Skip to content

API Reference

The public surface is split across the pages below, grouped by concern. Every symbol below is also re-exported at the top level — from jaxhybridmodels import MLPPredictor works exactly like from jaxhybridmodels.predictors import MLPPredictor.

Data: Experiments, Channels, Datasets

ChannelObs, Experiment, make_experiment, BucketPayload, Dataset, make_dataset, make_bootstrap_dataset, split_dataset, describe_buckets

Predictors: Trainable Components

Predictor, BoundScaler, BoundedPredictor, MLPPredictor, KANPredictor, NeuralNPolynomial, reinitialize_with_key, reinitialize_pytree_with_key

Penalties: Gradient-Safe Bound Handling

PenaltyPointSource, attach_penalty_state, bound_penalty, box_grid, box_violation, clip_ste, data_penalty_points, length_mask_keep, penalty_integral, penalty_vector_field, select_penalty_points, soft_inverse, soft_logit, softclip, strip_penalty_state, trajectory_saturation_penalty, validate_penalty_points

Transforms: Squash Shapes and Axis Warps

BoundTransform, BOUND_TRANSFORMS, register_bound_transform, Warp, WARPS, register_warp

Solver: ODE Integration

SolverConfig, SOLVER_REGISTRY, register_solver, ADJOINT_REGISTRY, register_adjoint

Training: Optax & Evosax Loops

OptaxTrainingConfig, train_with_optax, train_seed_ensemble, train_bootstrap_ensemble, EvosaxTrainingConfig, train_with_evosax, register_algorithm

Training Kernels: Build Your Own Loop

apply_length_mask, predict_bucket_obs, build_bucket_step, build_score_bucket, build_penalty_step, build_apply_update

Losses: Masked & Balanced Objectives

masked_mse, masked_mle, bal_mse, bal_mle, LOSS_REGISTRY, resolve_loss_fn

Trainable Masks: Freezing Leaves

default_trainable, trainable_mask, freeze_paths, freeze_modules_of_type, freeze_where, frozen_default_mask, count_trainable_params

Prediction: Forward Simulation

predict_bucket, predict_dataset, predict_dense, ensemble_predictions, evaluate_predictor

Metrics: Per-Channel Evaluation

ChannelMetrics, compute_metrics, print_metrics

Profiles: Time-Varying Inputs

constant_profile, step_profile, ramp_profile, piecewise_linear_profile

Schedules: Epoch-Scaled Annealing

annealing_schedule

Serialise: Save & Load Runs

save_predictors, load_predictors, save_run, load_run

UI: Training Dashboards

TrainingUI, EvosaxUI, SilentUI, RichTrainingUI, RichEvosaxUI

RNG: Named-Fold Keys

fold

Released under the BSD-3-Clause License.