https://blog.youkuaiyun.com/bullingzh-cnhttps://blog.youkuaiyun.com/https://blog.youkuaiyun.com/bulling/article/details/164251818https://blog.youkuaiyun.com/bulling/article/details/164251818bullingTue, 01 Sep 2026 09:43:12 +0800我们需要理解任务：根据用户提供的内容生成≤150字的文章摘要。用户提供了一篇长文，标题是“SHAP并非预测模型，而是模型事后解释工具...”，包含SHAP介绍、可视化、优化、与LIME对比等。摘要需要概括全文核心，不超过150字。 我们需要简洁地抽取关键点：SHAP是基于博弈论Shapley值的模型解释工具，用于归因预测结果到特征。它本身不优化模型，但可通过解释-诊断-优化驱动改进（数据清洗、特征工程、筛选、调参、业务校准）。与LIME相比，SHAP理论严谨、稳定、可加、支持全局聚合，但计算成]]>https://blog.youkuaiyun.com/bulling/article/details/164044149https://blog.youkuaiyun.com/bulling/article/details/164044149bullingTue, 25 Aug 2026 09:18:49 +0800https://blog.youkuaiyun.com/bulling/article/details/163865630https://blog.youkuaiyun.com/bulling/article/details/163865630bullingWed, 19 Aug 2026 00:16:34 +0800https://blog.youkuaiyun.com/bulling/article/details/163865388https://blog.youkuaiyun.com/bulling/article/details/163865388bullingTue, 18 Aug 2026 23:42:02 +0800https://blog.youkuaiyun.com/bulling/article/details/157330201https://blog.youkuaiyun.com/bulling/article/details/157330201bullingMon, 25 May 2026 10:07:56 +0800https://blog.youkuaiyun.com/bulling/article/details/157899820https://blog.youkuaiyun.com/bulling/article/details/157899820bullingSat, 28 Mar 2026 22:46:50 +0800https://blog.youkuaiyun.com/bulling/article/details/159089258https://blog.youkuaiyun.com/bulling/article/details/159089258bullingSat, 28 Mar 2026 22:41:55 +0800https://blog.youkuaiyun.com/bulling/article/details/158583565https://blog.youkuaiyun.com/bulling/article/details/158583565bullingFri, 13 Mar 2026 08:44:02 +0800https://blog.youkuaiyun.com/bulling/article/details/156982271https://blog.youkuaiyun.com/bulling/article/details/156982271bullingSun, 01 Mar 2026 22:14:37 +0800https://blog.youkuaiyun.com/bulling/article/details/156098290https://blog.youkuaiyun.com/bulling/article/details/156098290bullingFri, 19 Dec 2025 23:36:24 +0800https://blog.youkuaiyun.com/bulling/article/details/155676348https://blog.youkuaiyun.com/bulling/article/details/155676348bullingSun, 07 Dec 2025 21:52:31 +0800https://blog.youkuaiyun.com/bulling/article/details/155666732https://blog.youkuaiyun.com/bulling/article/details/155666732bullingSun, 07 Dec 2025 19:14:01 +0800https://blog.youkuaiyun.com/bulling/article/details/155647493https://blog.youkuaiyun.com/bulling/article/details/155647493bullingSat, 06 Dec 2025 21:10:57 +08001) 进行温度软化piTexp⁡ziT∑jexp⁡zjTpiT​∑j​expzj​Texpzi​T​当 (T) 较大时，概率分布会变得更平滑，保留了类别之间的相对相似度信息（即“暗知识”），这比硬标签（one‑hot）提供了更丰富的学习信号。]]>https://blog.youkuaiyun.com/bulling/article/details/155645788https://blog.youkuaiyun.com/bulling/article/details/155645788bullingSat, 06 Dec 2025 21:04:42 +0800https://blog.youkuaiyun.com/bulling/article/details/155428365https://blog.youkuaiyun.com/bulling/article/details/155428365bullingSun, 30 Nov 2025 22:32:27 +0800https://blog.youkuaiyun.com/bulling/article/details/155427213https://blog.youkuaiyun.com/bulling/article/details/155427213bullingSun, 30 Nov 2025 22:06:43 +0800https://blog.youkuaiyun.com/bulling/article/details/139926117https://blog.youkuaiyun.com/bulling/article/details/139926117bullingMon, 24 Jun 2024 14:50:33 +0800https://blog.youkuaiyun.com/bulling/article/details/139921729https://blog.youkuaiyun.com/bulling/article/details/139921729bullingMon, 24 Jun 2024 11:57:56 +0800https://blog.youkuaiyun.com/bulling/article/details/138323711https://blog.youkuaiyun.com/bulling/article/details/138323711bullingTue, 30 Apr 2024 10:06:23 +0800https://blog.youkuaiyun.com/bulling/article/details/138323634https://blog.youkuaiyun.com/bulling/article/details/138323634bullingMon, 29 Apr 2024 19:58:08 +0800