How jadesoanes/climate_ai_summer_school works
Project type and purpose
• This is not a monolithic application but a “learning‐materials” repository: a collection of Jupyter notebooks structured by session for the Climate Change AI Summer School.
• Purpose: Organize tutorials and user‐added notes on applying AI/ML to various climate‐related domains (land use, bioacoustics, NLP policy analysis, climate forecasting, building controls, power systems, transport modeling).
File structure
• Top level:
• README.md
• Session_X folders (Session_2 through Session_12, some with lowercase session_11)
• Each Session_X folder contains:
– A “.temp” subfolder (likely intermediate artifacts)
– One or more .ipynb tutorial notebooks
• No src/ or bin/—the code lives in notebooks.
• No Dockerfile or CI configs—assumes execution in Google Colab or local Jupyter.
Key insights from README
• Each session maps to a domain‐specific tutorial
• Notebooks run in Google Colab, use Python ML/AI libraries (PyTorch, scikit‐learn, GEE Python API, CityLearn, BOPTEST, PyTorch Geometric, etc.)
• Data sources and external frameworks vary by session (satellite imagery, audio recordings, text corpora, power‐grid simulation, transport datasets).
Main components
a. Repository & Directory Structure
• GitHub Repo “Climate Change AI Summer School”
• Session_X folders
• .temp artifacts
• README.md
b. Execution Environment
• Google Colab / Jupyter Notebook runtime
• Python runtime container (local/Colab)
c. Notebooks (per session)
• Data Ingestion & Preprocessing steps
• Model Training & Evaluation
• Post‐processing & Visualization
d. External Services & Data Sources
• Google Earth Engine for geospatial data
• Satellite imagery repositories (Session 6)
• Audio datasets (Session 3)
• Text corpora/APIs (Session 4)
• CityLearn simulator & BOPTEST environment (Session 9)
• Power‐grid simulators or data (Session 11)
• Transport/POI and bike‐share datasets (Session 12)
e. ML Frameworks & Libraries
• PyTorch (deep learning)
• scikit‐learn (classical ML)
• PyTorch Geometric (GNNs)
• Hugging Face Transformers (NLP)
• Custom Python scripts (data utils)
Relationships & data flows
• User (Engineer/Student) pulls the GitHub repo → opens a Session_X notebook in Colab/local Jupyter
• Notebook code fetches data from external sources or .temp cache → preprocesses data → feeds into ML frameworks → trains/evaluates models → outputs charts/tables back into notebook UI.
• Intermediate artifacts (model checkpoints, feature caches) can be written to .temp.
Architectural patterns & principles
• Modular organization by session (separation of concerns per domain)
• Notebook‐centric “pipeline” pattern: each notebook is a self‐contained ETL + modeling pipeline
• Reuse of common Python data‐utils across sessions (implied .temp caches)
• Execution environment abstraction (Colab vs. local)—notebooks do not assume specific underlying infra.
Detailed but not overthought
• Don’t try to break every cell into a block—treat each notebook as one component.
• Common elements (Python runtime + ML frameworks) under a “Compute Environment” container.
• Session_X folder as a repeatable pipeline block, optionally annotating domain specifics beside it.