Tag
该论文提出超越准确率评估agentic数据科学系统的方法，通过220项任务的大规模失败模式分析、可复用测试套件和与Kaggle专家方案的定性对比，发现多数agentic解决方案存在方法论缺陷（如数据泄漏），且数据泄漏与预测性能正相关。
Kirk Borne 分享并推荐了教材《Bayes Rules!: An Introduction to Applied Bayesian Modeling》，该书由 Alicia A. Johnson、Miles Q. Ott 和 Mine Dogucu 编写，涵盖共轭先验、MCMC、回归与层级模型，并配有 bayesrule R 包。
Michele Coscia 和合作者利用 LIST（拉丁铭文时空数据库）估算罗马帝国各行省的职业分布，并在校正身份偏见、性别偏见与史学偏见后计算其经济复杂度，成果发表于 npj Complexity。本文是该论文的博客解读，介绍数据、方法与主要发现。
博客文章回顾了 R 语言中管道操作符（%>% / |>）的历史，从 2012 年 StackOverflow 上的提问、Ben Bolker 的首个实现，到 dplyr 中 chain 与 %.% 操作符的引入与发展。
Adobe paper proposes evaluating agents against code that recomputes the correct answer from live data, with an AI grader matching human experts 29% better and using 16% fewer tokens than written descriptions on 53 test cases.
An O'Reilly article outlining 'The Agentic Data Science Playbook' — how AI-native data scientists delegate analysis to AI agents while designing workflows, verification, and feedback loops. Using a modified Elliptic Bitcoin fraud dataset, Claude Opus 5.0 initially reported inflated results (F1 0.87) due to a temporal-split error and a planted leaking feature; with proper guidance, corrected performance dropped to F1 0.70 and 0.21 recall on high-degree nodes, showing why expert oversight still matters.
This paper introduces ORCA, a comprehensive benchmark for evaluating Large Language Models on data science code translation tasks, highlighting challenges and proposing an intent-augmented method to improve performance.
Release of SmolDataEnvs, a collection of 5,000 verifiable RL environment tasks for training small models in code and data science, fully open source.
This article details the implementation of a Parquet file writer in Haskell for the DataHaskell/Dataframe library, enabling efficient data serialization and interoperability with the data science ecosystem.
This article promotes a free live workshop on building AI with Python, offering a roadmap for data scientists to transition to AI/DS System Builder roles with practical demos.
The author shares their past efforts to achieve efficient single-expert loading for high-quality token-level classification at scale, echoing Maxime Rivest's thoughts on the utility of domain-specific expert models.
The Khipu Field Guide is a web-based resource that provides detailed symbolic renderings and interactive analyses of over 600 Incan khipus, aiming to democratize khipu research through computational methods and data science.
TabPFN-3.5 is released, claiming state-of-the-art performance for tabular data beyond IID small data settings, with features like handling grouped and temporal data, uncertainty calibration, and faster inference.
MIT CSAIL shares a beginner's guide to statistics, providing foundational educational content for data science and AI.
The article questions why software engineering jobs remain high compared to data science roles despite AI advancements, referencing US and EU labor statistics and growth projections for ICT professionals.
This paper introduces Calendar-SPCA, a calendar-structured sparse principal component method for interpretable representation learning in multi-periodic electricity consumption profiles, demonstrating high explained variance with increased sparsity and coherence.
A user shares insights from analyzing a decade's worth of lunch data using a state-of-the-art LLM, highlighting interesting findings about expenses and common restaurants.
The paper introduces StatFormBench, a benchmark for evaluating LLMs on statistical problem formulation, and finds that current models have significant limitations in classifying problems and identifying variables.
DS-Lighting is a unified harness toolkit that makes harness design explicit for data-science automation, improving reproducibility, comparability, and reliability in end-to-end workflows.
This paper presents the first application of data science to evaluate the UK Honours system using natural language processing, introducing a novel sentiment analysis algorithm called Minos to assess public opinion on honours recipients.