{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.md","pages":[{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","unit":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","unitId":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","chapter":"course","number":1}],"aliases":[{"oldId":"b-1299637c877be362c84f0316","targetId":"b-1299637c877be362c84f0316","page":"reading-opening","operation":"move","beforeSHA256":"7d57f9e628741b07c2704a32d4b4c2e5ca44d512d696e3562c18003cb33df7d1","afterSHA256":"7d57f9e628741b07c2704a32d4b4c2e5ca44d512d696e3562c18003cb33df7d1","reason":"Same source body in authored semantic page"},{"oldId":"b-63e4c677ec72659b27ce9ef5","targetId":"b-63e4c677ec72659b27ce9ef5","page":"reading-opening","operation":"move","beforeSHA256":"4ccf0bbfa6ba80c69b24a8e406dbc2e523b49d2b5ea60244d5c5b35f5c0e9748","afterSHA256":"4ccf0bbfa6ba80c69b24a8e406dbc2e523b49d2b5ea60244d5c5b35f5c0e9748","reason":"Same source body in authored semantic page"},{"oldId":"b-57570207b465be53fc6dde99","targetId":"b-57570207b465be53fc6dde99","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-e307e84eccfdc4ae63ce05f0","targetId":"b-e307e84eccfdc4ae63ce05f0","page":"reading-opening","operation":"move","beforeSHA256":"75bc3d13109b5a891b82a4daefd668c001abde4ddf12e5ad4e20e6fe674663c6","afterSHA256":"75bc3d13109b5a891b82a4daefd668c001abde4ddf12e5ad4e20e6fe674663c6","reason":"Same source body in authored semantic page"},{"oldId":"b-d1556e276eb7455de408ae5e","targetId":"b-d1556e276eb7455de408ae5e","page":"reading-opening","operation":"move","beforeSHA256":"4ebb0f014e398dc2ae6bf8dbcba937edcb76b2b2c3c0eb1a5daff55a40961d07","afterSHA256":"4ebb0f014e398dc2ae6bf8dbcba937edcb76b2b2c3c0eb1a5daff55a40961d07","reason":"Same source body in authored semantic page"},{"oldId":"b-7834365f208d4877a7ea9fc9","targetId":"b-7834365f208d4877a7ea9fc9","page":"reading-opening","operation":"move","beforeSHA256":"0f2c32f89259030e3551404cdc0f2edfebe5a687a5c075df965b428bffb76dd6","afterSHA256":"0f2c32f89259030e3551404cdc0f2edfebe5a687a5c075df965b428bffb76dd6","reason":"Same source body in authored semantic page"},{"oldId":"b-09f0fe8fe85022d414ae1f91","targetId":"b-09f0fe8fe85022d414ae1f91","page":"reading-opening","operation":"move","beforeSHA256":"9243e65353229b7e0cf09235b672662b032235d9e19d00dd53a13edf6dccc534","afterSHA256":"9243e65353229b7e0cf09235b672662b032235d9e19d00dd53a13edf6dccc534","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-1299637c877be362c84f0316","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-1299637c877be362c84f0316","content":"[LAB 01 · 固定表示：标签预算与读出](/course/notebooks/4dd8605ed9904820919ea9e13d9e7d7c.html)","sha256":"7d57f9e628741b07c2704a32d4b4c2e5ca44d512d696e3562c18003cb33df7d1","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-63e4c677ec72659b27ce9ef5","prerequisites":[],"relatedBlocks":[]},{"id":"b-63e4c677ec72659b27ce9ef5","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-63e4c677ec72659b27ce9ef5","content":"# 