Remove unused feature type inference to stick with one way of doing it

[eroell]

Add binary_as parameter to infer_feature_types

Summary

• Feature: Add a binary_as parameter to ed.infer_feature_types that controls whether binary (0/1) features are classified as categorical or numeric.
• Modified: ed.infer_feature_types considers now integer series 0, ... , n as numeric, and not as categorical
• Removed: A private function _infer_numerical_column_indices has been removed which has previously been used in some imputation functions from ehrapy (Remove _infer_numerical_column_indices ehrapy#1014)

Fixes #214

Motivation

Binary 0/1 variables are ambiguous: they can represent true categorical labels (e.g., male/female) or numeric indicators (e.g., presence/absence of a diagnosis used in downstream arithmetic). Previously, infer_feature_types always classified them as categorical. This made it cumbersome when a dataset has many binary indicator columns that should be treated numerically — users had to manually correct each one with ed.replace_feature_types.

Change

infer_feature_types (and the internal _detect_feature_type) now accept a keyword argument:

binary_as: Literal["categorical", "numeric"] = "categorical"

• "categorical" (default): binary 0/1 features are classified as categorical — no change in behavior.
• "numeric": binary 0/1 features are classified as numeric instead.

Usage

import ehrdata as ed

edata = ed.dt.mimic_2()

# Default — binary features are categorical (existing behavior)
ed.infer_feature_types(edata)

# Opt in — treat binary 0/1 features as numeric
ed.infer_feature_types(edata, binary_as="numeric")

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB