covnorm 0.2.1

Robust conditional normalization of continuous markers using polynomial surface fitting over categorical and continuous covariates

covnorm

Scikit-learn compatible transformer for robust conditional Z-score normalization of a continuous marker against categorical and continuous covariates.

For each unique combination of categorical covariates, the transformer fits a polynomial surface over mu and sigma estimated per bin via Q-Q regression on Box-Cox transformed values, with iterative Tukey z-score outlier rejection. At transform time, each sample is normalized using the mu and sigma predicted from its covariate values.

Data requirement: target values must be strictly positive (Box-Cox transform). Shift your data if it contains zeros or negatives.

Supports up to 2 categorical and 2 continuous covariates.

Installation

pip install covnorm

Usage

import numpy as np
from covnorm import RobustConditionalNormalizer

# X columns: [sex, batch, age, cell_count, marker]
# sex=0, batch=1 → categorical
# age=2, cell_count=3 → continuous
# marker=4 → target to normalize

X = np.load("data.npy")  # shape (n_samples, 5)

normalizer = RobustConditionalNormalizer(
    categorical_cols=[0, 1],
    continuous_cols=[2, 3],
    target_col=4,
    n_bins=6,                      # bins per continuous covariate
    log_transform_continuous=True, # recommended when covariates span orders of magnitude
)

X_norm = normalizer.fit_transform(X)

The target column in X_norm contains Z-scores. All other columns are unchanged.

It follows the scikit-learn fit / transform / fit_transform API and is compatible with Pipeline.

Parameters

Parameter	Default	Description
categorical_cols	—	Column indices treated as categorical grouping variables
continuous_cols	—	Column indices used as continuous covariates for surface fitting
target_col	—	Column index of the marker to normalize
n_bins	6	Number of bins per continuous covariate axis
degree	3	Polynomial degree of the mu/sigma surface
n_iterations	3	Iterative Tukey z-score outlier-removal passes before binning
log_transform_continuous	False	Apply log10 to continuous covariates before fitting (recommended when they span orders of magnitude, e.g. age in years)