学习资源与参考 Learning Resources & References
官方文档、公开数据、论文——以及一条从入门到实战的学习路径。 Official docs, public data, papers — and a learning path from zero to practice.
射电天文 Python 工具链 Radio Astronomy Python Stack
astropy
FITS 读写、WCS、单位与坐标——一切天文 Python 的地基。
FITS I/O, WCS, units & coordinates — the foundation of astronomy in Python.
文档 Docs →spectral-cube
谱线数据立方体与矩图的标准库(本案例主角)。
The standard library for spectral cubes & moment maps (star of our case study).
文档 Docs →radio_beam
合成束(beam)处理:Jy/beam ↔ K ↔ 柱密度/质量换算。
Synthesized beam handling: Jy/beam ↔ K ↔ column density / mass.
文档 Docs →CASA 文档与指南
射电干涉阵校准、成像、immoments 的官方教程。
Official tutorials for calibration, imaging, immoments.
指南 Guides →pyuvdata
在 Python 中读写可见度数据(UVFITS / MS / CASA 格式)。
Read/write visibility data (UVFITS / MS / CASA) in Python.
文档 Docs →PINT / PRESTO
脉冲星计时(PINT)与折叠/搜索(PRESTO)。
Pulsar timing (PINT) and folding/searching (PRESTO).
PINT → PRESTO →blimpy / your
Filterbank 数据读取、去色散、FRB/瞬变搜索。
Filterbank I/O, dedispersion, FRB/transient searches.
blimpy → your →HiPS / 可视化
天文图像可视化进阶:WCSAxes(astropy)、aplpy、HiPS 切图。
Advanced visualization: WCSAxes (astropy), aplpy, HiPS tiles.
WCSAxes →AI 编程工具官方入口 AI Coding Tools — Official Entries
国内产品(推荐)与开源工具优先;国外闭源产品仅作了解。
Domestic (recommended) and open-source first; foreign closed-source for reference.
概念延伸阅读 Deeper Reading
DeepSeek-V4 技术报告
百万 token 上下文、高效注意力与 Agent 能力的官方技术报告(2026)。
Official technical report for million-token context & agent capabilities (2026).
arXiv →DeepSeek-V3 / R1 论文
2025 年开源里程碑:V3 的高效 MoE 架构与 R1 的强化学习推理。
The 2025 open milestones: V3's efficient MoE and R1's RL-based reasoning.
V3 arXiv → R1 arXiv →DeepSeek Harness 介绍
官方开源 Agent 运行时:"Everything is a Plugin",含文档、示例与社区。
The official open-source agent runtime: "Everything is a Plugin", with docs, examples and community.
GitHub →Kimi K3 技术报告
全球参数最大开源模型(约 2.8T)的技术报告(2026)。
Technical report of the largest open-weight model, ~2.8T params (2026).
GitHub →ReAct 论文
Agent 循环的奠基工作:Reasoning + Acting(ICLR 2023)。
The foundational paper of agent loops: Reasoning + Acting (ICLR 2023).
arXiv →MCP 协议
Model Context Protocol 官方规范:工具连接的开放标准。
The official MCP specification — the open standard for tool connectivity.
官网 Site →Building Effective Agents(Anthropic)
Anthropic 工程团队对"怎么设计好 Agent"的实战总结。
Anthropic engineering's practical take on designing effective agents.
阅读 Read →A Practical Guide to Building Agents(OpenAI)
OpenAI 的 Agent 构建实用指南。
OpenAI's practical guide to building agents.
PDF →建议学习路径 Suggested Learning Path
- 一周上手:读本站四篇概念页 → 装一个图形化工具(Trae 或 Qoder,免费)→ 让它帮你写一个读取 FITS 头的小脚本 Week 1: read the four concept pages → install a GUI tool (Trae or Qoder, free) → have it write a small FITS-header reader
- 一个月实战:跑通本案例(合成立方体 → 矩图 → CASA 对比)→ 换一份真实公开数据(THOR/HI4PI 切块)重复一遍 Month 1: reproduce this case (synthetic cube → moments → CASA check), then redo it with a real public survey cutout
- 进阶:让 Agent 帮你把管线打包成可复用模块(函数 + pytest + 文档);把流程写成 SKILL.md / SOP 沉淀进课题组;尝试把 CASA 任务封装为 MCP 工具,让 Agent 直接调用 Advanced: package the pipeline as reusable modules (functions + pytest + docs); write the process as a SKILL.md / SOP for your group; try wrapping CASA tasks as MCP tools your agent can call directly