Skip to contents

Changelog

Source: NEWS.md

mathmodels 0.0.14

Data cleaning toolkit (new)

  • Diagnostics: na_summary() (per-column missing counts), na_mcar_test() (Little’s MCAR test, Little 1988 / SAS-style pattern-mean statistic with EM-estimated mean and covariance, following naniar::mcar_test()), plot_na_heatmap() and plot_na_bar() for missingness visuals.
  • Imputation: impute() for simple fills (group mean / median / mode via .by slices, row-order linear and spline interpolation reusing interp_linear() / interp_spline(), and constants) and impute_model(), a unified model-based framework with k-nearest-neighbors (hand-written), linear regression, and decision trees (rpart). Both return the data frame of the same shape and record every changed cell in attr(df, "changes") (model fits are also summarized in attr(df, "details")).
  • Outliers: outlier() detects outliers with IQR / z-score / Hampel-MAD bounds (per .by slice). By default the data is untouched and the detection table is returned; to = "na" sets detected values to NA (ready for impute()), to = "cap" caps them at the fence values.

Pivot table (new)

  • pivot_table(): aggregates values by the cross of index and columns (.by-style tidy evaluation) and spreads them into a wide tibble; naming is built with names_glue so the value-column prefix is always present (e.g. sales_2023, v_c1), controlled by names_sep.
  • Dependencies (new): rpart (ships with R).

Multiple imputation toolkit (new)

  • impute_multiple(): FCS (fully conditional specification, chained-equations) multiple imputation — pmm/norm methods for numeric columns and logreg/polyreg for categorical ones; validated against mice in cross-validation tests.
  • rubin_pool(): pools scalar estimates across imputations with Rubin’s rules, including the Barnard-Rubin small-sample degrees-of-freedom adjustment.
  • pool_fit(): pools lm/glm fits (coefficients, SEs, t/p-values, pooled residual df) with automatic Barnard-Rubin edf.

Statistical inference toolkit (new)

  • Hypothesis tests (stat_infer.R): tidy data frames in, tidy tibbles out (one row per test, stat_infer class with a labelled print).
    • stat_t_test() / stat_wilcox_test(): one-sample, two-sample (Welch or pooled), and paired tests with Cohen’s d / rank-biserial effects; Wilcoxon signed-rank and Mann-Whitney U as the nonparametric counterparts.
    • stat_anova() (multi-factor with interactions, partial eta-squared), stat_kruskal() (epsilon-squared), stat_ancova() (with an optional parallel-slopes check), stat_cor() (Pearson/Spearman/Kendall with p-adjustment), stat_chisq() (Cramer’s V, low-expected-frequency warning, Fisher exact option), stat_normality() (Shapiro-Wilk plus skewness/kurtosis), stat_hov_test() (Levene/Bartlett/F).
    • All tests support .by slice grouping (repeat the test within each slice) and multi-column loops with p_adjust.
    • stat_describe(): grouped descriptive statistics.
  • Generic frameworks: stat_bootstrap() built on boot::boot() / boot::boot.ci() (BCa / percentile / basic CIs, parallel support) and stat_perm_test() (label / sign-flip / shuffle permutations with +1 corrected p-values; custom function(data, indices) statistics).
  • export_table(): three-line (booktabs) table export to Word (flextable) and LaTeX from descriptive statistics, stat_* results, and reg_*() regression results.
  • Dependencies (new): boot, flextable.

mathmodels 0.0.13

Statistical inference toolkit (new)

Breaking changes

mathmodels 0.0.12

Breaking changes

  • Renamed the differential equation model functions from model_*() to ode_*() for a consistent prefix with ode_solver(): model_malthus() → ode_malthus(), model_logistic() → ode_logistic(), model_si() → ode_si(), model_sis() → ode_sis(), model_sir() → ode_sir(), model_seir() → ode_seir(), model_lv() → ode_lv().
  • Removed the redundant method argument of plot_Rt_estimate() (the two methods were numerically identical); R_t = beta * S / (gamma * N) is now always used and N is inferred from the known compartments (S, E, I, R) only.

