Package index • edaphos

Pilar 1 — Causal AI

Backdoor adjustment, LLM-driven Knowledge Graphs, causal discovery, effect posteriors.

causal_4d_plot(): Plot a time-varying causal effect trajectory
causal_adjustment_set(): Suggest a backdoor-adjustment set from a DAG
causal_augment_dag(): Augment a base DAG with edges from a Knowledge Graph
causal_augment_diff(): Diff between a base DAG and an augmented DAG
causal_cerrado_dag(): DAG tailored to the bundled Cerrado dataset (br_cerrado)
causal_cerrado_real_dag(): Real-data Cerrado pedogenetic DAG
causal_clorpt_dag(): Canonical CLORPT pedogenetic DAG
causal_corpus_deduplicate(): Deduplicate a corpus by DOI or title
causal_corpus_openalex(): Query the OpenAlex corpus
causal_corpus_scielo(): Query the SciELO literature corpus
causal_effect_bootstrap(): Block-bootstrap the backdoor-adjusted direct effect
causal_effect_posterior(): Posterior distribution of a backdoor-adjusted direct effect
causal_effect_time_varying(): Time-varying causal effect beta(t) over a sliding window
causal_effect_trend_test(): Mann-Kendall trend test on a beta(t) trajectory
causal_estimate_effect(): Estimate a causal effect using DAG-guided backdoor adjustment
causal_iv_first_stage(): First-stage regression diagnostics for an IV design
causal_iv_fit_2sls(): Two-stage least squares (2SLS) instrumental variable estimator
causal_iv_from_embeddings(): Fit 2SLS using foundation-model (or proxy) embeddings as instruments
causal_iv_posterior(): Bootstrap posterior for a 2SLS effect as an edaphos_posterior
causal_iv_sargan_test(): Sargan test for instrument over-identification
causal_kg_add_edge(): Add a causal edge to a pedogenetic Knowledge Graph
causal_kg_alignment(): Align Knowledge-Graph node labels to a canonical vocabulary
causal_kg_edges(): Tidy edge list of a pedogenetic Knowledge Graph
causal_kg_load(): Load a Knowledge Graph from disk
causal_kg_new(): Create an empty pedogenetic Knowledge Graph
causal_kg_rank_edges(): Rank Knowledge-Graph edges by evidence strength
causal_kg_rename(): Rename Knowledge-Graph nodes from an alignment mapping
causal_kg_save(): Save a Knowledge Graph to disk
causal_kg_to_dagitty(): Export a Knowledge Graph to a dagitty DAG
causal_kg_to_turtle(): Export a Knowledge Graph to RDF 1.1 Turtle
causal_llm_extract(): Extract causal claims from text via an LLM backend
causal_llm_ingest_abstract(): Ingest an abstract into a pedogenetic Knowledge Graph
causal_llm_ingest_abstract_voted(): Ingest an abstract into a KG via multi-extractor voting
causal_llm_ingest_corpus(): Ingest a corpus of abstracts into a Knowledge Graph (resumable)
causal_llm_vote(): Multi-extractor consensus over LLM-extracted causal claims
causal_ontology_agrovoc(): Query the AGROVOC SPARQL endpoint
causal_ontology_agrovoc_align(): Live AGROVOC alignment for a vector of free-text terms
causal_ontology_agrovoc_align_batch(): Concurrent AGROVOC alignment for a large vocabulary
causal_ontology_cerrado(): Canonical Cerrado pedometric vocabulary
causal_ontology_envo(): Load an ENVO ontology from a local .obo file
causal_sensitivity_from_iv(): Sensitivity analysis of an edaphos_causal_iv fit
causal_sensitivity_from_lm(): Sensitivity analysis of an lm backdoor fit
causal_sensitivity_grid(): Bias-adjustment grid for a Cinelli & Hazlett sensitivity contour
causal_sensitivity_summary(): Cinelli & Hazlett (2020) sensitivity summary for a causal effect
causal_structure_learn(): Structure learning from horizon data -> Knowledge Graph
llm_annotation_export(): Export a reviewed JSONL into the canonical gold-standard format
llm_annotation_launch(): Launch the interactive gold-standard review app
llm_annotation_to_zenodo(): Package a reviewed gold-standard into a Zenodo-ready deposit bundle
llm_annotation_validate(): Validate a gold-standard JSONL file
llm_annotation_vocabulary(): Canonical pedometric vocabulary for LLM-KG claims
llm_benchmark_cost(): Estimate per-1 000-claim extraction cost
llm_benchmark_kappa(): Pairwise Cohen's kappa between backends on edge presence
llm_benchmark_match(): Match extracted LLM claims against a gold-standard set
llm_benchmark_metrics(): Compute precision / recall / F1 from a match table
llm_benchmark_simulate(): Simulate backend extractions from a gold-standard set
llm_preannotate(): Pre-annotate a corpus with an LLM to produce draft claims
summary(<edaphos_causal_kg>): One-line summary of a Knowledge Graph

