moderndive 0.1.0

📦 Now on PyPI: pip install moderndive — https://pypi.org/project/moderndive/0.1.0/

Initial release of the Python companion to ModernDive: Statistical Inference
via Data Science — a faithful port of the R moderndive and infer packages,
targeting full functional, dataset, and argument-level parity on a modern Python
stack (polars, plotly, plotnine, statsmodels).

Inference grammar (infer)

• Tidy pipeline specify → hypothesize → generate → calculate, plus fit()
  (per-term inference for multiple regression), observe(), and assume()
  (theoretical t/z/F/Chisq). specify() is also a DataFrame method, so
  df.specify(...) mirrors R's df %>% specify(...).
• Full calculate(stat=) vocabulary: mean, median, sum, sd, prop,
  count, diff in means/medians/props, ratio of means/props,
  odds ratio, slope, correlation, t, z, F, Chisq, plus any
  custom callable statistic.
• hypothesize() point nulls for the mean (mu), median (med), proportion
  (p), and one-sample mean z (sigma); a {level: probability} mapping
  drives the chi-square goodness-of-fit test.
• generate() bootstrap / permute / draw (with "simulate" as an alias for
  "draw"), and variables= to choose the permuted column.
• Confidence intervals (percentile, se, bias-corrected) and simulation
  p-values via get_confidence_interval / get_p_value. British-spelling and
  short aliases included.

Theory-based wrappers

• t_test, prop_test, chisq_test, t_stat, chisq_stat, plus the
  moderndive.theory module. prop_test mirrors R's prop.test (chi-square by
  default with Yates correction, z=, conf_int=/conf_level=, and a
  Wilson-score CI for one proportion); chisq_test(response=, p=) runs
  goodness-of-fit.

Regression & summary helpers (moderndive)

• get_regression_table (conf_level, exponentiate,
  default_categorical_levels), get_regression_points (newdata=, ID=,
  in-formula-transformation handling), get_regression_summaries,
  get_correlation (method=pearson/spearman/kendall, multiple predictors,
  wide=, na_rm=), pop_sd, tidy_summary, and count_missing.
• Both OLS and GLM models are supported (GLM points on the response scale,
  GLM-shaped summaries), via the formula and array statsmodels APIs.

Dual-engine plotting (plotly default, plotnine optional)

• Every plot takes engine="plotly" (default, interactive) or
  engine="plotnine": visualize / shade_p_value / shade_confidence_interval
  (with fill=, dens_color=, method="both"/"theoretical", and per-facet
  shading for visualize_fit), pairplot, gg_parallel_slopes /
  geom_parallel_slopes (alpha=), gg_categorical_model /
  geom_categorical_model, and plot_3d_regression. visualize() returns an
  InferPlot that composes with shading via + and renders in Jupyter/Quarto
  for both engines. View() renders an interactive table via the optional
  itables package. Static image export uses the optional moderndive[image]
  extra (kaleido); both extras are optional so the core stays pure-Python and
  Pyodide-installable.

Datasets

• 58 bundled datasets via load_*() loaders returning polars DataFrames — the R
  moderndive/infer, nycflights23, gapminder, ISLR2, and FiveThirtyEight
  data used in the book.

Quality & tooling

• Beginner-friendly, infer-style messages and errors (a summary line + →
  hint bullets; suppressible ModernDiveMessage notes).
• 100% statement coverage enforced in CI, including a vignette smoke test and
  R-validated reference values; parity-drift CI against the upstream R packages;
  docs on Read the Docs with runnable examples.