Bug fixes

  • pred_reg: fixed the Hosmer-Lemeshow statistic (wrong column reference and missing variance term — previously reported absurd chi-squares); fixed the factor-response coding in reg_logistic() (events were mapped to -1/0); logistic prediction CIs now use the link scale (previous CIs were far too narrow); .predict_lm() returns both confidence_interval and prediction_interval (the prediction interval was mislabeled); reg_predict() accepts a newdata argument and returns the data used; rank-deficient models no longer break the coefficient table; shapiro.test() is guarded for n > 5000.
  • eval_algorithms: topsis() validates index and w and rejects all-zero columns; sample names default to Sample1..n; rank_sum_ratio() rejects unknown method; grey_corr() guards the degenerate all-equal case and validates w.
  • eval_dea: super_SBM() now implements Tone (2002) super-efficiency (efficient DMUs get rho >= 1 instead of NA); SBM undesirable-output weights follow Tone (2004) (1/s1, 1/s2); basic_DEA(ud_outputs=) now actually treats undesirable outputs as inputs (Hailu-Veeman) instead of silently ignoring them; malmquist(type1 = "seq") implements sequential reference sets (was identical to “cont”); character periods sort numerically; strictly positive inputs/outputs are enforced; SBM rejects the inapplicable orientation argument; malmquist warns on panel irregularities.
  • eval_weights: critic_weight() / entropy_weight() accept matrices again (matrix inputs previously errored element-wise); non-negative data is enforced for entropy paths; combine_weights() rejects unknown type; game_linear() falls back to a pseudo-inverse for proportional weights; AHP() returns CI = CR = 0 for n <= 2 and rejects non-positive matrices; unknown index values are rejected instead of silently treated as “already normalized”.
  • eval_fuzzy: defuzzify() rejects unknown methods and returns the first maximum on ties; fuzzy_eval() validates R in [0, 1] with no NA; membership functions validate their parameters (strict ordering / positivity).
  • pred_grey / pred_ts: combine_preds() handles identical predictions (equal weights); GM models reject non-positive/NA data and near-constant series (a ~ 0) with informative errors; GM1N() returns C/P as documented; ts_back_transform() applies Box-Cox bias adjustment to point forecasts (bias_adj = TRUE default); Ljung-Box diagnostics use ARMA-only fitdf and no longer produce NaN rows for short series; ts_transform() / ts_forecast() validate differencing depth and duplicated levels.
  • diffEq: ode_solver() validates times (strictly increasing) and rejects unused compartments in init; equation parse errors name the offending variable; plot_incidence() warns on non-uniform time steps; epi_metrics() infers N from known compartments only and validates beta/gamma; the b parameter is documented as constant recruitment.
  • eval_system / eval_regional / eval_preprocess / interp_fit: coupling degree validates type and unit-sum weights; id_cols typos are rejected; HHI() handles single values and rejects negatives; entropy / CV weight functions reject NA; preprocessing helpers guard degenerate inputs (rescale_middle, to_positive, rescale_extreme, normalize, standardize); curve_fit()/growth_fit() report AIC/BIC on the original scale (fixes NaN at n = p) with non-finite start-value fallbacks; interp_poly() validates degree.
  • read_nbs: range = "A1" (version-robust header read).
  • DESCRIPTION: dplyr (>= 1.1.0) now declared (reframe() requires it).

Multivariate statistics toolkit (new)

  • Dimension reduction (mv_stats.R):
    • mv_pca(): Principal component analysis with tidy scores, loadings, and variance decomposition.
    • mv_efa(): Exploratory factor analysis (maximum likelihood) with loadings, uniquenesses, variance shares, and optional factor scores.
  • Clustering (mv_stats.R):
    • mv_hclust(): Hierarchical clustering with tidy cluster assignments, merge heights, and the raw hclust object for custom plots.
    • mv_kmeans(): K-means clustering with assignments, centers, and within-cluster sums of squares.
    • mv_cluster_metrics(): Clustering validity metrics computed in base R: per-observation silhouette widths, average silhouette, Davies-Bouldin, Calinski-Harabasz, and Dunn indices.
    • plot_mv_elbow() and plot_mv_silhouette(): Elbow plot (k-means or hierarchical) and silhouette plot for both clustering types; plot_mv_dendrogram() for hierarchical dendrograms.
  • Discriminant analysis (mv_stats.R):
    • mv_discrim(): LDA/QDA via MASS with training predictions, confusion matrix, accuracy, discriminant scores (LDA), and optional classification of new data.
  • Other multivariate methods (mv_stats.R):
    • mv_corresp(): Simple correspondence analysis with row/column coordinates and inertia decomposition.
    • mv_mds(): Metric (cmdscale) and non-metric (MASS::isoMDS) multidimensional scaling.
    • mv_cancor(): Canonical correlation analysis with Wilks’ lambda significance tests, coefficients, and structure correlations.
    • mv_pcor(): Partial correlation matrix with t-tests.
  • Visualization: plot_mv_scree() and plot_mv_scores() for PCA/EFA scree and score plots, plot_mv_corresp() and plot_mv_mds() for CA/MDS configuration maps.
  • Dependencies: unchanged. Tidy-select column interfaces are implemented through the existing dplyr dependency.
  • All functions follow a tidy interface: tidy-select columns in, tibble components out, and print() methods for compact summaries.

mathmodels 0.0.11

  • New function ts_arimax(): Fits ARIMA models with exogenous regressors (ARIMAX). Supports both automatic order selection (forecast::auto.arima()) and manual specification (forecast::Arima()), consistent with the existing ts_sarima() interface.
  • Updated ts_forecast(): Added newxreg parameter for forecasting ARIMAX models with future exogenous regressor values.
  • Improved compute_mf() / compute_mf_funs(): Added .builder parameter supporting Gaussian ("gauss") and sigmoid ("sigmoid") membership functions in addition to the default triangular/trapezoidal ("tri"). Custom builder functions are also accepted. Renamed thresholds parameter to knots for clarity. Fixed defuzzify() floating-point comparison in max_membership method.

