1 = torch.tensor([False, True, False])\ntensor2 = torch.tensor(True)\n\ntorch.isclose(input=tensor1, other=tensor2)\n# tensor([False, True, False])\n<\/pre>\n<p>\u7ec8\u4e8e\u4ecb\u7ecd\u5b8c\u5566\uff01\u5c0f\u4f19\u4f34\u4eec\uff0c\u8fd9\u7bc7\u5173\u4e8e\u300aPyTorch \u4e2d\u7684 isclose\u300b\u7684\u4ecb\u7ecd\u5e94\u8be5\u8ba9\u4f60\u6536\u83b7\u591a\u591a\u4e86\u5427\uff01\u6b22\u8fce\u5927\u5bb6\u6536\u85cf\u6216\u5206\u4eab\u7ed9\u66f4\u591a\u9700\u8981\u5b66\u4e60\u7684\u670b\u53cb\u5427~\u516c\u4f17\u53f7\u4e5f\u4f1a\u53d1\u5e03\u6587\u7ae0\u76f8\u5173\u77e5\u8bc6\uff0c\u5feb\u6765\u5173\u6ce8\u5427\uff01<\/p>\n<p>      \u7248\u672c\u58f0\u660e \u672c\u6587\u8f6c\u8f7d\u4e8e\uff1adev.to \u5982\u6709\u4fb5\u72af\uff0c\u8bf7\u8054\u7cfb\u5220\u9664<\/p>\n","protected":false},"excerpt":{"rendered":"<p>PyTorch \u4e2d\u7684 isclo&#46;&#46;&#46;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4925],"tags":[],"class_list":["post-204592","post","type-post","status-publish","format-standard","hentry","category-4925"],"_links":{"self":[{"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/posts\/204592","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/comments?post=204592"}],"version-history":[{"count":0,"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/posts\/204592\/revisions"}],"wp:attachment":[{"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/media?parent=204592"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/categories?post=204592"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/server.hk\/cnblog\/wp-json\/wp\/v2\/tags?post=204592"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}