{"topic":{"id":11826,"title":"optim lab 中一个很有趣的现象（与分数无关），求大神解答","slug":"topic","categoryId":58,"createdAt":"2024-04-27 04:05:29.82143","bumpedAt":"2024-06-09 12:41:50.588999","views":556,"postsCount":6,"highestPostNumber":6,"likeCount":6,"closed":false,"pinned":false,"pinnedGlobally":false,"pinnedAt":null,"pinnedUntil":null,"isBanner":false,"excerpt":"实验已经做完了，但是碰见一个十分神奇的现象，求大神解答：我在 CLI 中 make grade 的成绩一直是 100/100 闲着无聊将 CLI 中我的键盘输入 / 打分脚本的输出复制进 CLI_output.txt 中：显示所有分数是 100/100 使用./grade >> test.txt 将 CLI 中的结果重定向到 test.txt 中，发现有些的分数变成了 75/100 了具体的操作是：在命令行中敲击了 5…","deletedAt":null,"isAnonymous":false,"category":{"id":58,"name":"计算机系统导论","slug":"ics","color":"3AB54A"},"author":{"id":1647,"username":"carrot","name":"carrot","avatarUrl":"/uploads/image/ab/5a/ab5a7d5a15a8ec5d8851ab15b4c7ccb6183bfcad4cb05cbb2eb502862d84b2c9.avif"},"tags":[]},"posts":[{"id":30915,"postNumber":1,"raw":"实验已经做完了，但是碰见一个十分神奇的现象，求大神解答：\n\n1. 我在 CLI 中 make grade 的成绩一直是 100/100\n2. 闲着无聊将 CLI 中我的键盘输入 / 打分脚本的输出复制进 CLI_output.txt 中：显示所有分数是 100/100\n3. 使用./grade >> test.txt 将 CLI 中的结果重定向到 test.txt 中，发现有些的分数变成了 75/100 了\n\n具体的操作是：在命令行中敲击了 5 次 make grade 与 5 次./grade >> test.txt\n\n- CLI_output.txt 中的分数是 5 个 100/100（是复制进去的，所以很显然没变化）\n- test.txt 中分数分别是 75/100, 75/100, 100/100, 75/100, 100/100\n\n很想了解一下这里的打分程序 grader 是怎么运行的，主要是不太熟悉逆向工程，因此没有对 optim/grader 二进制文件进行操作\n\n[CLI_output.txt|attachment](/uploads/blob/eb/04/eb04861ef39d8ca40ff5615638cf572cb5e3367b06badcd7be4fc668eaa4e310.txt) (3.4 KB)\n[test.txt|attachment](/uploads/blob/d1/2a/d12a68271147bb5bfde0a6883d9214a6fad8bd85ac6736415eb68ede9ef9c1ea.txt) (2.8 KB)","cooked":"

实验已经做完了，但是碰见一个十分神奇的现象，求大神解答：

我在 CLI 中 make grade 的成绩一直是 100/100
闲着无聊将 CLI 中我的键盘输入 / 打分脚本的输出复制进 CLI_output.txt 中：显示所有分数是 100/100
使用./grade >> test.txt 将 CLI 中的结果重定向到 test.txt 中，发现有些的分数变成了 75/100 了

具体的操作是：在命令行中敲击了 5 次 make grade 与 5 次./grade >> test.txt

CLI_output.txt 中的分数是 5 个 100/100（是复制进去的，所以很显然没变化）
test.txt 中分数分别是 75/100, 75/100, 100/100, 75/100, 100/100

很想了解一下这里的打分程序 grader 是怎么运行的，主要是不太熟悉逆向工程，因此没有对 optim/grader 二进制文件进行操作

CLI_output.txt|attachment (3.4 KB)
\ntest.txt|attachment (2.8 KB)

\n","createdAt":"2024-04-27 04:05:29.927961","updatedAt":"2024-05-11 16:26:39.08651","replyToPostNumber":null,"replyCount":0,"likeCount":1,"wordCount":227,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":1047,"username":"anonymous_coward_old","name":"Anonymous Coward Old"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":1647,"username":"carrot","name":"carrot","avatarUrl":"/uploads/image/ab/5a/ab5a7d5a15a8ec5d8851ab15b4c7ccb6183bfcad4cb05cbb2eb502862d84b2c9.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":30917,"postNumber":2,"raw":"补充：\n所有从 100->75 都是在最后一个环节\n\n=============== Test 4: Measuring poly_optim() ================\nMeasured CPE of poly_optim(): \n\n从 25 -> 0","cooked":"

补充：
\n所有从 100->75 都是在最后一个环节

=============== Test 4: Measuring poly_optim() ================
\nMeasured CPE of poly_optim():

从 25 -> 0

\n","createdAt":"2024-04-27 04:13:25.711302","updatedAt":"2024-04-27 04:13:25.711302","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":30,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":1647,"username":"carrot","name":"carrot","avatarUrl":"/uploads/image/ab/5a/ab5a7d5a15a8ec5d8851ab15b4c7ccb6183bfcad4cb05cbb2eb502862d84b2c9.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":32019,"postNumber":3,"raw":"grader 是怎么运行的：查看 `Makefile` 可以知道 `grade` 是 `main.c`, `poly.c` 和 `measure_time_std.o` 编译而来的，其中 `measure_time_std.o` 是测量执行时间的标准实现编译后的目标文件。执行 `./grade` 时执行的便是 `main.c` 中的代码。\n\n现象的解释：首先我没能复现出这个现象，我尝试以几种方式修改 `measure_time()` 函数后，要么都是 100 分要么都是 75 分。\n我给出我对这一现象的原因的猜测如下：输出重定向到文件后最后一个 test 不再得分，并且测得的 CPE 都明显偏大，初步猜测是测量时间的函数没有预热 cache，而输出重定向到文件的操作比直接输出到终端要更复杂，从而使 cache miss 增加，测得的 CPE 就偏大。但这只是猜测的结论、很可能不正确，如果想进一步讨论可以将你的 `measure_time()` 实现私发给助教（我），我尝试复现现象之后再进行分析。","cooked":"

grader 是怎么运行的：查看 Makefile 可以知道 grade 是 main.c, poly.c 和 measure_time_std.o 编译而来的，其中 measure_time_std.o 是测量执行时间的标准实现编译后的目标文件。执行 ./grade 时执行的便是 main.c 中的代码。

