Getting started with ksCompare

ksCompare compares two data frames in the spirit of SAS PROC COMPARE — column-by-column, row-by-row, with tolerances and labelled-metadata awareness — and returns a structured ks_comparison object you can print, query, export, or render to HTML / Excel.

A first comparison

base <- data.frame(
  id   = 1:5,
  age  = c(34, 41, 28, 55, 19),
  arm  = c("A", "A", "B", "B", "A"),
  stringsAsFactors = FALSE
)
comp <- data.frame(
  id   = 1:5,
  age  = c(34, 41, 27, 55, 19),
  arm  = c("A", "A", "b", "B", "A"),
  stringsAsFactors = FALSE
)

cmp <- ks_compare(base, comp, by = "id")
#> ⚠ base vs comp — 2 value diffs across 2 columns
cmp
#> 
#> ── ksCompare comparison ────────────────────────────────────────────────────────
#> ℹ Verdict: 86.67% of matched cells equal; 2 diffs across 2 columns (top cause: letter case differs).
#> Base: base (5 × 3)
#> Comp: comp (5 × 3)
#> Keys: id
#> 
#> ── Schema ──
#> 
#> • Matched columns: 3
#> 
#> ── Rows ──
#> 
#> • Strategy: keyed match on "id" (unique on both sides)
#> • Matched: 5
#> • Base-only: 0
#> • Comp-only: 0
#> 
#> ── Values ──
#> 
#> ! 2 value differences across 2 columns.
#> • age: 1 diff
#> • arm: 1 diff

summary(cmp) returns a structured object with per-section counts; as_tibble(cmp) returns the long-format value-diff table.

summary(cmp)
#> 
#> ── ksCompare summary ───────────────────────────────────────────────────────────
#> Rows (base / comp): 5 / 5
#> Matched / base-only / comp-only rows: 5 / 0 / 0
#> Matched / base-only / comp-only columns: 3 / 0 / 0
#> Cells with diffs: 2 (in 2 columns)
tibble::as_tibble(cmp) |> head()
#> # A tibble: 2 × 11
#>   key_id base_row comp_row column_base column_comp kind      base  comp   diff
#>    <int>    <int>    <int> <chr>       <chr>       <chr>     <chr> <chr> <dbl>
#> 1      3        3        3 age         age         double    28    27        1
#> 2      3        3        3 arm         arm         character B     b        NA
#> # ℹ 2 more variables: na_flow <chr>, note <chr>

What’s inside `cmp`?

A ks_comparison is a list with named slots; you can read them directly when scripting QC dashboards:

names(cmp)
#>  [1] "meta"               "schema_diff"        "key_diff"          
#>  [4] "row_diff"           "value_diff"         "pattern_summary"   
#>  [7] "unmatched_rows"     "first_last_unequal" "options"           
#> [10] "tolerance"          "manifest"           "verdict"

cmp$schema_diff – one row per column pair (matched, base-only, or comp-only) with kind_match, label_match, format_match.
cmp$row_diff – one row per matched / unmatched row with key_id, base_row, comp_row, status.
cmp$value_diff – one row per differing cell with kind, base, comp, diff, note. Same as as_tibble(cmp) and ks_tidy(cmp).
cmp$pattern_summary – recurring shapes per column (e.g. case_only, constant_offset) when find_patterns = TRUE was requested; otherwise an empty tibble with the usual columns.
cmp$unmatched_rows – full base-only / comp-only observations, capped by max_unmatched_rows.
cmp$first_last_unequal – PROC COMPARE-style first / last differing observations per matched column.
cmp$meta – counts, key columns, row-matching strategy, and the rendered key labels used by reports.
cmp$manifest – input hashes + run timestamp for QC traceability.
cmp$verdict – executive headline used by print(cmp) and the HTML report.

cmp$schema_diff
#> # A tibble: 3 × 14
#>   base  comp  side    kind_base kind_comp kind_match label_base label_comp
#>   <chr> <chr> <chr>   <chr>     <chr>     <lgl>      <chr>      <chr>     
#> 1 id    id    matched integer   integer   TRUE       <NA>       <NA>      
#> 2 age   age   matched double    double    TRUE       <NA>       <NA>      
#> 3 arm   arm   matched character character TRUE       <NA>       <NA>      
#> # ℹ 6 more variables: label_match <lgl>, label_diff <chr>, format_base <chr>,
#> #   format_comp <chr>, format_match <lgl>, format_diff <chr>
cmp$value_diff
#> # A tibble: 2 × 11
#>   key_id base_row comp_row column_base column_comp kind      base  comp   diff
#>    <int>    <int>    <int> <chr>       <chr>       <chr>     <chr> <chr> <dbl>
#> 1      3        3        3 age         age         double    28    27        1
#> 2      3        3        3 arm         arm         character B     b        NA
#> # ℹ 2 more variables: na_flow <chr>, note <chr>
ks_glance(cmp)
#> # A tibble: 1 × 8
#>   n_base_rows n_comp_rows n_matched_rows n_base_only_rows n_comp_only_rows
#>         <int>       <int>          <int>            <int>            <int>
#> 1           5           5              5                0                0
#> # ℹ 3 more variables: n_matched_columns <int>, n_value_diffs <int>,
#> #   n_columns_with_diffs <int>