源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sha256":"4ccf0bbfa6ba80c69b24a8e406dbc2e523b49d2b5ea60244d5c5b35f5c0e9748","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1299637c877be362c84f0316","nextBlock":"b-57570207b465be53fc6dde99","prerequisites":[{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-57570207b465be53fc6dde99","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-57570207b465be53fc6dde99","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-63e4c677ec72659b27ce9ef5","nextBlock":"b-e307e84eccfdc4ae63ce05f0","prerequisites":[{"id":"b-63e4c677ec72659b27ce9ef5","url":"#b-63e4c677ec72659b27ce9ef5","role":"section-title"},{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e307e84eccfdc4ae63ce05f0","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-e307e84eccfdc4ae63ce05f0","content":"固定CNN与缓存特征、标签预算、kNN/linear和独立评价","sha256":"75bc3d13109b5a891b82a4daefd668c001abde4ddf12e5ad4e20e6fe674663c6","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-57570207b465be53fc6dde99","nextBlock":"b-d1556e276eb7455de408ae5e","prerequisites":[{"id":"b-63e4c677ec72659b27ce9ef5","url":"#b-63e4c677ec72659b27ce9ef5","role":"section-title"},{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d1556e276eb7455de408ae5e","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-d1556e276eb7455de408ae5e","content":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证。本册固定使用同一排列的第 12,000–12,639 张，共 640 张，沿用[L03-LAB 02 参数迁移实验](/course/lab09-transfer.html)的固定划分，分成 **256 训练、128 验证、256 保留评价**。它们不与源训练、源验证重叠。像素除以 255，再整体右移 2 像素，左侧补零、右侧裁切。目标是 `digit >= 5`。","sha256":"4ebb0f014e398dc2ae6bf8dbcba937edcb76b2b2c3c0eb1a5daff55a40961d07","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e307e84eccfdc4ae63ce05f0","nextBlock":"b-7834365f208d4877a7ea9fc9","prerequisites":[{"id":"b-63e4c677ec72659b27ce9ef5","url":"#b-63e4c677ec72659b27ce9ef5","role":"section-title"},{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7834365f208d4877a7ea9fc9","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-7834365f208d4877a7ea9fc9","content":"这检查的是样本索引分离。数据未提供写字者分组，不能声称写字者独立。目标数据仍来自同一 MNIST 数据集；它不是跨领域自然图像实验。","sha256":"0f2c32f89259030e3551404cdc0f2edfebe5a687a5c075df965b428bffb76dd6","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d1556e276eb7455de408ae5e","nextBlock":"b-09f0fe8fe85022d414ae1f91","prerequisites":[{"id":"b-63e4c677ec72659b27ce9ef5","url":"#b-63e4c677ec72659b27ce9ef5","role":"section-title"},{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-09f0fe8fe85022d414ae1f91","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/66d701fc5af94115b7ebe33ce49821d1.html#b-09f0fe8fe85022d414ae1f91","content":"**本册一次性固定协议：**读出采用 L2 单样本特征归一化；近邻 k=5、均匀投票；线性逻辑回归 C=10。它们不是经过充分调优的最佳基线。标签预算使用同一训练顺序的嵌套前缀 32/64/128/256，不根据评价结果重新抽样。保留集只在各臂拟合完、配置锁定之后评价；若阅读结果后再改方案，应另留新的评价集。","sha256":"9243e65353229b7e0cf09235b672662b032235d9e19d00dd53a13edf6dccc534","chapter":{"id":"course","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"unit":{"id":"notebook-blocks/66d701fc5af94115b7ebe33ce49821d1","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证"},"page":{"id":"reading-opening","title":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","number":1},"sectionTitle":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","currentQuestion":"源模型曾用 MNIST 训练排列的前 12,000 张学习十分类；排列末尾 10,000 张作源验证","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7834365f208d4877a7ea9fc9","nextBlock":null,"prerequisites":[{"id":"b-63e4c677ec72659b27ce9ef5","url":"#b-63e4c677ec72659b27ce9ef5","role":"section-title"},{"id":"b-1299637c877be362c84f0316","url":"#b-1299637c877be362c84f0316","role":"problem-context"}],"relatedBlocks":[]}]}