ML Python 300 - Impute Numeric with Multiple Methods

KNIME Python Script: Train and Save Multiple Iterative Imputation Models for Missing Data Handling for Numeric values

Short Summary

This script, designed for use inside a KNIME Python node, prepares tabular data for imputation by:

1. Loading input data from KNIME and extracting workflow variables such as the model path, excluded columns (like IDs), and the target label.

2. Separating features into numerical and categorical columns, while storing metadata about excluded, label, numeric, categorical, and remaining columns.

3. Initializing several machine learning regressors (ARDRegression, AdaBoost, Decision Trees, Extra Trees, KNN) as estimators for the IterativeImputer from scikit-learn.

4. Training an imputer model for each estimator on the numeric features, then saving the trained imputers as compressed .pkl files (with LZMA compression) in the given path.

5. Returning a dictionary of column classifications (excluded, label, numeric, categorical, rest) as the KNIME output object for downstream use.

👉 In essence, it creates a library of imputation models to handle missing values using different algorithms and saves them for later application.

MEDIUM BLOG

Data preparation for Machine Learning with KNIME and the Python “vtreat” package

https://medium.com/lp/efcaf58fa783

Learn the imputation models (numeric values only)

The results will be stored in the ../model/ folder. You can apply them in the lower section of the workflow to new /unseen data.

Download the whole workflow group "Missing Value Imputation":

https://hub.knime.com/s/smId4VcHVVVL5D_t

Apply the numeric imputation models learned in the code above