Pilar 2 — Physics-Informed ML

Pedogenetic ODE, Bayesian posterior, Neural ODE + ensemble.

piml_bayes_posterior(): Posterior predictive distribution from a Bayesian Pillar 2 fit
piml_hierarchical_fit(): Hierarchical Neural ODE over multiple pedons (Pillar 2 × Pillar 5)
piml_hierarchical_predict(): Predict depth profiles for new locations from covariates
piml_neural_ode_fit(): Fit a Neural ODE depth profile (Pillar 2, deep variant)
piml_neural_ode_fit_ensemble(): Train a deep ensemble of Neural ODEs for uncertainty quantification
piml_neural_ode_posterior(): Posterior predictive distribution from a Pillar 2 deep ensemble
piml_neural_ode_predict(): Predict a depth profile from a fitted Neural ODE
piml_profile_fit(): Fit a Physics-Informed depth-profile model (Pillar 2)
piml_profile_fit_bayesian(): Bayesian posterior for the Pillar 2 pedogenetic ODE
piml_profile_fit_group(): Fit the Pillar 2 profile model to a group of pedons independently
piml_profile_predict(): Forward-integrate a Physics-Informed depth profile
piml_qkrr_fit(): Fit a Physics-Informed Quantum Kernel Ridge Regression
piml_quantum_kernel(): Physics-informed quantum kernel via ODE-residual fusion
predict(<edaphos_piml_bayes>): Posterior predictive distribution of a Bayesian Pillar 2 fit
predict(<edaphos_piml_neural_ode_ensemble>): Predictive posterior from a Neural-ODE deep ensemble

Pilar 3 — 4D Pedometry

ConvLSTM, rollout, mass-balance physics loss, stochastic EnKF.

temporal_convlstm_cell(): Build a standalone ConvLSTM cell (Pillar 3 primitive)
temporal_convlstm_ensemble_fit(): K-seed deep ensemble of stacked ConvLSTMs
temporal_convlstm_ensemble_rollout(): Roll every ensemble member forward and stack the forecasts
temporal_convlstm_fit(): Fit a stacked ConvLSTM on a 4D covariate cube
temporal_convlstm_mcdropout_predict(): MC-dropout predictive draws from a ConvLSTM fit
temporal_convlstm_predict(): Predict with a fitted stacked ConvLSTM
temporal_convlstm_rollout(): Multi-step rollout forecast
temporal_cube_to_tensor(): Assemble a 4D input tensor-ready array from a synthetic cube
temporal_kalman_update(): Ensemble Kalman update of a Pillar 3 forecast by new point observations
temporal_piml_loss(): Physics-informed ConvLSTM mass-balance loss (Pilar 2 x Pilar 3)
temporal_synth_soc_cube(): Generate a synthetic 4D soil-dynamics cube (Pillar 3 helper)

Pilar 4 — Foundation Models

SimCLR + MoCo v2 pretraining, Zenodo-hosted weights, raster extraction.

foundation_build_cerrado_stack(): Build the Cerrado raster stack for IV / quantum-foundation benchmarks
foundation_embed_at_coords(): Extract foundation-model embeddings at a set of query coordinates
foundation_finetune_ensemble(): Deep-ensemble fine-tune of a Pillar 4 foundation encoder
foundation_fit_classifier(): Fine-tune or linearly probe a Pillar 4 encoder for classification
foundation_fit_regressor(): Fine-tune or linearly probe a Pillar 4 encoder for regression
foundation_mcdropout_predict(): MC-dropout predictive draws from a fine-tuned Pillar 4 fit
foundation_moco_embed(): Extract backbone embeddings from a fitted MoCo v2 encoder
foundation_moco_embed_raster(): Apply a fitted MoCo v2 encoder over a full raster mosaic
foundation_moco_pretrain(): Pillar 4 – MoCo v2 pre-training on raster covariate patches
foundation_moco_pretrain_tiles(): Dataset-backed MoCo v2 pre-training for planetary-scale corpora
foundation_simclr_embed(): Extract embeddings from a pretrained SimCLR encoder
foundation_simclr_pretrain(): SimCLR pre-training on raster covariate patches (Pillar 4 scaffold)
foundation_tile_align(): Align multiple raster sources onto a common analysis grid
foundation_tile_dataset(): Build a lazy patch dataset over a terra::SpatRaster
foundation_tile_source_era5(): ERA5 source stub (needs Copernicus CDS key)
foundation_tile_source_modis(): MODIS source stub (needs NASA EarthData credentials)
foundation_tile_source_soilgrids(): Fetch a SoilGrids 250 m stack over an AoI
foundation_tile_source_srtm(): Fetch an SRTM elevation raster over an AoI
foundation_tile_source_worldclim(): Fetch a WorldClim 2.1 climate stack over an AoI
foundation_weights_download(): Download a pretrained Pillar 4 encoder from Zenodo
foundation_weights_list(): Catalogue of pretrained Pillar 4 encoders
foundation_weights_load(): Load a pretrained encoder into an edaphos_foundation_moco wrapper
predict(<edaphos_foundation_classifier>): Predict class probabilities / labels from a fine-tuned classifier
predict(<edaphos_foundation_regressor>): Predict numeric targets from a fine-tuned regressor
predict(<edaphos_foundation_ensemble>): Predict with an edaphos_foundation_ensemble

