Optionaloptions: { confidence?: number; maxLag?: number }Augmented Dickey–Fuller unit-root test on a series column.
Optionaloptions: { lags?: number; regression?: "c" | "ct" | "n" }Accelerated life testing on a stress column.
Optionaloptions: {One-way ANCOVA: y by group adjusting for covariate.
One-way ANOVA of column across the levels of by (Minitab One-Way ANOVA).
Two-way ANOVA Type III via anovaTwoWay.
Optionaloptions: { interaction?: boolean }ARIMA / SARIMA / ARIMAX via CSS-ML.
Optionaloptions: {Auto-ARIMA (stepwise AIC/BIC).
Optionaloptions: {Automated model selection over the tree learners (+ OLS / logistic).
Optionaloptions: {Best Subsets over the predictor columns.
Optionaloptions: { include?: string[]; maxK?: number; nBest?: number }Boxplot statistics, optionally per level of by.
Optionalby: stringProcess capability (normal) of a column.
Capability Sixpack plot-ready panels.
CART on predictors → response.
Optionaloptions: CartOptionsChi-square test of independence between two categorical columns.
Optionaloptions: { correction?: boolean }Cluster Variables (correlation-distance hierarchical clustering).
Optionaloptions: {Optionalopts: ExecuteOptionsControl chart on a column (I-MR by default; pass subgroup / type for X̄-R / X̄-S / attributes).
Optionaloptions: ControlChartOptionsOptionaloptions: { columns?: string[]; method?: CorrMethod }Correlation between two columns with p-value and CI.
Optionaloptions: {Cox PH regression; predictors are numeric columns.
Optionaloptions: {Tabular CUSUM chart.
Optionaloptions: { h?: number; k?: number; mu?: number; sigma?: number; vMask?: boolean }Classical seasonal decomposition.
Display Descriptive Statistics for one or more columns, optionally by a grouping column.
Optionaloptions: { by?: string }Dunnett comparisons of every level of by against control.
Optionaloptions: { strict?: boolean }Test for equal variances: Levene / Brown–Forsythe (default, as Minitab), Bartlett, or Bonett.
Optionalmethod: "levene" | "bartlett" | "bonett"Equivalence test (TOST): one-sample by default, two-sample with by (test level − reference level,
pass reference to name it), paired with paired.
Exponential smoothing (SES / DES / Winters).
Optionaloptions: {EWMA chart.
Optionaloptions: { L?: number; lambda?: number; mu?: number; sigma?: number }Fine–Gray competing risks.
Fisher LSD pairwise comparisons (individual error rate).
Optionaloptions: { alpha?: number }Fitted Line Plot: polynomial (degree 1–3) regression of y on x with the fitted curve.
Optionaloptions: { confidence?: number; degree?: 2 | 1 | 3; logX?: boolean; logY?: boolean }Friedman test: column by treatment within block (one observation per cell).
StaticfromStaticfromParse Apache Arrow IPC bytes (stream or file format) — e.g. tableToIPC(table) from apache-arrow, a
DuckDB arrow() result, pyarrow / Polars output. Int8…Int64, UInt8…UInt64, Float16/32/64, Bool, Utf8,
LargeUtf8, Timestamp (any unit), Date32/64, Null and dictionary-encoded strings are accepted; nested,
decimal, binary and compressed batches throw with the column name. S is an assertion, not verified.
Optionaloptions: ArrowIpcReadOptionsStaticfromStaticfromOptionaloptions: { copy?: boolean }
Optionalcopy?: booleanDefault true: typed arrays and pre-encoded codes are copied, so the frame is immutable and later
writes to your arrays cannot change it (or a persist()ed result derived from it). false shares
your buffers zero-copy — faster for very large inputs, but the frame then aliases memory you can still
write to: its plans are never cached by persist(), and results depend on what the buffers hold when
collect() runs.
StaticfromParse CSV text. S is an assertion about the file, not something the compiler can verify.
Optionaloptions: ReadCsvOptionsStaticfromOptionaloptions: ReadCsvOptionsStaticfromOptionaloptions: ReadCsvOptionsStaticfromStaticfromBuild a frame from row objects; column dtypes are inferred from every row. BigInt cells follow
options.int64 (see Int64Policy): by default a column of safe BigInts becomes an exact f64 column and a
value beyond ±(2^53 − 1) raises PrecisionLossError; 'string' keeps exact decimal strings.
