omnibias.symbolic
Neural-jet equation discovery and interpretable surrogates. Because every jet
column is an exact omnibias fastpath, the discoverers recover identities
library-free — without being handed an exp, sin, or tanh basis.
import omnibias.symbolic as os_
This is a large package; the highlights below group the public surface by task.
Discovery engines
| Symbol | Role |
|---|---|
NeuralJetDiscoverer | library-free SINDy over jet coordinates x, y, dy, … |
FieldLawDiscoverer | lift the search to many variables / PDE laws |
discover_interpretable_surrogate | AutoML over Taylor/Fourier/hybrid libraries |
discover_pde_operator_law | recover PDE operator coefficients (e.g. diffusivity) |
discover_field_pde_law | recover space–time PDE laws |
discover_activation_identity | recover an activation's defining ODE |
discover_from_noisy_observations | end-to-end from noisy data |
Fields, jets & operators
NeuralFieldND, FieldJet, extract_field_jet, field_gradient,
field_divergence, field_curl, field_laplacian, field_hessian,
field_ito_generator, and dataset builders make_heat_field_split,
make_wave_field_split, make_burgers_field_split,
make_laplace_field_split, make_heat2d_field_split.
Geometry & exterior calculus discovery
laplace_beltrami, christoffel_symbols, covariant_hessian,
riemann_tensor, ricci_tensor, scalar_curvature, pullback_metric_field,
discover_geometric_heat_law; DifferentialForm, exterior_derivative,
hodge_star, codifferential, hodge_laplacian, wedge.
Model selection & uncertainty
- selection:
aic,aicc,bic,mdl,information_criterion,kfold_select,stability_selection,equation_information_criterion. - uncertainty:
bootstrap_coefficients,ridge_coefficient_covariance,certified_coefficient_intervals,attach_uncertainty. - dimensional analysis:
buckingham_pi_groups,dimension_matrix,is_dimensionless,n_dimensionless_groups.
Applied & verified reports
solve_blasius / discover_blasius_identity, latent-ODE discovery
(discover_latent_ode, takens_embedding), and a large family of
verify_* report builders for fluid / regularity / mass-gap proof-preparation
candidates.
The verify_* / candidate-artifact routines build and check finite proof
obligations and diagnostic reports. They are proof-preparation; they do not
close the infinite analytic obligations of the underlying open problems. See
scope boundaries.