Changes in version 0.1.0 (2026-10-08)                  

Major Features

Analysis Integration

  - Added comprehensive support for analytical methods with automatic
    metadata integration
  - PCA: compute_prcomp() performs PCA, adds scores to metadata and
    stores the full prcomp object (use store = FALSE to only add the
    scores)
  - Clustering:
      - compute_hclust(): Hierarchical clustering with dendrogram
        storage
      - compute_kmeans(): K-means clustering
  - Other embeddings: compute_mds(), compute_tsne() (Rtsne) and
    compute_umap() (umap)
  - Analysis results are added as columns to metadata with clear naming:
      - PCA: {name}_PC1, {name}_PC2, etc.
      - Clustering: {name}_cluster

Analysis Management

  - get_analysis(): Retrieve stored analysis objects (prcomp, hclust,
    kmeans)
  - list_analyses(): List all stored analyses
  - remove_analysis(): Remove specific or all stored analyses
  - check_analyses(): Check validity of stored analyses

Analysis Invalidation

  - Stored analysis objects are automatically removed when data is
    modified: filter(), slice(), arrange(), joins,
    summarize()/count()/tally(), t() and all matrix transformations
  - Metadata columns (PC scores, cluster assignments) are preserved even
    when analysis objects are removed
  - Clear warnings indicate which analyses were removed during data
    modification

Core Features (from initial development)

Data Structure

  - tidymatrix(): Create tidymatrix objects combining matrix data with
    row and column metadata
  - activate(): Switch context between rows, columns, or matrix
  - Automatic validation of matrix-metadata alignment

dplyr Integration

  - filter(): Filter rows or columns with corresponding matrix
    subsetting
  - select(): Select metadata columns
  - mutate(): Add or modify metadata columns
  - arrange(): Reorder rows or columns
  - slice(), slice_head(), slice_tail(), slice_sample(): Subset by
    position
  - rename(): Rename metadata columns
  - pull(): Extract metadata columns as vectors
  - relocate(): Reorder metadata columns

Grouping and Aggregation

  - group_by(): Group by metadata variables (creates grouped_tidymatrix)
  - summarize(): Aggregate grouped data with matrix aggregation
      - Default mean() for numeric matrices
      - Required .matrix_fn parameter for non-numeric matrices
      - Type-aware error messages with helpful suggestions
  - count(): Count observations by group with matrix aggregation
  - tally(): Count within existing groups
  - ungroup(): Remove grouping

Design Principles

1.  Tidy principle: Same data type in, same data type out
2.  Explicit naming: Clear, prefixed column names prevent conflicts
3.  Smart defaults: Sensible defaults with explicit requirements where
    needed
4.  Helpful errors: Type-aware error messages guide users
5.  Analysis objects: Store full R objects (prcomp, hclust) for later
    use
6.  Metadata persistence: Metadata columns persist even when analysis
    objects are invalidated

Row- and column-wise statistics

  - compute_across() applies any function to every row or column, with
    access to the metadata of the other dimension
  - Wrappers: compute_ttest(), compute_wilcox(), compute_anova(),
    compute_kruskal(), compute_correlation(), compute_lm(),
    compute_lm_simple()
  - In two-group tests, groups that are not given explicitly are
    inferred in a deterministic order (factor levels, otherwise sorted
    values) and a message reports the comparison. Missing group labels
    are ignored.

Matrix operations

  - transform_matrix(), scale(), center(), clip_values(),
    log_transform(), t() and add_stats()
  - Extra arguments to transform_matrix() are evaluated with the
    metadata of the other dimension as data mask, so values can be
    transformed depending on their metadata (e.g. reverse-scoring
    questionnaire items)

Joins

  - All six dplyr joins on rows or columns. The matrix keeps its storage
    type and row/column names; rows or columns that only exist in the
    joined table are filled with NA

Data and documentation

  - New example datasets: a simulated Big Five personality survey
    (big5_responses, big5_respondents, big5_items) and a country table
    (big5_countries)
  - Vignettes rewritten around the survey data: a getting-started tour
    plus vignettes on dplyr verbs, joins, matrix operations, PCA and
    clustering, other embeddings, row-wise statistics, and plotting and
    exporting