Extending METIS¶

METIS is designed to be extensible. You can add new metrics and new generators without modifying core code.

Add a New Metric¶

Step 1 — Choose the base class¶

Family	Base class	Location
Fidelity (numeric)	`NumericColumnMetric`	`metis/infrastructure/metrics/fidelity/fidelity_base.py`
Fidelity (categorical)	`CategoricalColumnMetric`	same
Fidelity (num↔num)	`NumNumPairMetric`	same
Fidelity (num↔cat)	`NumCatPairMetric`	same
Fidelity (cat↔cat)	`CatCatPairMetric`	same
Privacy	`DatasetPrivacyMetric`	`metis/infrastructure/metrics/privacy/privacy_base.py`
Utility	Inherit from utility base	`metis/infrastructure/metrics/utility/`

Step 2 — Create the metric file¶

# metis/infrastructure/metrics/fidelity/marginal/tails/my_metric.py

import pandas as pd
from scipy.stats import some_test

from metis.infrastructure.metrics.registry import register
from ...fidelity_base import NumericColumnMetric


@register("fidelity.my_metric")
class MyMetric(NumericColumnMetric):
    """Brief description of what this metric measures."""

    name: str = "my_metric"
    is_distance: bool = True  # True if lower raw value = better quality

    def _compute_column(self, real_col: pd.Series, synth_col: pd.Series) -> float:
        """Compute metric for a single numeric column."""
        statistic, _ = some_test(real_col.values, synth_col.values)
        return float(statistic)

Note

The @register("fidelity.my_metric") decorator is all you need for auto-discovery.

Step 3 — Import in registry¶

Add the import in metis/infrastructure/metrics/registry.py:

def _register_default_metrics():
    # ... existing imports ...
    from .fidelity.marginal.tails.my_metric import MyMetric

Step 4 — Add to taxonomy¶

In metis/domain/taxonomy.py:

FIDELITY_METRICS = {
    "marginal": {
        "subcategories": {
            "tails": [
                "fidelity.ks",
                # ...
                "fidelity.my_metric",  # ← add here
            ],
        },
    },
}

Step 5 — Use in config¶

metrics:
  - "fidelity.my_metric"        # by exact ID
  - "fidelity.marginal.tails"   # or via shortcut (auto-included)

Add a New Generator¶

Step 1 — Create the generator class¶

# metis/sota_models/generators/my_generator.py

import pandas as pd
from .base import BaseGenerator


class MyGenerator(BaseGenerator):
    """Brief description."""

    def __init__(self, name="MyGenerator", my_param=100, random_state=None, **kwargs):
        super().__init__(name=name, **kwargs)
        self.my_param = my_param
        self.random_state = random_state

    def fit(self, real_data, categorical_columns=None, ordinal_columns=None,
            continuous_columns=None):
        # Learn distribution
        self._is_fitted = True

    def generate(self, n_samples: int) -> pd.DataFrame:
        if not self._is_fitted:
            raise RuntimeError(f"{self.name} must be fitted first")
        # Generate synthetic data
        return synth_df

Step 2 — Register¶

In metis/sota_models/generators/__init__.py:

from .registry import GeneratorRegistry
from .my_generator import MyGenerator

GeneratorRegistry.register("my_generator", MyGenerator)

Step 3 — Use in YAML¶

benchmark:
  generators:
    - name: "my_generator"
      params:
        my_param: 200
        random_state: 42