boxcrete: V2 strength GP — final-form module layout

Introduces the V2 strength GP — a multiplicatively-gated kernel architecture
that structurally enforces the physics constraint f(x, t=0) = 0 with zero
posterior variance — plus the supporting public API and the saved model
artifact.

Module layout (each module focused on one concept):

  * boxcrete/concrete_model.py — SustainableConcreteModel (the
    multi-output joint GWP/slump/strength model) and AppendDerivedFeatures.
  * boxcrete/strength_model.py — public fit/load API for V2:
    fit_strength_gp() and load_pretrained_strength_gp(). Hosts the
    V2 fit factory and ChainedInputTransform composition.
  * boxcrete/kernels.py — kernel primitives: TimeGatedKernel,
    within_group_prior, ard_matern_with_within_group_prior,
    additive_time_kernel, build_strength_kernel_for_aug_dim,
    make_gated_strength_kernel_builder.
  * boxcrete/features.py — features and transforms:
    F5_ALLLOG_FEATURES, GATE_TAU, IDX, FEATURE_BUILDERS,
    append_engineered_features_callable, augmented_bounds, max_scale_Y,
    LearnableLogTimeTransform.
  * boxcrete/likelihoods.py — GatedGaussianLikelihood (heteroscedastic
    Gaussian with h(t)² gating, paired with the gated kernel) and
    PartialFixedNoiseLikelihood.
  * boxcrete/priors.py — WithinGroupShrinkagePrior + the binder /
    aggregate group definitions used for ARD lengthscale shrinkage.
  * boxcrete/slump_model.py — fit_slump_gp() for the slump model.
  * boxcrete/strength_model_legacy.py — V1 helpers retained for
    backward compatibility (get_strength_gp_input_transform).

Replaces the legacy boxcrete/models.py (758 lines) with the focused
modules above.

Persisted artifact:

  * docs/model/strength_model.pt — PyTorch state_dict for the deployed
    V2 strength GP. load_pretrained_strength_gp() reconstructs the
    full SingleTaskGP from this without re-fitting.

Tests:

  * test/test_strength_model_parity.py — guards that the V2 fit
    factory produces byte-equivalent posteriors to the research-side
    catalog at experiments/model_variant_study.py.
  * test/test_pretrained_loader_fidelity.py — guards that the
    state_dict round-trip reproduces the trained model exactly.
  * test/test_strength_curve_monotonicity.py — physical-constraint
    regression for monotonic strength evolution.
  * test/test_physical_constraints.py — physical-constraint regression
    for f(x, t=0) = 0 with zero variance under the gated kernel.
  * test/test_public_api.py — public API surface guard.

Author: SebastianAment

README.md

@@ -16,7 +16,7 @@ This repository contains probabilistic models and data for the
 2) The associated global warming potential (GWP)
 3) Slump prediction using Gaussian Process regression with derived features
 
-as a function of their composition, consisting of cement, fly ash, slag, fine and coarse aggregate, admixtures, and water, to name a few basic ingredients. See `boxcrete/models.py` for implementation details.
+as a function of their composition, consisting of cement, fly ash, slag, fine and coarse aggregate, admixtures, and water, to name a few basic ingredients. See `boxcrete/concrete_model.py` for implementation details.
 
