Tutorial Notebooks¶
These notebooks are executable tutorials checked into the repository. The primary
links open them in Google Colab, where each notebook installs seahorse-stpp from
PyPI. Notebooks 01 and 02 generate a small demo dataset inside the notebook; the
case study loads a real dataset from the Hub.
Available Notebooks¶
| Notebook | What it covers | Runtime notes |
|---|---|---|
| 01 Run One Model With The Python API | Generate demo JSONL data, fit AutoSTPP and PoissonGMM, evaluate, call predict_next, and plot sampled next locations. |
CPU-only, no Hugging Face dependency. |
| 02 Benchmark Models With The CLI | Generate demo JSONL data, run python -m seahorse bench with poisson_gmm, hawkes_gmm, auto_stpp, and deep_stpp, then inspect benchmark tables. |
CPU-only, uses one seed and one epoch. |
| 03 Case Study: NYC Citibike | Load the real Citibike dataset from the Hub, explore it on a map, fit PoissonGMM and DeepSTPP, compare them, and visualize predictions. |
CPU-only; downloads Citibike from Hugging Face. |
Running Locally¶
From the repository root:
python -m venv .venv
source .venv/bin/activate
python -m pip install seahorse-stpp notebook
jupyter notebook docs/notebooks/
Notebooks 01 and 02 create their own demo data under runs/tutorials/; the case
study downloads Citibike from the Hub.
Execution Notes¶
The notebooks are designed for a fresh Colab runtime and do not require a GPU. Notebooks 01 and 02 need no external data; the case study downloads the Citibike dataset from the Hub. They also run locally.