Build a frame from row objects; column dtypes are inferred from every row. BigInt cells follow
options.int64 (see Int64Policy): by default a column of safe BigInts becomes an exact f64 column and a
value beyond ±(2^53 − 1) raises PrecisionLossError; 'string' keeps exact decimal strings.
Optionaloptions: FromRowsOptions<"number" | "error">Build a frame from row objects; column dtypes are inferred from every row. BigInt cells follow
options.int64 (see Int64Policy): by default a column of safe BigInts becomes an exact f64 column and a
value beyond ±(2^53 − 1) raises PrecisionLossError; 'string' keeps exact decimal strings.
Gage R&R (ANOVA) with part / operator / measurement columns.
Games–Howell pairwise comparisons.
Optionaloptions: { alpha?: number }G chart (opportunities between events) on a count column.
Optionaloptions: { alpha?: number; estimator?: "mle" | "mvue"; limits?: "probability" | "sigma" }The runtime plans derived from this frame execute on.
Generalized linear model; family: 'binomial' is Binary Logistic, 'poisson' Poisson Regression. trials / offset / weights name columns.
GLMM (binomial/Poisson/negbin) with random intercept (+ optional slope).
Graphical Summary: descriptives, Anderson–Darling, CIs, histogram and boxplot data.
Optionaloptions: { bins?: number; confidence?: number }Interaction plot data (cell means) for two factor columns.
Interval plot data: group means with confidence intervals.
Optionaloptions: { confidence?: number; pooled?: boolean }Item Analysis (Cronbach's α) over item columns.
Kaplan–Meier estimate.
Optionaloptions: { censor?: string; confidence?: number }k-means on the named numeric columns.
KPSS stationarity test on a series column.
Optionaloptions: { lags?: number; regression?: "c" | "ct" }Kruskal–Wallis test across the levels of by.
Two-sample Kolmogorov–Smirnov between the two levels of by.
Lasso regression of y on predictor columns.
Optionaloptions: { alpha?: number; maxIter?: number; tol?: number }Regression with life data (AFT) with optional censor column (0 exact, 1 right, 2 left, 3 interval + time2).
Optionaloptions: {General Linear Model from a formula over the columns: df.linearModel('y ~ a*b + x').
Optionaloptions: { confidence?: number; factors?: string[] }Binary Logistic Regression (glm with the binomial family).
Optionaloptions: Omit<GlmOptions, "names" | "weights" | "trials" | "offset" | "family"> & {Mantel–Haenszel / Wilcoxon / Tarone–Ware log-rank across levels of group.
Optionaloptions: {Main effects plot data for several factor columns.
Mann–Whitney test between the two levels of by (first level − second, sorted).
Optionaloptions: MannWhitneyOptionsOne-way MANOVA on response columns by group.
General MANOVA: responses on a model formula right-hand side, e.g. 'a*b + x'.
Optionaloptions: { factors?: string[] }MEWMA chart on several columns.
Optionaloptions: { arl0?: number; h?: number; lambda?: number }Random-intercept (+ optional slope / nested) linear mixed model.
Nominal Logistic Regression (multinomial).
Optionaloptions: { confidence?: number; maxIter?: number; reference?: string | number }Nonlinear regression of column y on column(s) x with a model function and start values.
Normality test: Anderson–Darling (default, as Minitab), Ryan–Joiner, Kolmogorov–Smirnov or Shapiro–Wilk.
Optionalmethod: NormalityMethodPer-column unique counts (sync peek).
Ordinal Logistic Regression (proportional odds) of a categorical / integer response.
Optionaloptions: { confidence?: number; maxIter?: number }Orthogonal (Deming) regression of y on x.
Optionaloptions: { confidence?: number; errorVarianceRatio?: number }Outlier test: Grubbs (default) or Dixon.
Optionaloptions: {Pareto of a categorical column.
Optionaloptions: { weights?: string }Partial correlation controlling for one or more columns.
Optionaloptions: { alternative?: Alternative; method?: "pearson" | "spearman" }PCA on the named numeric columns.
Optionaloptions: { nComponents?: number; scale?: boolean }Periodogram / spectral analysis of a series column.
Optionaloptions: { detrend?: "mean" | "none" | "linear"; spans?: number[]; taper?: number }Partial least squares of one or more response columns on the predictor columns.
Optionaloptions: Omit<PlsOptions, "names" | "responseNames">Goodness-of-fit test for Poisson on a column of counts.
Optionaloptions: { minExpected?: number }Power-law NHPP (Crow–AMSAA) on a column of failure times, optionally per system.
Probit analysis from events / trials / stress columns.
