Skip to main content
Version: 1.0.1

Missing

missing.py replaces values in specified columns with NaN. It is designed to support controlled experimentation with missing data in machine learning workflows.

Function Signature​

from pucktrick.missing import missing

error_code, modified_df = missing(df, strategy)
# or, for mode="extended" / mode="composed":
error_code, modified_df = missing(df, strategy, original_df=clean_df)

Strategy Parameters​

No module-specific parameters are required inside perturbate_data. Use the base strategy parameters to configure which rows and columns to corrupt.

Example​

from pucktrick.missing import missing

strategy = {
    "affected_features": ["age", "income"],
    "selection_criteria": "all",
    "percentage": 0.15,
    "mode": "new",
    "perturbate_data": {"sampling": "random", "distribution": "random"}
}

err, df_corrupted = missing(df, strategy)

Modes​

ModeBehaviour
newIntroduces NaN into clean columns up to the specified percentage.
extendedIncrementally adds NaN values to columns that may already contain missing data, reaching the cumulative percentage target. Requires original_df.
composedIntroduces NaN only into rows already modified by a previous operator. Requires original_df.