Duplicated

duplicated.py injects duplicated records into a dataset. These utilities simulate real-world issues such as data imbalance, bias from oversampling, and noise from repeated observations.

Function Signature

from pucktrick.duplicated import duplicated

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

Strategy Parameters

Set "function" at the top level of the strategy to apply a text transformation to the duplicated rows:

Value	Description
"shuffle_words"	Shuffles the words in string fields
"abbreviate_text"	Abbreviates text values
"replace_punctuation"	Replaces punctuation characters
"remove_replace"	Removes or replaces characters
"upper_lower"	Randomly changes case

If "function" is omitted, rows are duplicated without modification.

Example

from pucktrick.duplicated import duplicated

strategy = {
    "affected_features": ["name", "description"],
    "selection_criteria": "all",
    "percentage": 0.10,
    "mode": "new",
    "function": "shuffle_words",
    "perturbate_data": {"sampling": "random", "distribution": "random"}
}

err, df_corrupted = duplicated(df, strategy)

Modes

Mode	Behaviour
new	Duplicates rows from the clean dataset up to the specified percentage.
extended	Adds more duplicated rows to a dataset that may already contain repeats, reaching the cumulative target. Requires original_df.
composed	Duplicates only rows already modified by a previous operator. Requires original_df.