2023-01-03更新：
第一题我后来想到可以用Symbolic Method求解，参考后来写的英文blog: Symbolic Method to solve a probability problem.
刷知乎刷到一个问题
https://www.zhihu.com/question/419625606
咱看了看一时间想试试看。
Continue reading网传电子系概率论期末解答
1
2020九月·略不世出
题记
足下涉西河，虏魏王，禽夏说，引兵下井陉，诛成安君，徇赵，胁燕，定齐，南摧楚人之兵二十万，东杀龙且，西向以报，此所谓功无二于天下，而略不世出者也。
——《史记 · 淮阴侯列传》
计划
- 继续研究PAIP争取完成逻辑编程的部分
- 继续研究Logic of Science，争取把第一部分弄扎实
- 严格执行训练、控体重
- 有时间看一些非技术类的书籍
总结(2020.09.30)
- PAIP逻辑编程搞了一部分，后面的例子兴趣不是很大
- Logic of Science复习了前三章，并给出了不放回的抽样公式
- 体重有点难以启齿，摔跤技术倒是提升了
- 非技术类书看了《沉默的真相》和原著
Symbolic Mathematics in Erlang
This is Chapter 8 of the book PAIP. It still uses pattern-match and rules to design several simple algorithms to do symbolic math simplification. I just go through the ideas in the book, and design my own program to pass all the cases in the book. The code is in https://github.com/kainwen/paip/tree/master/sym_math2/.
Continue reading2020八月·粉身碎骨浑不怕
题记
千锤万凿出深山，烈火焚烧若等闲。
《石灰吟》于谦
粉骨碎身浑不怕，要留清白在人间。
计划
- 继续阅读PAIP和Logic of Science
- PAIP完成STUDENT
- Logic of Science到第八章
- 整理SIGMOD2020的论文，并选择2篇进行分析阅读
- BinDex: A Two-Layered Index for Fast and Robust Scans
- SuRF: Practical Range Query Filtering with Fast Succinct Tries (趁机研究下Trie和它的算法分析）
- 希望可以把体重控制到78kg以下，并提升摔跤技术
总结(2020.08.31)
- PAIP完成度很满意，跳过了一些无趣的章节，完成了微积分的程序。不过这些技术还都是解释器的技术。
- Logic of Science后期花费的时间不够多，开始重头捋细节
- 论文这两篇都看不下去，没有发现有趣的论文，也没有回顾经典的论文。这方面有所欠缺。
- 体重别提了，我爸来北京后，饭局不断，9月背水一战，必须成功。。。。
- 摔跤技术应该有提升，赢了苏正红和龙哥几下，不排除他们让我的，但是有进步。肘关节伤病实在让人烦。
Eliza in Erlang
Introduction
This is Chapter 5 of the book PAIP. It implemented an early-day algorithm to dialog with a machine. It had the appearance of AI, but actually it was just another interpreter: programmer designs some rules at first, and then eval the user’s input. The main topic is pattern-match programming.
Pattern-match is really my favorite language feature. Erlang and many functional programming languages (Oz, ML, …) support it quite well. C language offers switch statement which can only match constant expression in each clause. Different languages may have different scopes for pattern variables.
GPS in Erlang
The book PAIP
GPS is the acronym for General Problem Solver. The book Norvig, Peter. Paradigms of artificial intelligence programming: case studies in Common LISP. Morgan Kaufmann, 1992 spent the whole Chapter 4 to show how you can program a GPS using common lisp and pointed out several “bugs” that you should pay attention to. This is a great book that I highly recommend it.
Since EOPL, I am convinced that Erlang is a very good language to implement other language or write AI paradigm program. My several projects are:
- finish EOPL’s several language model and exercises: https://github.com/kainwen/eopl
- Jack_{vm, assembler, compiler}:
More details on Common Lisp implementation can be found in the open-source repo: https://github.com/norvig/paip-lisp.
Continue readingElementary sampling theory
The problems are from the Book Jaynes, Edwin T. Probability theory: The logic of science. Cambridge university press, 2003. Chapter 3 of the book is on the topic of Elementary sampling theory. These problems can be easily solved using the symbolic method of analytical combinatorics.
Continue readingVolatile Functions in Query on Broadcast-like Relation in Greenplum
Some background knowledge
Greenplum is MPP database. The SQL query’s planning and executing model is a distributed computing model. Greenplum introduces the concept of locus to model a path’s data distribution.
- Function scan of generate_series is General locus (because the information is complete at every single place, but in Greenplum 5 this function is not treated as General, this may be the historical reason)
- the scan path of a replicated table is segmenteGeneral locus (because data only stored in segments, not master, and the information is complete at every single segment)
- ……
For more details, please refer to one of my talk Greenplum 查询优化器关键技术介绍.
Continue reading2020七月·自强不息 厚德载物
CM (count min) is not MC (min count)
Papers and stories
This month, my most spare time has been spent on the problem Cardinality Estimate. When I finished the analysis of hyperloglog (see the blog A Tour of Understanding Hyperloglog), I just remember a recent paper of SIGMOD2019: Approximate Distinct Counts for Billions of Datasets. Its idea is to combine CountMin and Hyperloglog. Before diving into that paper, I decide to study CountMin. Another paper on cardinality estimate survey is VLDB2017’s Cardinality Estimation: An Experimental Survey mentioned a method called MinCount.
CM sketch was introduced by the paper: An Improved Data Stream Summary: The Count-Min Sketch and its Applications. I do not want to dive into that paper, but just reword the proof here (maybe in a better way).
Continue reading