KNN imputation of categorical values

Once all the categorical columns in the DataFrame have been converted to ordinal values, the DataFrame is ready to be imputed. Imputing using statistical models like K-Nearest Neighbors (KNN) provides better imputations.

In this exercise, you'll

Use the KNN() function from fancyimpute to impute the missing values in the ordinally encoded DataFrame users.
Convert the ordinal values back to their respective categories using the ordinal encoder's .inverse_transform() method.

Remember, ordinal_enc_dict stores sklearn's OrdinalEncoder() for each column. The users DataFrame stores the encoded values (ordinal values) for each column.

The KNN() function, the dictionary of OrdinalEncoder()s ordinal_enc_dict and the users DataFrame have already been loaded for you.

This exercise is part of the course

Dealing with Missing Data in Python

View Course

Exercise instructions

Impute the users DataFrame using KNN_imputer's fit_transform() method. These transformed values are rounded to get integers.
Iterate over columns in users.
Select the column's OrdinalEncoder() from ordinal_enc_dict and perform .inverse_transform() on the reshaped array reshaped.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Create KNN imputer
KNN_imputer = KNN()

# Impute 'users' DataFrame. It is rounded to get integer values
users_KNN_imputed.iloc[:, :] = np.round(___)

# Loop over the column names in 'users'
for col_name in ___:
    
    # Reshape the column data
    reshaped = users_KNN_imputed[col_name].values.reshape(-1, 1)
    
    # Select the column's Encoder and perform inverse transform on 'reshaped'
    users_KNN_imputed[col_name] = ___

Edit and Run Code