跳过导航至主内容
Skip to article

研究笔记 · Research Note

Python 科学计算

Zhenyu He · Jobs Stroustrup 阅读约 1 分钟

Python 科学计算 (Python Scientific Computing)

熟练度

精通

描述

以 Python 为主要工具进行科学计算和数据分析。具体能力包括:

  • 数值模拟数据的处理与可视化(netCDF、HDF5 格式)
  • 数据管线构建(QA 过滤、空间重映射、时间聚合)
  • 空间数据加速处理(KD-tree 索引、网格化聚合)
  • 统计分析与敏感性实验设计
  • MCMC 等反演方法实现

🌟 研究级 signature capabilities

Aggregate scale signature

  • ~175k+ lines total aggregated 跨 7 年 research (PKU Walker 27,882 MATLAB+NCL lines + PKU Aerosol ~6,400 MATLAB lines + Caltech DSCOVR ~3,000+ lines + PhD 115,363 lines / 62 ipynb / 3,793 code cells — Python 主 + 跨语言 auxiliary)
  • 62 ipynb notebooks in PhD research notebook (UC Berkeley Romps Lab aggregate)

Multi-language fluency (context-switch discipline)

  • Caltech DSCOVR 11-month 项目内 5 种语言 fluency: IDL (Yuk Yung lab OOP MCMC class) + MATLAB (Hansen csvd) + Python (Kawahara CUDA + emcee + healpy) + Mathematica (symbolic 推导) + R (Metropolis-Hastings)
  • PhD 阶段 Fortran 77/90 (NCAR CAM3 + DAM LES source mods, 见 数值模式源代码修改)

Filename-as-version-control discipline

  • Jupyter notebook 用 filename 编码 methodology version (e.g., enforceEpsilon_z_iteration_v3.ipynb — filename encoding methodology iteration)
  • Aerosol 代码目录: hzy_experi1_package_collaborator_revised_now_no_error_above_height_4 (filename 编码 collaboration history)
  • 见 Inconvenient Evidence Retention Discipline (private)

Research-grade Python libraries

  • emcee (Foreman-Mackey 2013) MCMC
  • healpy HEALPix 球面网格
  • scipy.linalg Cholesky (assume_a="pos") + slogdet + SVD
  • KDTree spatial indexing (TROPOMI 大规模 coordinate match)
  • pandas / xarray / netCDF4 / h5py
  • matplotlib / cartopy geographic plotting

在哪些经历中用到