How jtoleary/spinn works
SPINN learns hidden drift and diffusion physics in stochastic differential equations by propagating uncertainty through known dynamics and fitting neural networks to moment estimates from data. The repository is compact: the sampled implementation is concentrated in core.py, while the README supplies the broader framework description. Data.zip and Examples.zip are visible in the tree, but their contents were not inspected, so their roles and connections are not assumed.
Main entry points: core.py Model training · Data.zip Data archive
Components 9
Moment methods
- Sigma points
core.py - Drift targets
core.py - Diffusion targets
core.py
Hidden physics learning
- Neural networks
core.py - Model training
core.py - Physics reconstruction
core.py
Repository artifacts
- Data archive
Data.zip - Examples archive
Examples.zip
Other components
- Researcher
Connections 8
- Researcher → Data archive: supplies data
- Data archive → Drift targets: provides moments
- Sigma points → Drift targets: supports estimates
core.py - Sigma points → Diffusion targets: supports estimates
core.py - Drift targets → Model training: sets targets
core.py - Diffusion targets → Model training: sets targets
core.py - Neural networks → Model training: trains models
core.py - Model training → Physics reconstruction: enables reconstruction
core.py