Optionaloptions: {1 or 2 Proportions: without by, sums eventsCol and trials for propTest1; with by (2 levels),
aggregates per group (binary column or events + trials columns) for propTest2.
Random Forest on predictors → response.
Optionaloptions: CartOptions & { nTrees?: number }StaticreadRead Apache Arrow IPC (stream or file / Feather v2) from a path, URL or bytes under the IO policy.
Optionaloptions: IoPolicy & {Optionalcontent?: booleanTreat a string source as raw content, never as a path (alias of mode: 'text').
Optionalencoding?: stringText encoding when reading bytes/files (default utf-8).
Optionalfloors?: IoPolicy[]Additional policy floors — a Session's policy, for instance. Every floor is enforced like the
process-wide one (each must pass; byte caps and timeouts take the minimum), and per-call fields
cannot widen any of them.
Optionalmode?: IoSourceModeInterpret a bare string as content / path / URL instead of guessing (default 'auto').
StaticreadStaticreadOptionaloptions: ReadCsvOptionsStaticreadAlias of readSql.
Optionaloptions: ReadSqlOptionsStaticreadStaticreadRead from a browser File / Blob (CSV/JSON inferred by name or format).
Optionaloptions: { format?: "csv" | "json" | "parquet" | "excel" } & IoPolicy & {Optionalformat?: "csv" | "json" | "parquet" | "excel"Optionalcontent?: booleanTreat a string source as raw content, never as a path (alias of mode: 'text').
Optionalencoding?: stringText encoding when reading bytes/files (default utf-8).
Optionalfloors?: IoPolicy[]Additional policy floors — a Session's policy, for instance. Every floor is enforced like the
process-wide one (each must pass; byte caps and timeouts take the minimum), and per-call fields
cannot widen any of them.
Optionalmode?: IoSourceModeInterpret a bare string as content / path / URL instead of guessing (default 'auto').
Optionalcomment?: stringLines starting with this prefix are ignored.
Optionaldecimal?: stringDecimal mark (default .).
Optionaldelimiter?: stringAlias of separator (pandas-style).
Optionaldtypes?: Record<string, DType>Force column dtypes after parse.
OptionalfalseValues?: string | string[]Values treated as boolean false.
OptionalhasHeader?: booleanPolars alias for header !== false.
Optionalheader?: number | booleanFirst row is header (default true). Number = header row index after skipRows.
Optionalnames?: string[]Explicit column names (used when header is false, or to override).
OptionalnRows?: numberMax data rows to read (after header).
OptionalnullValues?: string | string[]Values treated as null.
OptionalquoteChar?: stringQuote character (default ").
Optionalseparator?: stringColumn separator (default ,). Alias of delimiter.
OptionalskipBlankLines?: booleanSkip empty lines (default true).
OptionalskipInitialSpace?: booleanTrim whitespace after delimiter (default false).
OptionalskipRows?: numberRows to skip at the start of the file (before header).
Optionalthousands?: stringThousands separator to strip before parsing numbers.
OptionaltrueValues?: string | string[]Values treated as boolean true.
Optionalusecols?: (string | number)[]Keep only these columns (names or 0-based indices).
Optionalint64?: Int64PolicyInteger literals beyond ±(2^53 − 1) (JSON.parse would round them silently): 'error' (default) throws
PrecisionLossError, 'string' keeps the column as exact decimal strings, 'number' accepts the nearest double.
Optionallines?: booleanRead newline-delimited JSON (NDJSON). Alias of orient:'lines'.
OptionalnRows?: numberMax records to read.
Optionalorient?: "records" | "columns" | "values" | "lines"records = array of objects (default); columns = {col: values[]}; lines = NDJSON.
OptionalskipRows?: numberSkip first N lines (NDJSON) or N array elements.
StaticreadStaticreadConsume a bounded batch from Kafka into a DataFrame.
Nested JSON values are flattened to dotted columns by default
(user.address.city). Driver: optional peer kafkajs.
Optionaloptions: Partial<ReadKafkaOptions>StaticreadStaticreadLoad a query result from a database (pandas read_sql / polars read_database).
Built-in dialects: postgres, mssql, clickhouse, mysql, sqlite.
Drivers are optional peer packages (pg, mssql, @clickhouse/client, mysql2, better-sqlite3).
Optionaloptions: ReadSqlOptionsRegression (OLS) of y on the predictor columns: coefficient table, ANOVA, R², diagnostics, predict().
Optionaloptions: Omit<OlsOptions, "names" | "weights"> & { weights?: string }Parametric reliability fit (Weibull / lognormal / exponential).
