Guides

Custom Rules Guide

Learn how to write custom deterministic data quality validation rules

Featuresmith is fully extensible. You can easily add your own deterministic rules by extending the BaseRule interface and defining rule parameters.

The BaseRule Interface

All rules must extend BaseRule and implement the following properties and methods:

id: A unique dotted identifier (e.g., statistical.zero_variance).
name: A descriptive rule title.
category: Group category (e.g., quality, statistical, leakage).
severity: Finding severity level (e.g., RuleSeverity.INFO, WARNING, or CRITICAL).
evaluate(profile: ProfileResult) -> list[RuleFinding]: Core audit logic.

Custom Rule Implementation

Here is a complete example of a rule to detect numerical columns that have zero standard deviation (constant value columns):

python

1from typing import Any
2from featuresmith.rules.base import BaseRule
3from featuresmith.core.rule_finding import RuleFinding, RuleSeverity
4from featuresmith.core.profile_result import ProfileResult
5
6class ZeroVarianceRule(BaseRule):
7    """Custom rule to detect columns with zero variance."""
8
9    def __init__(self, **kwargs: Any) -> None:
10        super().__init__()
11        self.enabled = kwargs.get("enabled", True)
12
13    @property
14    def id(self) -> str:
15        return "statistical.zero_variance"
16
17    @property
18    def name(self) -> str:
19        return "Zero Variance Columns"
20
21    @property
22    def category(self) -> str:
23        return "statistical"
24
25    @property
26    def severity(self) -> RuleSeverity:
27        return RuleSeverity.WARNING
28
29    def evaluate(self, profile: ProfileResult) -> list[RuleFinding]:
30        findings = []
31        for col_name, col_profile in profile.column_profiles.items():
32            numeric_stats = col_profile.numeric_stats
33            if numeric_stats is not None and numeric_stats.std == 0.0:
34                findings.append(
35                    self.create_finding(
36                        column_name=col_name,
37                        title="Zero Variance Detected",
38                        description=f"Column '{col_name}' has standard deviation of 0.0 (no variance).",
39                        evidence={"std": 0.0}
40                    )
41                )
42        return findings

Evaluating Custom Rules

You can instantiate your rule and run it directly in Python:

python

1import featuresmith as fs
2
3dataset = fs.load("data.csv")
4rule = ZeroVarianceRule()
5
6# Run rule directly against computed statistical profiles
7profile = fs.profile(dataset)
8findings = rule.evaluate(profile)
9
10for finding in findings:
11    print(f"[{finding.severity}] {finding.title} in {finding.column_name}")

1from typing import Any 2from featuresmith.rules.base import BaseRule 3from featuresmith.core.rule_finding import RuleFinding, RuleSeverity 4from featuresmith.core.profile_result import ProfileResult 5 6class ZeroVarianceRule(BaseRule): 7 """Custom rule to detect columns with zero variance.""" 8 9 def __init__(self, **kwargs: Any) -> None: 10 super().__init__() 11 self.enabled = kwargs.get("enabled", True) 12 13 @property 14 def id(self) -> str: 15 return "statistical.zero_variance" 16 17 @property 18 def name(self) -> str: 19 return "Zero Variance Columns" 20 21 @property 22 def category(self) -> str: 23 return "statistical" 24 25 @property 26 def severity(self) -> RuleSeverity: 27 return RuleSeverity.WARNING 28 29 def evaluate(self, profile: ProfileResult) -> list[RuleFinding]: 30 findings = [] 31 for col_name, col_profile in profile.column_profiles.items(): 32 numeric_stats = col_profile.numeric_stats 33 if numeric_stats is not None and numeric_stats.std == 0.0: 34 findings.append( 35 self.create_finding( 36 column_name=col_name, 37 title="Zero Variance Detected", 38 description=f"Column '{col_name}' has standard deviation of 0.0 (no variance).", 39 evidence={"std": 0.0} 40 ) 41 ) 42 return findings

1import featuresmith as fs 2 3dataset = fs.load("data.csv") 4rule = ZeroVarianceRule() 5 6# Run rule directly against computed statistical profiles 7profile = fs.profile(dataset) 8findings = rule.evaluate(profile) 9 10for finding in findings: 11 print(f"[{finding.severity}] {finding.title} in {finding.column_name}")