Triage helpers

When a comparison is not clean, these helpers narrow the search space quickly:

ks_cause_summary(cmp)
#> # A tibble: 2 × 4
#>   cause               n_cells n_columns columns
#>   <chr>                 <int>     <int> <chr>  
#> 1 letter case differs       1         1 arm    
#> 2 plain value change        1         1 age
ks_row_diff_summary(cmp)
#> # A tibble: 1 × 6
#>   key_id base_row comp_row key_label n_diffs columns 
#>    <int>    <int>    <int> <chr>       <int> <chr>   
#> 1      3        3        3 id = 3          2 age, arm

Use ks_unmatched_rows(cmp) to inspect full base-only / comp-only records, and rerun with find_patterns = TRUE when you want cmp$pattern_summary populated.

Tolerances

ks_compare(
  data.frame(id = 1, x = 1.0),
  data.frame(id = 1, x = 1.0001),
  by = "id",
  tolerance = ks_tol(abs = 1e-3)
)
#> ✔ data.frame(id = 1, x = 1) vs data.frame(id = 1, x = 1.0001) — identical

Per-column overrides:

ks_tol(
  abs = 0,
  per_column = list(
    weight = ks_tol(abs = 0.01),
    height = ks_tol(rel = 0.001)
  )
)

ULP tolerance is useful when you only want to absorb floating-point round-trip noise:

ks_compare(base, comp, by = "id", tolerance = ks_tol(ulp = 4))

Options

ks_comp_options() controls semantics that apply to all columns:

na_equal: treat NA == NA as equal (default TRUE).
sas_special_missing: distinguish .A-.Z and ._ SAS missings (haven-imported numerics only).
compare_labels / compare_formats: schema-level checks. Format comparison is trailing-dot- and case-tolerant (DATE9. == DATE9 == date9.).
str_trim, str_case, str_norm: string normalisation. Strings are always compared encoding-safely (Latin1 vs UTF-8 will not produce false diffs).
tz: "preserve" (default), "UTC", or "strip" for POSIXct.

ks_compare(
  base, comp, by = "id",
  options = ks_comp_options(str_trim = TRUE, str_case = "fold")
)

Renamed columns

Use mapping = for renamed non-key columns and a named by = for a renamed key:

ks_compare(
  base, comp,
  by      = c(USUBJID = "SUBJID"),
  mapping = c(AGE = "AGE_YEARS")
)

Reading files directly

ks_compare() accepts file paths in addition to in-memory data frames and dispatches the right reader by extension:

# SAS transport file
ks_compare("prod/adsl.xpt", "qc/adsl.xpt", by = "USUBJID")

# .sas7bdat
ks_compare("prod/adsl.sas7bdat", "qc/adsl.sas7bdat", by = "USUBJID")

# Parquet / Feather
ks_compare("a.parquet", "b.parquet", by = "id")

# RDS / RData / CSV / TSV are also recognised
ks_compare("a.rds", "b.rds", by = "id")

When both paths share the same basename (prod/adsl.xpt vs qc/adsl.xpt), display names are automatically disambiguated using the parent folder, so print(cmp) and the HTML report show prod/adsl vs qc/adsl instead of two identical labels.

Duplicate keys

By default dup_keys = "first" keeps the first occurrence per side (emitting a ksCompare_dup_keys_resolved info message). Other strategies:

"last" – keep the last per side.
"keep_all" – pair duplicates positionally within each key.
"all_pairs" – cartesian-pair duplicates (warns when group sizes differ).
"error" – raise on any duplicate key.

ks_compare(base, comp, by = "id", dup_keys = "keep_all")

Next steps

From SAS: see vignette("from-proc-compare") for a side-by-side mapping of PROC COMPARE options.
Smart features: see vignette("smart-features") for auto-key inference, duplicate-key handling, pattern detection, and SAS-style first / last unequal summaries.
Reports: see vignette("reports") for HTML and Excel output.
CI / pipelines: see vignette("pipeline-gates") for using comparisons as assertions.