zedda.compare()

Compare two datasets for schema drift, distribution shift, and target-column health.

zedda.compare(
    path_a,
    path_b,
    sample_size=None,
    correlate=False,
) -> None

Compare two datasets for schema drift, distribution shift, and target-column health. Prints a Rich comparison table. Returns None.

Arguments

Argument Type Default Description
path_a str / Path / DataFrame (required) First dataset (typically train)
path_b str / Path / DataFrame (required) Second dataset (typically test)
sample_size int or None None Sample size in rows applied to both datasets
correlate bool False Force O(N²) correlation on each dataset

Returns

None. compare() is print-first.

Output

The comparison table has five sections:

  1. Schema diff — columns in A but missing in B (and vice versa), and type mismatches on shared columns.
  2. Null-rate changes — for each shared column, the delta in null percentage from A to B.
  3. Distribution shift — for each shared numeric column, the relative shift in mean from A to B. ID-like and binary-target columns are skipped. Stable if |shift| < 5%, shift if |shift| ≥ 10%.
  4. New categories — for each shared string column, categories present in B but not in A.
  5. Verdict — PASS, REVIEW, or FAIL.

Verdict logic

Verdict Triggered when
FAIL Any non-target-like column missing in B, or any type mismatch on a shared column
REVIEW Any distribution shift on a shared numeric column, or any new category in a string column
PASS None of the above

A column is treated as target-like if its name starts with survived, target, label, y, class, outcome, is_, or has_. Target-like columns missing in B are downgraded from a FAIL trigger to a REVIEW trigger.

Example

import zedda as zd

zd.compare("train.csv", "test.csv")

# Compare two DataFrames
import pandas as pd
train = pd.read_csv("train.csv")
test = pd.read_csv("test.csv")
zd.compare(train, test)

Programmatic access

compare() returns None. For CI, the underlying helpers are exposed in zedda._compare:

  • compute_schema_diff(profile_a, profile_b) — returns the diff structure
  • compute_distribution_shift(profile_a, profile_b) — returns per-column shift data
  • compute_verdict(schema_diff, distribution_shift, ...) — returns the verdict string

See also