Pilar 5 — Active Learning

Hybrid policy, BatchBALD, physics gate, posterior calibration.

al_fit(): Fit a Quantile Regression Forest for Active Learning
al_history(): Extract the learning curve from a fitted Active-Learning model
al_initial_design(): Initial sampling design via Conditioned Latin Hypercube (cLHS)
al_loop(): Closed-loop Autonomous Active Learning for soil mapping
al_physics_gate_piml(): Build a physics gate from a PIML profile fit
al_physics_gate_piml_hierarchical(): Per-location physics gate backed by a hierarchical PIML fit
al_query(): Query the most informative unlabeled candidates
al_query_batchbald(): BatchBALD information-theoretic batch acquisition
al_query_bhs(): Bayesian Hierarchical Active Learning (Pilar 7 x Pilar 5)
al_query_causal(): Causal Active Learning: query the next sample(s) that most reduce the uncertainty of a targeted causal effect
al_query_diffusion(): Diffusion-posterior-driven AL (Pilar 9 x Pilar 5)
al_query_neural_operator(): Causal-driven AL via Neural Operator disagreement (Pilar 8 x Pilar 5)
al_query_temporal(): Temporal Active Learning: rank candidate cells by their Kalman gain norm after the latest EnKF assimilation
al_update(): Append newly labeled samples and refit the model
active_learning_posterior(): Posterior predictive distribution of an edaphos Active-Learning fit

Pilar 6 — Quantum ML

ZZFeatureMap, Quantum KRR, Qiskit VQE, organo-mineral Hamiltonians.

quantum_feature_map(): Quantum feature map (Pillar 6)
quantum_hamiltonian(): Build a quantum Hamiltonian from Pauli-string coefficients
quantum_hamiltonian_from_pyscf(): Build a quantum Hamiltonian from a molecular geometry
quantum_hamiltonian_h2(): Molecular H2 in the Bravyi-Kitaev-tapered 2-qubit basis
quantum_hamiltonian_ising_1d(): Transverse-field Ising Hamiltonian on an n-qubit chain
quantum_hamiltonian_organo_mineral(): Toy organo-mineral Hamiltonian (4-qubit Fe + ligand coupling)
quantum_hamiltonian_organo_mineral_nature(): Organo-mineral Hamiltonians derived from ab initio molecular models
quantum_ibmq_available(): Check whether an IBM Quantum backend is reachable
quantum_ibmq_backends(): List IBM Quantum backends available to the current account
quantum_ibmq_least_busy(): Pick the least-busy operational IBM Quantum backend
quantum_ibmq_submit(): Submit a single expectation-value PUB to IBM Quantum hardware
quantum_kernel(): Quantum kernel Gram matrix via ZZFeatureMap overlap
quantum_krr_fit(): Fit a Quantum Kernel Ridge Regression (Pillar 6)
quantum_krr_posterior(): GP-equivalent posterior for a Quantum Kernel Ridge Regression fit
quantum_nature_available(): Check whether the qiskit-nature + PySCF stack is available
quantum_nature_total_energy(): Total molecular energy from an active-space VQE result
quantum_scale(): Rescale a covariate matrix into [lower, upper] column-wise
quantum_vqe_exact(): Exact ground-state energy via classical diagonalisation
quantum_vqe_fit(): Variational Quantum Eigensolver (Pillar 6 main entry point)

Pilar 7 — Bayesian Hierarchical Spatial (v2.3.0+)

Gaussian process + Gibbs sampler (R fast-path / RcppArmadillo / spBayes).

bhs_fit(): Fit a Bayesian hierarchical spatial linear model (Pilar 7)
predict(<edaphos_bhs>): Predict at new sites from a fitted Bayesian hierarchical spatial model

Pilar 8 — Neural Operators (v2.4.0+)

DeepONet + FNO over depth function space.

no_deeponet_fit(): Fit a DeepONet for depth-profile operators
no_fno_fit(): Fit a 1-D Fourier Neural Operator for depth-profile operators

