stick function for the EDA in time series

Which programming language to use for coding interviews

Lựa chọn ngôn ngữ lập trình (Python, Java) cho phỏng vấn coding ảnh hưởng lớn đến hiệu suất, nhưng quan trọng nhất là sự quen thuộc của bạn với ngôn ngữ đó. Trừ trường hợp vị trí đặc thù, không nên học ngôn ngữ mới chỉ để phục vụ phỏng vấn.

Lập trình viên nên đọc bài này để tránh rủi ro mất điểm do không biết chọn ngôn ngữ phù hợp với yêu cầu của các câu hỏi trong cuộc phỏng vấn kỹ thuật.

Google Consent Mode v2

Google Consent Mode là tính năng quan trọng giúp quản lý sự đồng thuận của người dùng đối với cookie và dữ liệu, phiên bản v2 bổ sung các tham số mới (như ad_user_data, ad_personalization) để tối ưu hóa quảng cáo theo quy định. Có thể triển khai thông qua Google Tag Manager, SDK hoặc chỉnh sửa mã trực tiếp.

Lập trình viên nên đọc bài này để hiểu cách tích hợp Google Consent Mode v2 vào dự án của mình để bảo vệ quyền riêng tư người dùng, tối ưu hóa hiệu suất quảng cáo và tránh bị phạt từ các quy định GDPR/CCPA.

How to Ace Data and ML Behavioural Interviews

Behavioural interviews are often the deciding factor for levelling in data science and ML roles, yet most candidates neglect them. The post outlines a preparation strategy built around three pillars: building a 'story vault' of 2–3 deep, reusable project examples; researching the target company's culture and value principles and mapping stories to each; and using the R-STAR-L framework — an extension of the classic STAR method that adds a Repeat step (to confirm understanding and buy thinking time) and a Link Back step (to explicitly tie your answer to the company's stated values). A detailed worked example for a DoorDash data scientist interview illustrates how to naturally weave in a culture-fit signal without sounding scripted.

Schneier on Security

A database of nearly one million passports was leaked online after being stored in an ID verification system used by cannabis dispensaries. The core issue highlighted is that a high-value credential (a passport) was entrusted to a low-value, ancillary authentication system, and when that weaker system was breached, the high-value credentials were exposed. This illustrates the risk of using sensitive identity documents in third-party systems with weaker security postures.

Creating the NVIDIA Nemotron 3 Ultra NVFP4 Checkpoint with NVIDIA Model Optimizer

A deep dive into how NVIDIA quantized the 550B Nemotron 3 Ultra model to NVFP4 using NVIDIA Model Optimizer. The post explains why different layers use different precision formats (NVFP4, FP8, BF16), compares quantization scaling strategies including max scaling, MSE scaling, and the novel four-over-six approach, and shows how bits-per-element sweeps identified 5.03 BPE as the accuracy-performance sweet spot. The result is a 3.2x model size reduction (1,121 GB to 352.3 GB) with near-BF16 accuracy and up to 5.9x higher inference throughput. Practical code examples and YAML recipes for reproducing the pipeline with any Hugging Face model are included, along with a comparison of HuggingFace Transformers vs. Megatron-LM quantization paths (120 min vs. 45 min total).

Flowcharts that belong in the analysis pipeline

The Gmisc R package's flowchart() function has been improved to support reproducible, code-generated research diagrams such as CONSORT charts, cohort derivation flows, data validation pipelines, and longitudinal follow-up diagrams. Unlike drawing tools, these flowcharts live inside the analysis pipeline — counts update automatically, exclusions stay in sync with the analysis script, and figures survive multiple reviewer rounds without manual editing. The post walks through several practical examples with full R code, covering grouped treatment arms, side exclusion branches, dotted return arrows for censored participants, and custom box shapes for data engineering workflows. The update targets Gmisc 3.4.0.

Building Tree Structures From Hierarchical Data In ColdFusion

A practical walkthrough of converting flat hierarchical data (parent-child relationships) into a nested tree structure in ColdFusion using a two-pass O(n) algorithm. The first pass groups elements by parentID into an index, and the second pass injects a children array into each element in-place. The post includes full ColdFusion code with helper UDFs (arrayGroupBy, arraySortOn) and demonstrates both unsorted and sorted tree outputs from a book category dataset.