现象的解释：首先我没能复现出这个现象，我尝试以几种方式修改 measure_time() 函数后，要么都是 100 分要么都是 75 分。
\n我给出我对这一现象的原因的猜测如下：输出重定向到文件后最后一个 test 不再得分，并且测得的 CPE 都明显偏大，初步猜测是测量时间的函数没有预热 cache，而输出重定向到文件的操作比直接输出到终端要更复杂，从而使 cache miss 增加，测得的 CPE 就偏大。但这只是猜测的结论、很可能不正确，如果想进一步讨论可以将你的 measure_time() 实现私发给助教（我），我尝试复现现象之后再进行分析。

\n","createdAt":"2024-05-11 16:26:18.335158","updatedAt":"2024-05-11 16:26:18.335158","replyToPostNumber":null,"replyCount":0,"likeCount":2,"wordCount":276,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":2,"users":[{"id":1047,"username":"anonymous_coward_old","name":"Anonymous Coward Old"},{"id":1647,"username":"carrot","name":"carrot"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":3373,"username":"zxia","name":null,"avatarUrl":null,"admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":35249,"postNumber":4,"raw":"[a.c|attachment](/uploads/blob/24/49/244946ed6318e9210be8afffe509d1ec03c1144b8b5da08d7bbc85783b871216.c) (716 字节)\n原问题对应的 measure_time() 函数内，在测量时间之前，已经用 poly 预热一遍了，不知道是否是处理不当 / 还是说有其他错误？","cooked":"

a.c|attachment (716 字节)
\n原问题对应的 measure_time() 函数内，在测量时间之前，已经用 poly 预热一遍了，不知道是否是处理不当 / 还是说有其他错误？

\n","createdAt":"2024-06-09 12:31:58.79955","updatedAt":"2024-06-09 12:31:58.79955","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":54,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":1647,"username":"carrot","name":"carrot","avatarUrl":"/uploads/image/ab/5a/ab5a7d5a15a8ec5d8851ab15b4c7ccb6183bfcad4cb05cbb2eb502862d84b2c9.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":35250,"postNumber":5,"raw":"我看到了你的源码，你的 measure_time 函数在两次 clock_gettime() 之间只执行了一次 poly 函数，可能波动会比较大，多执行几次 poly 函数的话波动会被多次的执行均摊，测量结果会更稳定。\n\n我放几张图就很清晰了：\n\n原版：\n![original](/uploads/image/24/48/244879ae3d5b0068d8bd7ab4c5d0d93fc619bb15566637e8cdfbecc0bdd8e61b.avif)\n\n执行 50 次 poly 取均值：\n![截屏 2024-06-09 20.30.07|690x475](/uploads/image/a5/4a/a54a143e53859b0851cf0eb00a76562fc3fc7282aa9fa1cb58a1a16acb16d03a.avif)\n\n可以看到取均值的方法得到的数据线性化程度更好，我个人认为这足以说明这样测得的数据更可信。\n\n题外话，可以看到图中有一些离群点，我在打分程序（main.c）中使用了 RANSAC 算法尝试减少离群点带来的影响。但实际上由于离群点数量远小于样本数量，不用 RANSAC 算法结果也已经足够令人满意（如上图所示）。","cooked":"

我看到了你的源码，你的 measure_time 函数在两次 clock_gettime() 之间只执行了一次 poly 函数，可能波动会比较大，多执行几次 poly 函数的话波动会被多次的执行均摊，测量结果会更稳定。

我放几张图就很清晰了：

原版：
\n $\"original\"$

执行 50 次 poly 取均值：
\n $\"截屏$

可以看到取均值的方法得到的数据线性化程度更好，我个人认为这足以说明这样测得的数据更可信。

题外话，可以看到图中有一些离群点，我在打分程序（main.c）中使用了 RANSAC 算法尝试减少离群点带来的影响。但实际上由于离群点数量远小于样本数量，不用 RANSAC 算法结果也已经足够令人满意（如上图所示）。

\n","createdAt":"2024-06-09 12:34:07.684792","updatedAt":"2024-06-09 12:34:07.684792","replyToPostNumber":null,"replyCount":0,"likeCount":3,"wordCount":229,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":3,"users":[{"id":1647,"username":"carrot","name":"carrot"},{"id":3735,"username":"1834325559","name":"这扯淡的青春"},{"id":4668,"username":"IO","name":null}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":3373,"username":"zxia","name":null,"avatarUrl":null,"admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}},{"id":35251,"postNumber":6,"raw":"好的明白，谢谢老师！","cooked":"

好的明白，谢谢老师！

\n","createdAt":"2024-06-09 12:41:50.588999","updatedAt":"2024-06-09 12:41:50.588999","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":10,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":1647,"username":"carrot","name":"carrot","avatarUrl":"/uploads/image/ab/5a/ab5a7d5a15a8ec5d8851ab15b4c7ccb6183bfcad4cb05cbb2eb502862d84b2c9.avif","admin":false,"moderator":false,"signatureRaw":null,"signatureCooked":null}}],"hasMore":false,"lastReadPostNumber":3,"hasUrlPostNumber":true}