Extending

Normalizers, metrics, dataset formats and export formats are plug-ins chosen by name from a Registry. A plug-in registered in your own package is available in the configuration, the CLI and the GUI — whose parameter form is generated from the plug-in's parameter type.

A metric

import numpy as np
from pydantic import Field
from pydantic.dataclasses import dataclass

from erdr.core.metrics import METRICS, round_distances, traits
from erdr.core.registry import PARAMS_CONFIG


@dataclass(frozen=True, config=PARAMS_CONFIG)
class LorentzianParams:
    scale: float = Field(default=1.0, gt=0.0)


@METRICS.register(
    "lorentzian", summary="Σ ln(1 + |x′ − y′| / s) on I", params_type=LorentzianParams
)
@traits(domain="quantitative")
def lorentzian(A, B, quantitative, decimals, params=None, /):
    s = (params or LorentzianParams()).scale
    total = np.zeros((A.shape[0], B.shape[0]))
    for j in range(A.shape[1]):
        total += np.log1p(np.abs(A[:, j, None] - B[None, :, j]) / s)
    return round_distances(total, decimals)

A metric receives the unified values of the selected features and returns the rounded distance matrix. traits declares its domain (any, quantitative, nominal), an optional fit step (statistics over E₀, e.g. the covariance of mahalanobis) and an optional check of its parameters against the selected features. No metric reads a class label.

A normalizer

A normalizer (X, quantitative, params) -> Scaling estimates its constants on E₀; see erdr.core.normalizers (minmax, z-score, …) for the pattern. Register it with @NORMALIZERS.register("name", summary=…, params_type=…).

A dataset format or an export format

from erdr.data.loaders import LOADERS  # (path, LoadOptions) -> Dataset
from erdr.export import EXPORTERS  # (run, folder, ExportOptions) -> list[Path]

From another package

Entry points make the plug-ins available without importing anything:

[project.entry-points."erdr.metrics"]
lorentzian = "my_package.metrics:lorentzian"

Groups: erdr.metrics, erdr.normalizers, erdr.loaders, erdr.exporters (Registry.load_entry_points()).

A dataset

Drop <Name> (m, n, c).dat and .csv in the extended layout into a family folder of the type folder that fits its features — datasets/Quantitative/, Nominal/ or Heterogeneous/; erdr data check verifies it, including the type folder and the author's file conventions (erdr.data.write_extended writes them). The metric follows its type (Datasets). To build a small dataset into the package, copy its .csv to src/erdr/data/datasets/ and add it to erdr.data.builtin with its type.

A reading

The readings A, B, C and the scope of z are typed switches (erdr.core.readings); a new open point of the text would become a new field of Readings and ReadingsConfig, a branch in erdr.core.algorithm.run_erdr, a line in reading_notes (so it is flagged) and a column of the readings table — and must be agreed with the author first (it changes numerical results).