Optionaloptions: {Ridge regression of y on predictor columns.
Optionaloptions: { alpha?: number }Run chart tests on a column.
Optionaloptions: { alpha?: number }Runs test for randomness about the mean (or k).
Optionaloptions: { correction?: boolean; k?: number }Print head(n) as markdown to the console; returns this for chaining.
Optionaln: number1-Sample Sign test.
Optionaloptions: { alternative?: Alternative; confidence?: number; median?: number }Stability Study: response vs time with a batch factor; shelf life against spec limits.
Stepwise / forward / backward selection among the predictor columns.
Optionaloptions: Omit<StepwiseOptions, "names">STL (LOESS) seasonal-trend decomposition.
Hotelling T² chart on several columns (optionally subgrouped).
Optionaloptions: { alpha?: number; phase?: 2 | 1; subgroup?: string }T chart (time between events).
Optionaloptions: { alpha?: number; distribution?: "exponential" | "weibull" }Rows as objects, typed by the schema.
Apache Arrow IPC bytes — the streaming format by default ({ format: 'file' } for Feather v2). Zero-copy
consumers: tableFromIPC() of apache-arrow, DuckDB(-Wasm) insertArrowFromIPCStream, pyarrow.ipc,
polars.read_ipc_stream, pandas.read_feather (file format). f64/f32/i32/u32/bool/utf8 map 1:1,
category → Dictionary<Int32, Utf8>, datetime → Timestamp(ms). Round-trips with DataFrame.fromArrowIpc.
Optionaloptions: ArrowIpcWriteOptionscolumna's own JSON columnar form (schema + plain arrays). Not Apache Arrow — see toArrowIpc().
Optionalformat: "csv" | "json" | "parquet" | "parquet-like"Optionaloptions: CSV text. Options mirror pandas to_csv / ReadCsvOptions (separator/delimiter, encoding,
quoteChar, header, naRep, decimal, lineTerminator, columns). { escapeFormulas: true }
neutralises text cells a spreadsheet would run as formulas — off by default.
Optionaloptions: CsvWriteOptionsOptionalmaxRowsOrOpts: number | PresentWriteOptionsJSON text (orient: records | columns | values | lines).
Optionaloptions: JsonWriteOptionsTolerance interval for a column.
Optionaloptions: { confidence?: number; coverage?: number; method?: "normal" | "nonparametric" }OptionalmaxRowsOrOpts: number | PresentWriteOptionsTrend Analysis on a time-series column.
Optionaloptions: { horizon?: number; model?: "linear" | "quadratic" | "exponential" | "s-curve" }t-test on a column (Minitab 1-Sample t / 2-Sample t / Paired t): df.ttest('x', { mu: 50 }) // one-sample, H0: mean = 50 df.ttest('x', { by: 'group' }) // two-sample (Welch) between the two levels of group df.ttest('after', { paired: 'before' }) // paired, after − before
Optionaloptions: DataFrameTTestOptionsTukey HSD pairwise comparisons across the levels of by.
Optionaloptions: { alpha?: number }1 Variance: χ² (default) or Bonett.
1-Sample Wilcoxon signed-rank test.
Optionaloptions: {Optionaloptions: {Optionalexpr?: AnyExprOptionalmethod?: RankMethodrank only: tie handling, default 'average' (pandas / Minitab)
Optionaloffset?: numberOptionalorderBy?: (string | AnyExpr | { descending: boolean; expr: AnyExpr })[]OptionalpartitionBy?: string[]Write Apache Arrow IPC to a Node path (.arrow / .feather with { format: 'file' }) or return the bytes.
Optionalpath: stringOptionaloptions: ArrowIpcWriteOptionsWrite CSV to a path (streamed, memory bounded by one chunk; resolves to '') or return the text when no path is given.
Optionalpath: stringOptionaloptions: CsvWriteOptionsWrite Excel (.xlsx) bytes; optionally to a Node path.
Optionalpath: stringOptionaloptions: ExcelWriteOptionsWrite JSON to a path (resolves to '') or return the text when no path is given.
Optionalpath: stringOptionaloptions: JsonWriteOptionsWrite Apache Parquet (hyparquet-writer; round-trips with readParquet). Default codec SNAPPY.
Optionalpath: stringOptionaloptions: ParquetWriteOptionsWrite columna's JSON "parquet-like" format (round-trips with readParquetLike, not Apache Parquet).
Optionalpath: stringOptionaloptions: ParquetLikeWriteOptions1-Sample Z with known sigma.
Sample ACF.