Representativeness Metric

The Representativeness metric compares the distribution of your data to a target reference distribution (such as Normal or Uniform). It ensures that the dataset conforms to a given specification or requirement.

What It Measures

Representativeness checks if your data follows a known distribution. Use it to:

• Validate synthetic data — Ensure generated data matches expected patterns
• Check for data drift — Detect changes in data distribution over time
• Ensure balanced datasets — Verify class distributions in classification tasks

Available Statistical Tests

Test	What It Does	Best For
Chi-Square (χ²)	Goodness-of-fit test	Categorical/binned data
Kolmogorov-Smirnov (KS)	Non-parametric test	Continuous distributions
Shannon Entropy	Measures information diversity	General diversity
GRTE	Granular Relative Theoretical Entropy	Complex distributions (Confiance.ai)

Use Cases

• Validate that training data matches expected distributions
• Detect distribution shifts in production data
• Check if data augmentation preserved original distribution
• Verify synthetic data quality

Processor Information

• Class: RepresentativenessProcessor
• Package: dqm-ml-core
• Type Name: representativeness

Configuration Parameters

• input_columns: Numeric columns to analyze.
• distribution: Reference distribution name (normal, uniform).
• metrics: List of statistical tests to run (chi-square, kolmogorov-smirnov, shannon-entropy, grte).
• bins: Number of bins for histogram-based tests (default: 10).
• distribution_params: (Optional) Parameters for the reference distribution (e.g., mean, std for normal).

Example YAML Configuration

metrics_processor:
  dist_check:
    type: representativeness
    input_columns: ["feature_x"]
    distribution: "normal"
    metrics: ["chi-square", "kolmogorov-smirnov"]
    bins: 30
    distribution_params:
      mean: 0.0
      std: 1.0

Output

The processor returns a dictionary with scores for each requested metric and column:

• <column_name>_<metric_name>_score: The statistical test score.
• <column_name>_<metric_name>_pvalue: The p-value associated with the test.