[DOC]: What does imputation_type="simple" do?

[jameshfisher]

pycaret version checks

• [X] I have checked that the issue still exists on the latest versions of the docs here

Location of the documentation

I'm reading the regression API docs here, which say

imputation_type: str or None, default = ‘simple’

The type of imputation to use. Can be either ‘simple’ or ‘iterative’. If None, no imputation of missing values is performed.

Documentation problem

What does imputation_type="simple", the default, actually do?

I found a good description of imputation_type="iterative" in this unofficial article. I can also see that you can set the iterative imputer, which defaults to lightgbm.

But there is no description of what simple means or does.

Suggested fix for documentation

Describe this in the API docs somewhere, or link to something that describes it.

[jameshfisher]

I'm looking at https://github.com/pycaret/pycaret/blob/ff1a8c905ad722d9de4fd8edfe793794a25775c1/pycaret/internal/preprocess/preprocessor.py#L414-L466

Which suggests imputation_type="simple" means:

• For numeric features, impute with the mean. (More precisely: impute with a constant value chosen by the method given in the numeric_imputation param.)
• For categorical features, impute with the mode. (More precisely: impute with a constant value chosen by the method given in the categorical_imputation param.)

Do I have this right?

[Yard1]

Not exactly. Simple imputation consists of filling the values with mean/median for numeric features and most frequent value/constant for categorical values. See https://scikit-learn.org/stable/modules/generated/sklearn.impute.SimpleImputer.html

Iterative imputation starts with simple imputation and then fits ML models on each column with missing values, using the values of other columns to predict what should be imputed. This is done several times, thus iterative imputation. Hope that helps - please close the issue if you do not have any more questions.

[jameshfisher]

That's helpful, thank you!