Changes in version 0.1.1 (2026-07-19)                  

Packaging and build

  - R: The R package now compiles the C++ core directly through Makevars
    instead of CMake, with no network access or downloaded dependencies
    at install time. Eigen is provided by RcppEigen; nlohmann/json and
    pcg headers are vendored under inst/include. This makes the package
    installable on CRAN's offline build machines. (fmt and csv-parser,
    used only by the CLI, are no longer part of the R build.)
  - R: Self-registering strategies are compiled directly into the shared
    object, removing the previous whole-archive linking workaround.
  - R: Compiler flags for the direct build mirror the standalone C++
    build — EIGEN_NO_AUTOMATIC_RESIZING on all platforms and
    EIGEN_DONT_VECTORIZE on Windows.
  - Core: Replaced C++20 designated initializers and fixed
    member-initialization order in stats/GroupPartition and
    stats/Simulation so the code compiles warning-free under a strict
    C++17 GCC (-Wall -Wextra -pedantic).
  - Core: A compile-time EIGEN_VERSION_AT_LEAST(3, 4, 0) guard fails the
    build with a clear message if an incompatible Eigen is supplied
    (e.g. via RcppEigen).
  - make r-vendor-deps re-vendors the committed json/pcg headers after a
    version bump in core/Dependencies.cmake.

Documentation

  - R: DESCRIPTION uses Authors@R and cites the projection-pursuit tree
    and forest references with DOIs.
  - R: Examples for the parsnip and plot methods use \donttest with
    requireNamespace() guards instead of \dontrun, so they run under
    --run-donttest when the suggested packages are available.

CRAN

  - Added cran-comments.md. The package passes R CMD check --as-cran
    with no errors or warnings; remaining notes (new submission,
    cosmetic pragmas in the vendored nlohmann/json headers) are
    documented for the reviewer.

                        Changes in version 0.1.0                        

New features

  - Core: Projection-pursuit oblique decision trees and random forests
    for classification, using LDA/PDA optimization.
  - Core: Random uniform variable selection per split for forest
    diversity.
  - Core: Three variable importance measures: permuted (VI1),
    projections (VI2), and weighted projections (VI3).
  - Core: Out-of-bag error and confusion matrix for forests, with
    bootstrap sample indices persisted for recomputation. oob_error is
    NA_real_ (R) / "not available" (CLI) when no observation has any
    out-of-bag tree.
  - Core: Degenerate split detection when projection-pursuit cannot find
    a useful projection.
  - Core: OpenMP multi-threaded forest training.
  - Core: JSON serialization for trained models. Optional metrics fields
    use a uniform null-or-value representation so downstream tooling can
    distinguish "computed but empty" from other shapes without
    special-casing.
  - Core: Cross-platform reproducibility — identical results for the
    same seed on Linux, macOS, and Windows, enforced by golden-file
    tests in CI.
  - R: pptr() and pprf() with formula and matrix interfaces. Returned
    models carry an S3 class vector identifying both model type and mode
    (e.g. c("pprf_classification", "pprf", "ppmodel")).
  - R: predict() returns group labels (type = "class") or vote
    proportions (type = "prob") for classification.
  - R: summary() displays training and OOB confusion matrices.
  - R: Lazy OOB accessors — oob_error(), oob_predictions(),
    oob_samples(), bag_samples(), permuted_importance(),
    weighted_importance() — compute from the training data stored on the
    model on first access and memoize in an environment cache, so
    training is fast and repeated access is free. oob_predictions()
    returns a factor with NA for rows with no OOB tree.
  - R: Permuted variable importance may be negative; this is meaningful
    signal ("within noise") rather than a sentinel, so callers should
    rely on the ranking rather than clipping at zero. Weighted
    projection importance is non-negative by construction.
  - R: Degenerate split warnings when projection-pursuit cannot separate
    groups.
  - R: save_json() and load_json() for model persistence.
  - R: tidymodels/parsnip integration: pp_tree() and pp_rand_forest()
    model specifications.
  - R: ggplot2 visualizations — tree diagrams, variable importance
    plots, projection histograms, and decision boundary plots.
  - R: Bundled classification datasets: crab, crabs, fishcatch, glass,
    image, leukemia, lymphoma, NCI60, olive, parkinson, and wine. (Use
    datasets::iris from base R for iris examples.)
  - CLI: train fits a tree or forest from CSV and saves as JSON.
  - CLI: predict applies a saved model to new data.
  - CLI: evaluate runs train/test evaluation with smart convergence.
  - CLI: summarize displays model configuration, data summary, and
    metrics from a saved model JSON. --data recomputes metrics from
    training data.
  - CLI: benchmark runs multi-scenario performance benchmarks with
    baseline comparison.
  - CLI: serve exposes a saved model over HTTP — GET / returns the model
    summary as JSON (or an HTML dashboard for browsers showing
    configuration, training metrics, and variable importance), GET
    /health is a liveness probe, and POST /predict accepts a feature CSV
    and returns predictions (JSON for API clients, an HTML predictions
    page for browsers, with a Download CSV button and confusion matrix
    when the request CSV includes a response column). Results are cached
    in-memory with shareable ?id=… URLs; the dashboard binds to
    127.0.0.1:8080 by default.

Experimental features

Regression support is included but untested in production workloads. API
surface and defaults may change in future releases.

  - Core: Regression training via a ByCutpoint grouping strategy that
    quantile-slices the continuous response, a MeanResponse leaf, and
    MinSize / MinVariance / CompositeStop (stop::any) stop rules.
  - Core: Regression metrics — MSE, MAE, and R² — computed for training
    and out-of-bag predictions. Forest OOB error is reported as MSE.
  - R: Regression auto-detected when y is numeric (not a factor).
    predict() returns a numeric vector (type = "response").
  - R: Regression strategy wrappers: grouping_by_cutpoint(),
    leaf_mean_response(), stop_min_size(), stop_min_variance(),
    stop_any().
  - R: summary() displays MSE / MAE / R² for regression models.
    oob_predictions() returns a numeric vector with NA_real_ for rows
    with no OOB tree.
  - R: save_json() / load_json() preserve regression mode; parsnip
    pp_tree() / pp_rand_forest() accept mode = "regression".
  - CLI: --mode classification|regression selects the training mode;
    regression reads the last CSV column as the continuous response.
    predict returns numeric predictions and MSE/MAE/R²; evaluate reports
    MSE for regression.
  - R: Bundled regression dataset california_housing (20,433 × 9,
    predict median_house_value). For smaller regression examples use
    datasets::mtcars from base R.