Skip to content

Colab notebooks

Five self-contained notebooks that run top to bottom on a free Google Colab GPU runtime (Runtime > Change runtime type > T4 GPU), or on a CPU (slower). The first cell installs PINNeAPPle from PyPI (pinneapple>=0.7.0) and picks the GPU when there is one; the parameter cell (SMOKE, epochs, grid sizes) sets how big each run is.

Notebook What it shows Open
01 First PINN: heat in a plate a Laplace problem solved with pp.solve by a PINN and by FEM, then its trust card: PDE residual, boundary conditions, conservation, maximum principle, reference Open In Colab
02 Airfoil: geometry, surrogate, validation CST airfoil geometry, Cl / Cd / Cm from a trained ensemble with its spread, validated against stored OpenFOAM runs, XFOIL and thin-airfoil theory Open In Colab
03 Equal-budget benchmark three PINN architectures and FEM with the same epochs, points and seeds; error against budget; trust card of the winners; pp.benchmark when installed Open In Colab
04 Lunar rover terramechanics + dashboard NASA VIPER on lunar regolith (Wong-Reece wheels, multibody, deformable terrain), the testbed validation, a PINN of the wheel-soil contact and the animated dashboard as a GIF Open In Colab
05 Verify a model the trust card catching an under-trained PINN, a harmonic field with the wrong boundary, and extrapolation, without a reference solution Open In Colab

Files the notebooks read (the airfoil model and CFD results, the trust-card module on releases older than the one that ships pp.trust_card) come from a local checkout when there is one, otherwise from the repository on GitHub.

Running them headless

python scripts/run_colab_notebooks.py --smoke            # every notebook with tiny sizes (what CI runs)
python scripts/run_colab_notebooks.py 02                 # one notebook at full size
python scripts/run_colab_notebooks.py --check-clean      # committed notebooks must carry no outputs

The committed notebooks have their outputs cleared; --clear strips them after editing.

Coming next

The airfoil and rover notebooks use the model zoo (pp.zoo) when it is installed, and fall back to the repository files or to training on PINNeAPPle 0.7.0. The equal-budget benchmark API (pp.benchmark) and the geometry library (pp.geometries) are being merged: the benchmark notebook already has the cell that switches to pp.benchmark (marked Switch to pp.benchmark) and runs it only when the API is installed.