bunker-stats · Notebook UX

>>> nb.robust_summary(df)

Robust + classical descriptives per column: count, mean, std, median, MAD, IQR, Qn scale, trimmed mean, skew, kurtosis. Non-finite values are dropped and counted in n_missing.

>>> nb.describe_fast(df)

A faster, richer df.describe().T backed by the Rust kernels. Adds a robust block (MAD, IQR, Qn, trimmed mean, skew, kurtosis) alongside the quartiles.

>>> nb.outlier_report(df, method="iqr", k=1.5)

Per-column outlier counts, percentages and fence bounds. Methods: iqr, zscore, robust_zscore (median/MAD fences).

>>> nb.normality_report(df)

Jarque-Bera and Anderson-Darling diagnostics. The normal/conclusion verdict uses the JB p-value; A-D reports its statistic only (no p-value in the kernel).

>>> nb.correlation_report(df, pvalues=True)

Correlation between numeric columns using pairwise-complete rows. Long form (pvalues=True) gives one row per pair with the test statistic and p-value; the default returns a square matrix.

>>> nb.missingness_report(df)

The data-quality audit: separates NaN from ±inf from finite, per column, for every dtype. The one helper that counts rather than drops.

>>> nb.rolling_report(df, "price", window=5)

Rolling-window features via the fused Rust kernel (all stats in one pass). Right-aligned and index-preserving. Showing the first 10 rows.

>>> nb.bootstrap_ci_report(df, stat="mean", random_state=0)

Bootstrap point estimate and confidence interval per column, via BootstrapConfig. Deterministic given random_state.

>>> nb.scale_columns(df, ["price", "latency"], method="robust")

Batch scaling (robust / zscore / minmax). Fit on finite values only, scattered back so NaN positions and row order are preserved. Showing the new columns beside their sources.

>>> nb.winsorize_columns(df, ["price"], lower_q=0.05, upper_q=0.95)

Batch winsorization — clip tails at the given quantile fractions. Note how the 210 and 38 outliers are pulled to the fences while other rows are untouched.

>>> nb.outlier_style(df, ["price", "demand"])

Highlights outlier cells across many numeric columns at once (red). Non-finite cells are left unstyled. Showing the first 16 rows.

>>> nb.corr_heatmap(df)

Correlation matrix as a diverging background-gradient Styler (needs matplotlib). price↔demand shows the planted correlation.

>>> nb.style_significance(results, alpha=0.05)

Shades a results table by significance tier (p<α/50, p<α/5, p<α, then non-significant). NaN p-values stay unstyled.

>>> nb.style_effect_size(results, "cohens_d")

Shades an effect-size column by |magnitude| against thresholds (default Cohen's 0.2/0.5/0.8: negligible→small→medium→large).

>>> nb.demean_style(df, "price")

Legacy single-column styler (hardened): adds a demeaned column and colors each cell above (green) / below (red) the mean. Showing 14 rows.

>>> nb.zscore_style(df, "price", threshold=2.0)

Adds a z-score column and highlights scores beyond ±threshold (high = orange, low = blue). Showing 14 rows.

>>> nb.iqr_outlier_style(df, "price", k=1.5)

Single-column IQR outlier highlight — a thin wrapper over the multi-column outlier_style. Showing 16 rows.