Package: MetaHunt 0.1.0

MetaHunt: Privacy-Preserving Meta-Analysis via Low-Rank Basis Hunting

Tools for privacy-preserving meta-analysis of function-valued quantities across heterogeneous studies. Implements the 'MetaHunt' pipeline, including the denoised functional Successive Projection Algorithm (d-fSPA) for basis hunting, constrained weight estimation, Dirichlet regression of weights on study-level covariates, target prediction, and split/cross conformal prediction intervals. Operates on aggregate-level function evaluations, so individual-level data from source studies are not required. Methodology described in Shi, Imai, and Zhang (2026) <doi:10.48550/arXiv.2604.23847>.

Authors:Wenqi Shi [aut, cre], Kosuke Imai [aut], Yi Zhang [aut]

MetaHunt_0.1.0.tar.gz
MetaHunt_0.1.0.zip(r-4.7)MetaHunt_0.1.0.zip(r-4.6)MetaHunt_0.1.0.zip(r-4.5)
MetaHunt_0.1.0.tgz(r-4.6-any)MetaHunt_0.1.0.tgz(r-4.5-any)
MetaHunt_0.1.0.tar.gz(r-4.7-any)MetaHunt_0.1.0.tar.gz(r-4.6-any)
MetaHunt_0.1.0.tgz(r-4.6-emscripten)
manual.pdf |manual.html✨
DESCRIPTION |NEWS
card.svg |card.png
MetaHunt/json (API)

# Install 'MetaHunt' in R:
install.packages('MetaHunt', repos = c('https://wshi18.r-universe.dev', 'https://cloud.r-project.org'))

Bug tracker:https://github.com/wshi18/metahunt/issues

Pkgdown/docs site:https://wshi18.github.io

On CRAN:

Conda:

5.20 score 2 stars 501 downloads 16 exports 11 dependencies

Last updated from:ea2aa310d6. Checks:9 OK. Indexed: yes.

TargetResultTimeFilesSyslog
linux-devel-x86_64OK131
source / vignettesOK190
linux-release-x86_64OK211
macos-release-arm64OK150
macos-oldrel-arm64OK146
windows-develOK100
windows-releaseOK98
windows-oldrelOK82
wasm-releaseOK99

Exports:apply_wrapperbuild_gridconformal_from_fitcoveragecross_conformalcv_error_curvedfspaf_hat_from_modelsfit_weight_modelmetahuntminmax_regretpredict_targetproject_to_simplexreconstruction_error_curveselect_denoising_paramssplit_conformal

Dependencies:digestDirichletRegFormulagenericslatticemaxLikmiscToolsquadprogsandwichwithrzoo

An introduction to MetaHunt
The problem: function-valued meta-analysis | Key assumptions | A1. Low-rank cross-study heterogeneity | A2. Weight model | A3. Exchangeability | A4. Estimation error control | The three-step pipeline | Step 1. Basis hunting via d-fSPA | Step 2. Fitting weight model | Step 3. Target prediction | A worked end-to-end example | Where to next | References
Rendered frommetahunt-intro.Rmdusingknitr::rmarkdown

Last update: 2026-05-07
Started: 2026-04-28

Understanding grid_weights and the L2(mu) norm
What grid_weights does | When the default is fine | When non-uniform grid_weights matter | A worked comparison: uniform vs non-uniform | Practical guidance | Functions that accept grid_weights
Rendered fromgrid-weights.Rmdusingknitr::rmarkdown

Last update: 2026-05-03
Started: 2026-04-29

Choosing K and the denoising parameters
Why this matters | A small standalone simulation | Unsupervised diagnostic: reconstruction error vs K | Supervised diagnostic: cross-validated prediction error vs K | The d-fSPA denoising knobs (N, Delta) | Bypassing denoising | Setting (N, Delta) by hand | Tuning (N, Delta) by CV | Practical recipe | See also
Rendered fromchoosing-k-denoising.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

Conformal prediction with different choices
Why conformal prediction here | A small standalone simulation | The three flavours | Split conformal | Cross conformal | Pre-fit conformal | Pointwise vs scalar bands | Small-m warning on cal_frac | See also
Rendered fromconformal-prediction.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

Get started with MetaHunt
What this is | Inputs in 30 seconds | A minimal end-to-end run | Visualising the result | Where to next
Rendered fromget-started.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

Preparing your data: from fitted models to F_hat
Why this step matters | build_grid() | f_hat_from_models() | Dispatch table | ranger example (not run) | grf::causal_forest example (not run) | Worked lm example | predict_fn for custom S4 / bespoke classes | Multi-dimensional grids (3 covariates) | Sanity checks | Common pitfalls
Rendered fromdata-prep.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

Scalar summaries with wrapper
Why scalar summaries | The wrapper protocol | An ATE example with grf::causal_forest | Three custom wrappers | Plain mean | Restricted positive mean | Endpoint contrast | Conformal coverage with a wrapper | Pointwise vs scalar — quick reference | See also
Rendered fromwrapper-scalar.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

When to prefer minmax_regret
What minmax_regret() does | When to use which | A small standalone simulation | Running minmax_regret() | Visual comparison: minmax_regret vs metahunt | Small-m users | See also
Rendered fromminmax-baseline.Rmdusingknitr::rmarkdown

Last update: 2026-04-29
Started: 2026-04-29

Readme and manuals

Help Manual

Help pageTopics
Reduce predicted functions to scalars via a user-supplied wrapperapply_wrapper
Build a shared evaluation grid from a reference datasetbuild_grid
Extract coefficients from a MetaHunt weight modelcoef.metahunt_weight_model
Split conformal intervals from a pre-fit MetaHunt pipelineconformal_from_fit
Empirical coverage of a conformal prediction-interval objectcoverage
Cross-conformal prediction intervals (pooled K-fold scores)cross_conformal
Cross-validated prediction-error curve for basis-rank selectioncv_error_curve
Denoised functional Successive Projection Algorithm (d-fSPA)dfspa
Build the 'F_hat' matrix from a list of fitted study-level modelsf_hat_from_models
Fit a weight model mapping study-level covariates to simplex weightsfit_weight_model
Fit the full MetaHunt pipelinemetahunt
Minimax-regret aggregator for multisite function-valued estimandsminmax_regret
Plot recovered basis functions from a MetaHunt fitplot.metahunt
Plot a conformal prediction-interval objectplot.metahunt_conformal
Predict the target function for new study-level covariatespredict_target
Predict target functions (or scalar summaries) from a MetaHunt fitpredict.metahunt
Predict simplex weights for new study-level covariatespredict.metahunt_weight_model
Print method for d-fSPA denoising parameter search resultsprint.metahunt_denoising_search
Print a 'summary.metahunt' objectprint.summary.metahunt
Project study-level functions onto the simplex spanned by basis functionsproject_to_simplex
Reconstruction-error curve for basis-rank selectionreconstruction_error_curve
Choose d-fSPA denoising parameters by cross-validationselect_denoising_params
Split conformal prediction intervals for target-function predictionssplit_conformal
Summarise a MetaHunt fitsummary.metahunt
Summarise a conformal prediction-interval objectsummary.metahunt_conformal