stats

The stats collection provides common statistical operations — descriptive statistics, hypothesis tests and regression models — over tabular data. It is written in R, which has first-class, well-tested implementations of these methods in its base stats package, so the blocks are thin, dependency-light wrappers around functions statisticians already trust (summary, cor, t.test, aov, chisq.test, table, lm, glm).

The descriptive and test blocks share one shape: a single data input (type: file, format: CSV) and a single result output (type: json). They compose behind any source that produces a CSV — including data.read, fiadb.fullreport, or upstream sae/base blocks — and feed any downstream consumer that accepts JSON. All other configuration (column names, methods, test options) is passed as scalar args, never wired inputs.

Descriptive statistics:

• stats.summary — per-column descriptive statistics.
• stats.frequency — one-way or cross-tabulated frequency tables.
• stats.correlation — a correlation matrix over numeric columns.

Hypothesis tests:

• stats.t_test — one- or two-sample Student's t-test.
• stats.chisq_test — chi-squared test of independence.
• stats.anova — one-way analysis of variance.

Models:

• stats.lm — linear regression by least squares.
• stats.glm — generalised linear models: gaussian, binomial, poisson, Gamma.

The model blocks follow the model-fitting contract shared with the sae, ml and models collections. They read Parquet or CSV, and name the model either with an R formula or with a target and a features column selector (names, globs and exclusions such as B*,MB*,-B0), plus id_columns, weights, categorical and na_action. A missing or misspelled column fails with near-match suggestions and the list of columns that exist. Each writes a fitted model (a SpadeModel file), a coefficient table and fit statistics, with optional cross-validation by rows or by groups (cv_group). The model feeds models.predict, which scores it without R, and models.linear / models.glm fit the same models on tables too large for memory.

All blocks in this collection declare network: false — they operate purely on data already produced elsewhere in the pipeline.

Blocks🔗

Installation🔗

spade install file:///path/to/blocks/stats