Pilar 9 — Diffusion Models (v2.5.0+)

DDPM with cosine schedule + ancestral sampling.

dm_cosine_schedule(): Build a DDPM noise schedule
dm_fit(): Train a tiny DDPM on a collection of soil-map patches
dm_sample(): Sample new soil-map patches from a trained DDPM

Pilar 10 — Graph Attention Networks (v2.6.0+)

k-NN co-location graphs + multi-head attention; v3.6.0 sparse.

gnn_build_graph(): Build a k-NN co-location graph from a profile frame
gnn_causal_discovery(): Graph-based causal discovery (Pilar 10 x Pilar 1)
gnn_embed(): Node-level embeddings from a fitted GAT
gnn_fit(): Fit a Graph Attention Network on a WoSIS-style co-location graph

Cross-pillar bridges (v1.9.x + v2.0.0 + v3.0.0)

Foundation embeddings as causal instruments; quantum kernels over foundation embeddings; six v3.0.0 bridges (P7/8/9 x P5 + P10 x P1 + P2 x P3 + P6 x P10).

qf_embed_reduce(): Reduce foundation-model embeddings to PCs and scale to quantum range
qf_kernel_compare(): Compare quantum, RBF, and linear kernels on the same feature set
qf_krr_benchmark(): Benchmark quantum-foundation KRR against classical baselines
qf_krr_fit(): Quantum Kernel Ridge Regression on foundation embeddings
qf_krr_on_gat_embeddings(): Quantum KRR over GAT node embeddings (Pilar 6 x Pilar 10)
al_query_batchbald(): BatchBALD information-theoretic batch acquisition
al_query_bhs(): Bayesian Hierarchical Active Learning (Pilar 7 x Pilar 5)
al_query_causal(): Causal Active Learning: query the next sample(s) that most reduce the uncertainty of a targeted causal effect
al_query_diffusion(): Diffusion-posterior-driven AL (Pilar 9 x Pilar 5)
al_query_neural_operator(): Causal-driven AL via Neural Operator disagreement (Pilar 8 x Pilar 5)
al_query_temporal(): Temporal Active Learning: rank candidate cells by their Kalman gain norm after the latest EnKF assimilation
gnn_causal_discovery(): Graph-based causal discovery (Pilar 10 x Pilar 1)
temporal_piml_loss(): Physics-informed ConvLSTM mass-balance loss (Pilar 2 x Pilar 3)

Unified uncertainty API (v1.6.0)

Common edaphos_posterior class + single calibration diagnostic.

uncertainty_calibrate(): Calibration diagnostics for an edaphos_posterior
uncertainty_plot_reliability(): Reliability diagram from a calibration result
edaphos_posterior(): Unified posterior object for the edaphos pillars
as_edaphos_posterior(): Coerce a native pillar object to edaphos_posterior
autoplot.edaphos_posterior(): Default ggplot for an edaphos_posterior
edaphos_bias(): Mean bias (observed minus predicted)
edaphos_ece(): Expected calibration error (ECE) for a regression reliability diagram
edaphos_interval_score(): Interval score (Gneiting and Raftery 2007)
edaphos_mae(): Mean absolute error
edaphos_metrics_summary(): Summarise a pointwise + interval prediction against observations
edaphos_picp(): Prediction-interval coverage probability (PICP)
edaphos_r2(): Coefficient of determination (Nash-Sutcliffe efficiency)
edaphos_rmse(): Root-mean-square error
edaphos_zenodo_release(): Build a Zenodo-ready release bundle for the edaphos package

Benchmark wrappers (v3.1.0)

Plug-and-play wrappers for the 6-pilar WoSIS Cerrado benchmark.

benchmark_fit_p10_gat(): Benchmark wrapper: Pilar 10 – GAT seed-ensemble on k-NN graph
benchmark_fit_p1_causal(): Benchmark wrapper: Pilar 1 – DAG-adjusted OLS + parametric bootstrap
benchmark_fit_p6_quantum(): Benchmark wrapper: Pilar 6 – bootstrap-ensembled quantum KRR

LLM-KG production pipeline (v3.10.0)

Resumable 10k+ abstract orchestrator + Ollama pre-flight.

llm_kg_ollama_check(): Check whether a local Ollama server is reachable
llm_kg_pipeline_run(): Run the Pilar 1 LLM-KG pipeline on a (potentially large) corpus

Bundled datasets

br_cerrado: Synthetic Cerrado soil sample for edaphos vignettes
br_amazon: Synthetic Amazon-rainforest soil sample (NW Brazil)
br_pantanal: Synthetic Pantanal-wetland soil sample (MS, Brazil)

Package roadmap

edaphos-roadmap: Roadmap and status of the six pillars of edaphos