Interpolation & curve fitting toolkit (new)

  • Interpolation (interp_fit.R):
  • Curve fitting (interp_fit.R):
    • poly_fit(): Polynomial regression.
    • curve_fit(): Unified interface for linearizable curves via variable transformation (exponential, power-law, logarithmic, hyperbolic).
    • growth_fit(): Nonlinear growth curve fitting via minpack.lm::nlsLM() with automatic starting values (Logistic, Gompertz, exponential saturation, Michaelis-Menten).
  • Refinements:
    • poly_fit() now returns explicit polynomial equations (e.g., y = 3 + 2 * x - 0.5 * x^2) instead of R formula strings.
    • Internal ._compute_fit_stats() avoids redundant AIC/BIC computation for lm models (uses stats::AIC() / stats::BIC() directly; normal-approximation fallback for nls).
    • curve_fit() documentation now notes log-transform back-transformation bias for "exp" and "power" types.
  • Dependencies (new): minpack.lm for Levenberg-Marquardt nonlinear least squares.

mathmodels 0.0.10

Regression prediction toolkit (new)

  • Model fitting (pred_reg.R):
    • reg_lm(): OLS multiple linear regression with optional stepwise selection (forward / backward / both), returning coefficients with CIs and fit statistics.
    • reg_logistic(): Binary logistic regression (glm(binomial)) with stepwise selection, odds ratios, and Hosmer-Lemeshow goodness-of-fit test.
    • reg_poisson(): Poisson regression for count data, with dispersion check.
    • reg_negbin(): Negative binomial regression via MASS::glm.nb() for over-dispersed counts.
  • Diagnostics and prediction (pred_reg.R):
    • reg_diagnostics(): Unified diagnostics — Breusch-Pagan, Durbin-Watson, Shapiro-Wilk (LM); Hosmer-Lemeshow, dispersion/deviance (GLM).
    • reg_predict(): Draw new predictor data from the empirical distribution and generate fitted or predicted values with 95% confidence intervals.
  • Visualization (pred_reg.R):
    • plot_reg_residuals(): Residual diagnostic plots (residuals vs. fitted, normal Q-Q, density).
    • plot_reg_predict(): Prediction plots with confidence bands — supports both fitted (training data) and predicted (new data) modes.

mathmodels 0.0.9

Time series toolkit (new)

Epidemic visualization & metrics overhaul

  • Renamed epidemic_plot.R → epi_plots.R. Consolidated from 3 files into 1.
  • Removed compute_incidence(), plot_infectious_curve(), plot_cumulative_infection().
  • Merged epidemic_metrics() → epi_metrics(): simplified to 4 core scalars (R0, peak_infection, peak_time, attack_rate).
  • Refined plot_incidence(): now shows ΔI (daily new infections) with peak time annotation.

Testing

  • New: test-ts.R (52 tests), test-epi_plots.R (19 tests).
  • Full suite: 539 tests, 0 failures.

mathmodels 0.0.8

New features

  • Differential equation models (diffEq_models.R):
    • ode_solver(): General-purpose string-formula ODE solver powered by deSolve.
    • model_malthus(), model_logistic(), model_si(), model_sis(), model_sir(), model_seir(), model_lv() — all using a unified init + params interface with physics-meaningful named parameters (e.g., beta, gamma, sigma).
  • Epidemic visualization (epidemic_plot.R):
    • plot_compartments(): Faceted/overlaid line plot of selected compartments.
    • compute_incidence(): Daily incidence from S or I differences.
    • plot_incidence(), plot_infectious_curve(), plot_cumulative_infection(): Infection curve visualizations.
    • plot_phase_si(): S–I phase portrait.
    • plot_Rt_estimate(): Effective reproduction number trajectory.
  • Epidemic metrics (epidemic_metrics.R):
    • epidemic_metrics(): Comprehensive epidemic summary (R0, peak, attack rate, trajectory with Rt and growth rate).
  • Markov chain prediction (markov.R):
    • markov_chain(): Transition probability matrix, multi-step prediction, stationary distribution.
    • GM11_markov(): Grey–Markov combined prediction with GM(1,1) and Markov correction.

Dependencies

  • Imports (new): Added deSolve for ODE solving.

