On this page

Constraints

Enforce business rules that survive every row of generation. Constraints run at generation time, not as post-processing.

Types#

InequalityConstraint#

Ensures column_a OP column_b on every row. Violations are fixed by nudging column_a.

Python
from misata.constraints import InequalityConstraint

c = InequalityConstraint("price", ">", "cost")
df = c.apply(df)

In YAML:

YAML
constraints:
  - name: profit_margin
    table: orders
    type: inequality
    column_a: amount
    operator: ">"
    column_b: cost

Operators: >, >=, <, <=

ColumnRangeConstraint#

Clips a column to [low_col, high_col] per row.

Python
from misata.constraints import ColumnRangeConstraint

c = ColumnRangeConstraint("price", low_col="min_price", high_col="max_price")
df = c.apply(df)

RatioConstraint#

Enforces a target proportion of a categorical value.

Python
from misata.constraints import RatioConstraint

c = RatioConstraint(column="plan", value="free", ratio=0.70)
df = c.apply(df)
# 70% of rows will have plan == "free"

UniqueConstraint#

Removes duplicate values (or duplicate composite keys).

Python
from misata.constraints import UniqueConstraint

c = UniqueConstraint(columns=["user_id", "date"])
df = c.apply(df)

SumConstraint#

Ensures values in a column sum to a target within a group.

Python
from misata.constraints import SumConstraint

c = SumConstraint(column="hours", group_by="employee_id", target=8.0)
df = c.apply(df)
# Total hours per employee == 8.0

NotNullConstraint#

Replaces nulls with a fallback value.

Python
from misata.constraints import NotNullConstraint

c = NotNullConstraint(column="email", fallback="unknown@example.com")
df = c.apply(df)

Applying multiple constraints#

Python
from misata.constraints import ConstraintEngine

engine = ConstraintEngine([
    InequalityConstraint("price", ">", "cost"),
    RatioConstraint("status", "active", 0.85),
    UniqueConstraint(["order_id"]),
])

df = engine.apply(df)
Last updated Edit this page.md