nxs-computational-tutorial-2025

Agenda

Set up

1. Log into analysis.sns.gov
2. Run script start_jupyter.sh provided by Jean B. in /EXAMPLES/NXS2025/

Intro presentation (powerpoint) 30-45 mins.

• What is scientific software (MG)
• Version control: Git and GitHub (MG)
• Environment management (micromamba, pixi and pip) (JB):
• Running python options: scripts, python interpreter, IDE, jupyter (JB)
• Intro to file systems at ORNL. Where are my neutron data stored? Oncat (AS)
• AI (YZ)

Tutorial

Malcolm tutorial

• create a git repo

Jean tutorial

• Open notebook, Explanation of notebook (shift enter, shift enter...)
• Cell: imports:
• Exercise 1: Import data from ascii to numpy array. Do this multiple ways. Mention pandas.
• Exercise 2: Plot with matplotlib. Make it interactive. Show errors?
• Exercise 3: Extend script to for loop over multiple files
• Exercise 4: Create widget to do Exercise 3.

BREAK (AS)

Zach Tutorial 2

• Exercise 4 (SciPy): Set up fit to a peak: initial conditions, define to fit, define residual, define fit range, interpret errors (variance-covariance matrix)
• Exercise 5: Use LMFIT for same process.
• Advanced Exercise 1: Event data: Inspect nxs file with HDFView,
• Load neutron data and log metadata from nxs file with h5py.
• Advanced Ex 2: histogram events (with log binning)
• Super Advanced Ex 3: Re-use fitting script, fit peaks, plot position versus experimental log.

AI Tutorial 3

• Brief intro to LLM and relevant techniques (e.g., RAG), from a research user point of view.
• Tools for research
  • notebookLM
  • NapkinAI
  • Perplexity
  • image generation models/tools.
• Tools for programming
  • Web-based chat services (chat.com, gemini, claude, etc.)
  • IDEs, such as Cursor, GitHub Copilot in VSCode, Zed, VSCodium, ... you name it...
  • CLI tools, Codex by OpenAI, Claude Code by Anthropic, Gemini CLI by Google, etc.
• Integrations
  • Unified platform for LLMs, e.g., OpenRouter
  • Self-hosted options
    • Personal service with access to LLMs through APIs.
    • Pay by API call, counting by input/output tokens
    • LobeHub, Dify, etc.
    • Dify as an example for demo
      • LLMs access
      • Tools integration, search via Google & Perplexity, Slack, DALL-E, etc.
      • Personal knowledge base for RAG
  • Implementation in workflow platforms
    • n8n as an example
      • Chat to GPT models in Slack
      • AI summary and auto posting

References

Environment management

• micromamba
• pixi
• pip

Python editors

• jupyter
• visual code
• pycharm

Python librairies

• matplotlib
• numpy
• widgets

__