 ### Included Data
 
@@ -51,7 +51,7 @@ pip install -e ".[notebooks]"
 ```python
 import torch
 from boxcrete.utils import load_concrete_strength, get_bounds
-from boxcrete.models import SustainableConcreteModel
+from boxcrete.concrete_model import SustainableConcreteModel
 from boxcrete.plotting import plot_strength_curve
 
 # Load data and fit models
@@ -105,7 +105,7 @@ The models can be used for a variety of tasks, including but not limited to
 
 ## Compressive Strength Model
 
-The `SustainableConcreteModel` in [`boxcrete/models.py`](boxcrete/models.py) includes a strength_model that predicts the evolution of compressive strength as a function of mixture composition. A demo is provided in [`notebooks/strength_curve_prediction_demo.ipynb`](notebooks/strength_curve_prediction_demo.ipynb), which demonstrates how the model can be used to predict the full strength development curve for any user-specified mix. A comprehensive tutorial covering prediction, calibration, Pareto frontiers, and gradient-based experimental design is available in [`notebooks/prediction_and_optimization_tutorial.ipynb`](notebooks/prediction_and_optimization_tutorial.ipynb). The model is based on Gaussian Process (GP) regression and incorporates custom modeling steps to ensure physically consistent strength evolution and calibrated uncertainty.
+The `SustainableConcreteModel` in [`boxcrete/concrete_model.py`](boxcrete/concrete_model.py) includes a strength_model that predicts the evolution of compressive strength as a function of mixture composition. A demo is provided in [`notebooks/strength_curve_prediction_demo.ipynb`](notebooks/strength_curve_prediction_demo.ipynb), which demonstrates how the model can be used to predict the full strength development curve for any user-specified mix. A comprehensive tutorial covering prediction, calibration, Pareto frontiers, and gradient-based experimental design is available in [`notebooks/prediction_and_optimization_tutorial.ipynb`](notebooks/prediction_and_optimization_tutorial.ipynb). The model is based on Gaussian Process (GP) regression and incorporates custom modeling steps to ensure physically consistent strength evolution and calibrated uncertainty.
 
 ### Strength Curve Predictions
 

boxcrete/__init__.py

@@ -4,22 +4,37 @@
 # This source code is licensed under the MIT license found in the
 # LICENSE file in the root directory of this source tree.
 
-from boxcrete.model_utils import FixedFeatureModel, LinearModel
-from boxcrete.models import (
+from boxcrete.concrete_model import SustainableConcreteModel
+from boxcrete.features import (
     AppendDerivedFeatures,
-    fit_slump_gp,
-    fit_strength_gp,
-    get_strength_gp_input_transform,
-    PartialFixedNoiseLikelihood,
-    SustainableConcreteModel,
-    WithinGroupShrinkagePrior,
+    F5_ALLLOG_FEATURES,
+    FEATURE_BUILDERS,
+    GATE_TAU,
+    IDX,
+    LearnableLogTimeTransform,
+    append_engineered_features_callable,
+    augmented_bounds,
+    max_scale_Y,
+)
+from boxcrete.kernels import (
+    TimeGatedKernel,
+    additive_time_kernel,
+    ard_matern_with_within_group_prior,
+    build_strength_kernel_for_aug_dim,
+    make_gated_strength_kernel_builder,
 )
+from boxcrete.likelihoods import GatedGaussianLikelihood, PartialFixedNoiseLikelihood
+from boxcrete.model_utils import FixedFeatureModel, LinearModel
 from boxcrete.plotting import (
     compute_loo_cv,
     plot_calibration,
     plot_feature_importance,
     plot_strength_curve,
 )
+from boxcrete.priors import WithinGroupShrinkagePrior, within_group_prior
+from boxcrete.slump_model import fit_slump_gp
+from boxcrete.strength_model import fit_strength_gp, load_pretrained_strength_gp
+from boxcrete.strength_model_legacy import get_strength_gp_input_transform
 from boxcrete.units import (
     convert_slump,
     convert_strength,
@@ -69,7 +84,13 @@
     "DEFAULT_X_COLUMNS",
     "DEFAULT_Y_COLUMNS",
     "DEFAULT_YSTD_COLUMNS",
+    "F5_ALLLOG_FEATURES",
+    "FEATURE_BUILDERS",
     "FixedFeatureModel",
+    "GATE_TAU",
+    "GatedGaussianLikelihood",
+    "IDX",
+    "LearnableLogTimeTransform",
     "LinearModel",
     "MORTAR_BOUNDS_DICT",
     "MORTAR_CONSTRAINTS",
@@ -80,8 +101,14 @@
     "STRENGTH_DISPLAY_SCALE",
     "SustainableConcreteDataset",
     "SustainableConcreteModel",
+    "TimeGatedKernel",
     "UnitSystem",
     "WithinGroupShrinkagePrior",
+    "additive_time_kernel",
+    "append_engineered_features_callable",
+    "ard_matern_with_within_group_prior",
+    "augmented_bounds",
+    "build_strength_kernel_for_aug_dim",
     "compute_loo_cv",
     "convert_slump",
     "convert_strength",
@@ -93,12 +120,16 @@
     "get_reference_point",
     "get_strength_gp_input_transform",
     "load_concrete_strength",
+    "load_pretrained_strength_gp",
+    "make_gated_strength_kernel_builder",
     "make_linear_coefficients",
+    "max_scale_Y",
     "plot_calibration",
     "plot_feature_importance",
     "plot_strength_curve",
     "predict_pareto",
     "reduce_to_optimization_space",
     "slump_label",
     "strength_label",
+    "within_group_prior",
 ]