About关于

Sources来源

The pages are short explanations of standard ideas: confounding, matching, community structure, a message spreading on a graph, overfitting, regularization, and reading text as counts or as places on a map. The books below are where those ideas are taught.这些页是对一些标准想法的短说明：混杂、匹配、人群怎么分成一块一块、一句话在谈话网络上怎么传开、过拟合、正则化，以及把文字看成词数或看成地图上的位置。下面的书是这些想法的出处。

Books书

Causal因果

• Miguel A. Hernán and James M. Robins, Causal Inference: What If. Topics: confounding, exchangeability, and why an association is not yet an effect. This site’s page is the straight-line case in which one common cause is left out.。主题：混杂、可交换性（两组人本来可以互换），以及为什么“一起出现”还不是“造成了”。这一页是一条直线的特例：漏掉了一个共同原因。 Author page for the book作者的书页.
• Judea Pearl, Causality. Topic: causal diagrams. A confounder is drawn as a common cause. This site uses one small diagram, not a general proof of what can be identified.。主题：因果图。混杂画成一个共同原因。这一页只用一张小图，不是一套“什么能够识别出来”的一般证明。 Book site图书网站.
• Joshua D. Angrist and Jörn-Steffen Pischke, Mostly Harmless Econometrics (Princeton University Press). Topic: regression and omitted variables.（普林斯顿大学出版社）。主题：回归，以及表上漏掉的那一列。
• Scott Cunningham, Causal Inference: The Mixtape. Topic: confounding and regression adjustment, taught around the question of whether a slope is an effect.。主题：混杂与回归调整，围绕“斜率是不是效应”来讲。 Book site图书网站.
• Paul R. Rosenbaum and Donald B. Rubin (1983), “The central role of the propensity score in observational studies for causal effects,” Biometrika. Topic: a score that summarizes how the columns on the sheet predict who joined. The matching page has only prior math, so matching on that score and matching on prior math line the students up in the same order.。主题：用一个分数概括表上的列如何预测谁会参加。匹配这一页只有先前数学，所以按这个分数匹配，和按先前数学匹配，学生的顺序是一样的。
• Elizabeth A. Stuart (2010), “Matching methods for causal inference: a review and a look forward,” Statistical Science. Topics: calipers, balance on the columns you measured, and the fact that unmatched people leave the comparison. This page pairs each joiner with one nearby non-joiner and does not reuse a partner. It is not the review’s full list of estimators.。主题：卡尺、量过的那些列有没有配平，以及没配上的人会离开比较。这一页给每个参加者配一个附近的未参加者，同一个人不重复使用。它不是那篇综述里的全部估计方法。

Networks网络

• Stanley Wasserman and Katherine Faust, Social Network Analysis: Methods and Applications (Cambridge University Press). Topics: talks, paths, components, and subgroups. The page reports components and the raw share of talks inside a faculty, not a full assortativity index.（剑桥大学出版社）。主题：谈话、路径、连通分量（被一串谈话连上的一组人）和子群。这一页报告连通分量，以及学院内部谈话的原始占比，不是完整的同配指数。
• Paul W. Holland, Kathryn Blackmond Laskey, and Samuel Leinhardt (1983) on stochastic block models, as cited in the NetworkX stochastic block model documentation. Topic: talks drawn independently from a chance that depends on the pair of groups. The page is handed the faculties. It does not estimate them.。主题：按两组人之间的概率，独立地抽出有没有说过话。学院是交给这一页的。它不去估计学院。
• Mark Newman, Networks: An Introduction (Oxford University Press). Topic: an epidemic on a graph. The page about spare-seat news is a susceptible–infected process on the same two-faculty graph: having heard can pass to someone you talked to, and it does not fade. The news cannot leave the group connected to the student who said it first. A bridge is what lets it reach the other faculty.（牛津大学出版社）。主题：图上的流行病。选修课名额那一页，是同一张两个学院的图上的易感—感染过程：听到了可以传给聊过的人，而且不会淡掉。消息离不开和最先说出来的人连在一起的那一组。桥让它到达另一个学院。

Machine learning机器学习

• Gareth James, Daniela Witten, Trevor Hastie, and Robert Tibshirani, An Introduction to Statistical Learning. Topics: the bias–variance tradeoff, overfitting, and the comparison of training and test error, including polynomial fits. The regularization lab uses the same book’s ridge and lasso idea: a penalty that shrinks slopes, and, for the absolute penalty, can set some slopes to zero.。主题：偏差—方差权衡、过拟合，以及训练误差和测试误差的比较，包括多项式拟合。正则化那一页用同一本书里的岭回归和套索：惩罚把斜率缩短；用绝对值的那种惩罚，还可以把一些斜率设成零。 Book site图书网站. Both labs use this site’s own sine-wave simulation.。两页都用本站自己编的一条正弦波。
• Trevor Hastie, Robert Tibshirani, and Jerome Friedman, The Elements of Statistical Learning. Topic, at a high level: a more flexible curve sits closer to the districts used to draw it, and can change shape more from one draw to the next. The page does not estimate bias and variance as two separate pieces.。主题，往大处说：曲线越灵活，越贴近用来画它的那些区，换一批区时形状也可能变得更多。这一页不把偏差和方差拆成两项来估计。 Book site图书网站.

Text as data文字当数据

• Justin Grimmer, Margaret E. Roberts, and Brandon M. Stewart, Text as Data文字当数据 (Princeton University Press). Topics: choosing how to represent text, the bag of words, the document–term matrix, topic models as mixtures over K word distributions, and giving words a few coordinates. The pages use sentences written here. The topic-count page clusters word counts and says so. The word-map page places coordinates by hand. Neither fits a topic model or trains an embedding, and neither repeats an analysis from the book.（普林斯顿大学出版社）。这本书的英文书名是 Text as Data，讲的是把文字当成可以数的数据。主题包括：怎么表示一段文字、词袋、文档—词矩阵（一行一篇，一列一个词）、主题模型（每篇是 K 个词分布的混合），以及给词几个坐标。这些页用的是这里写的句子。主题个数那一页只是给词数分组，并且如此说明。词图那一页的坐标是手写的。两页都不去拟合主题模型，也不训练嵌入，也不重复书里的语料分析。

Interactive references交互式参考

These works are why the site teaches with pictures you can move. No one else’s visualisation code or research data is included.这些作品是这个网站用“可以动手调的图”来讲方法的原因。这里没有放进别人的可视化代码，也没有放进别人的研究数据。

• Seeing Theory — reference for learning through adjustable visualisations.——通过可调图形学习的参考。
• Bynum and colleagues: An Interactive Introduction to Causal Inference — VISxAI 2022; source repository源代码仓库.
• NetworkX stochastic block model documentation — model definition and the Holland, Laskey & Leinhardt (1983) reference.——模型定义，以及 Holland、Laskey 与 Leinhardt（1983）的文献。
• scikit-learn: Underfitting vs. Overfitting — reference for the polynomial-regression teaching example; this site uses its own sine-wave simulation and validation sample.——多项式回归教学例子的参考；本站用自己的正弦波模拟和验证样本。
• Math Insight applets and MIT Mathlets — references for parameter-driven mathematical exploration.——参数驱动的数学探索的参考。
• Complexity Explorables — reference for interactive complex-systems models.——交互式复杂系统模型的参考。