scikit-normod

Scikit-normod is a scientific library designed for normative modeling, offering a familiar scikit-learn-like API. This package facilitates the fitting and validation of normative models, the estimation of centiles, the transformation of data to z-scores, and the calculation of summary atypicality scores.

Warning

The library is in its early development stages. Expect bugs, missing features, and breaking changes.

Warning

Really

Key Features

Models :columns: auto

• Gaussian Homoskedastic

  • Ordinary Least Squares (OLS)

  • Generalized Additive Models (GAM)

• Gaussian Heteroskedastic (Location-Scale):

  • Generalized Additive Models for Location Scale (GAMLS)

• Location-Scale-Shape

  • Generalized Additive Models for Location Scale Shape (GAMLSS)

• Meta-Regression

  • Gaussian Homoskedastic

  • Gaussian Heteroskedastic

Model Methods :columns: auto

• Estimate quantiles

• Transform data to z-scores

• Transform data to quantiles

• (not-implemented) Quantiles and z-scores with uncertainity

• (not-implemented) Transfer/recalibrate model to new sites

Anomaly Detection :columns: auto

• Various anomaly scores including mean-z, min/max-z, z-over-threshold, and others

Validation :columns: auto

• Whole Model

  • Logarithmic score

  • (not-implemented) AIC, BIC, GAIC

• Calibration

  • Mean, standard deviation, skewness, kurtosis, W

• (not-implemented) Diagnostic plots

• (untested, probably broken) scikit-learn grid search etc.

Ploting :columns: auto

• Plot centiles (should be better)

• (not-implemented) calibration plots:

  • worm-plots

  • bucket-plots

  • qq

  • pp

  • centiles/distribution

scikit-learn Compatibility :columns: auto

• Grid search, pipelines, CV scorers, etc.

• In theory should work

• In practice not-tested, probably broken, maybe eventually