Testing

  • 68 tests for diffEq, 23 for epidemic plots, 21 for epidemic metrics, 14 for Markov chain models. Zero errors.

mathmodels 0.0.7

New features

  • DEA models: New DEA.R module implementing Data Envelopment Analysis with:
    • basic_DEA(): Radial models (CCR/BCC) with input/output orientation support, returning both Shephard distances and Farrell efficiencies, plus slacks, targets, and lambda weights.
    • super_DEA(): Super-efficiency radial models, fully compatible with CRS/VRS settings.
    • basic_SBM(): Slacks-Based Measure (non-radial) with CRS/VRS support.
    • super_SBM(): Super-efficiency SBM for distinguishing efficient DMUs.
    • malmquist(): Malmquist productivity index supporting contemporaneous, sequential, and global references with FGNZ and Ray-Desli decompositions.
  • All DEA models use lpSolveAPI as the sole LP solver, with zero external DEA package dependency.

Dependencies

  • Imports (new): Added lpSolveAPI for linear programming in DEA models.
  • Suggests (removed): Removed deaR from Suggests; all DEA functionality is now provided natively.

Parameter validation

  • Added input validation for all DEA functions, including checks for NA values, correct column types, and valid orientation/rts/type parameters.

Testing

  • Added 30 new test cases in test-DEA.R, covering:
    • Basic DEA model validation (CCR, BCC, SBM)
    • Super-efficiency models
    • Malmquist index with all 6 type1 × type2 combinations
    • Undesirable outputs handling
    • Input validation (NA detection, minimum column requirements)
  • All 208 tests pass with zero errors.

Documentation

  • Added @examples sections to all five DEA functions with self-contained inline data.
  • Regenerated man/DEA.Rd with devtools::document().

mathmodels 0.0.6

Bug fixes

  • AHP(): Extended the Random Index (RI) table from n = 11 to n = 15, and added validation for n > 15. Previously, AHP() would silently return NA for consistency ratios with more than 11 criteria.
  • grey_corr(): Fixed incorrect default weight direction in the internal validation; weights are now correctly validated against nrow(cmp) (number of evaluation objects) rather than ncol(cmp) (number of indicators).
  • DGM21(): Fixed the list(pred = fitted, ...) return value where fitted was undefined; corrected to list(fitted = pred, ...). Also fixed the initialization value x0 from a hard-coded constant to the first element of the input series X[1].
  • grey_corr_topsis(): Fixed an internal transpose error in the call to grey_corr() that caused dimension mismatch in non-square matrices.
  • z_mf(), pi_mf(), s_mf(): Rewrote the two-stage spline algorithm for all three membership functions. The original implementation used a single-stage approximation with logically unreachable second-phase conditions, causing values to fall outside the valid [0, 1] range for certain parameter configurations. The fix ensures proper two-stage spline behavior with exact midpoint values of 0.5.

Dependencies

  • Imports (new): Added MASS, purrr, readxl, rlang, stats, stringr, and tibble to DESCRIPTION/Imports. These were already used in code via @importFrom in NAMESPACE but were missing from the formal dependency declaration.
  • Suggests (new): Added deaR to DESCRIPTION/Suggests for DEA model functionality.

Parameter validation

  • Added input type and dimension validation (stopifnot() checks) for all exported functions, including checks for:
    • Data frame / matrix type verification
    • Vector length consistency
    • Non-negativity / positivity of weights
    • Matrix dimension bounds
    • Parameter range constraints

Testing

  • Added 178 test cases across 18 test files, achieving full coverage of all core modules:
Module Test file Tests
AHP test-AHP.R 8
Combine predictions test-combine_preds.R 5
Combine weights test-combine_weights.R 7
CRITIC weighting test-critic.R 5
CV weighting test-cv.R 4
Entropy weighting test-entropy.R 7
Fuzzy evaluation test-fuzzy.R, test-fuzzy_more.R 25
Grey analysis test-grey.R 10
Grey models test-grey_models.R 16
Inequality measures test-inequality.R 12
Linear sum test-linear_sum.R 4
Membership functions test-membership.R 19
PCA weighting test-pca.R 4
Preprocessing test-preprocess.R 21
Rank sum ratio test-rsr.R 5
System evaluation test-system_evaluation.R 9
TOPSIS test-topsis.R 5

Documentation

  • Updated function examples and documentation to reflect all parameter validation rules.
  • Added pkgdown website configuration (_pkgdown.yml). Once pushed to GitHub, the reference site will be automatically built and deployed to https://zhjx19.github.io/mathmodels/ via GitHub Actions.
  • Fixed a typo in the GitHub installation URL within README.md.
  • Updated README.md version badge from 0.0.5 to 0.0.6.
  • The comprehensive online manual is available at: https://zhjx19.github.io/mathmodels-book/

mathmodels 0.0.0.9000

  • Initial development version.