Coming from R¶

If you know the R moderndive and infer packages, this page maps the API to Python. The grammar is the same; the main differences are Python method-chaining (.hypothesize() instead of the |>/%>% pipe) and polars DataFrames.

The infer pipeline¶

# R
pennies %>%
  specify(response = year) %>%
  hypothesize(null = "point", mu = 1995) %>%
  generate(reps = 1000, type = "bootstrap") %>%
  calculate(stat = "mean")

# Python — verbs are methods on the returned objects
(
    md.load_pennies()
    .specify(response="year")
    .hypothesize(null="point", mu=1995)
    .generate(reps=1000, type="bootstrap", seed=1)
    .calculate(stat="mean")
)

specify(formula="y ~ x") works just like R’s formula interface; success= marks the success level for categorical responses.

Most names are identical¶

The overwhelming majority of functions keep the same name (including the British-spelling and short-form aliases). They’re called just as in R — as methods on the pipeline where applicable:

specify, hypothesize/hypothesise, generate, calculate, fit, assume, observe, get_p_value/get_pvalue, get_confidence_interval/get_ci, visualize/visualise, shade_p_value/shade_pvalue, shade_confidence_interval/shade_ci, t_test, prop_test, chisq_test, t_stat, chisq_stat, rep_sample_n/rep_slice_sample, get_regression_table, get_regression_points, get_regression_summaries, get_correlation, pop_sd, tidy_summary, geom_parallel_slopes, geom_categorical_model.

What’s actually different¶

R	Python	Why
`x %>% f(...)` / `x \|> f(...)`	`x.f(...)` (method chaining)	no pipe operator in Python
ggplot2 `geom_*` layers	plotly by default; `engine="plotnine"` for ggplot-style	dual-engine plotting
`lm(y ~ x, data)` object	a fitted statsmodels model: `smf.ols("y ~ x", data=df.to_pandas()).fit()`	regression backend
`get_correlation(df, y ~ x)`	`get_correlation(df, "y ~ x")` or `get_correlation(df, x="x", y="y")`	formula passed as a string

Other things to know¶

Plots compose with + in both engines, and default to plotly (interactive); pass engine="plotnine" anywhere for ggplot-style output.
DataFrames are polars in and out; pass .to_pandas() when a downstream tool needs pandas.
Reproducibility: pass seed= to generate() (R uses set.seed()).

Same datasets¶

Most R moderndive/infer datasets are bundled here under the same name — load_pennies(), load_mythbusters_yawn(), load_gss(), etc. See Datasets.