Target encoding in ML Python - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to calculate the mean target value for each category.

ML Python

mean_target = df.groupby('category')['target'].[1]()

Drag options to blanks, or click blank then click option'

Asum

Bmax

Cmean

Dcount

Attempts:

3 left

💡 Hint

Common Mistakes

Using sum instead of mean

Using count which counts entries, not averages

The mean() function calculates the average target value for each category, which is the basis of target encoding.

2fill in blank

medium

Complete the code to map the target mean to the original dataframe's category column.

ML Python

df['category_encoded'] = df['category'].[1](mean_target)

Drag options to blanks, or click blank then click option'

Amap

Bapply

Creplace

Dfillna

Attempts:

3 left

💡 Hint

Common Mistakes

Using apply which is slower and less direct

Using replace which is for replacing values but not with a mapping dictionary

The map() function replaces each category with its corresponding mean target value from mean_target.

3fill in blank

hard

Fix the error in the code to avoid data leakage by computing target encoding only on the training set.

ML Python

mean_target_train = train_df.groupby('category')['target'].[1]()
train_df['category_encoded'] = train_df['category'].map(mean_target_train)
test_df['category_encoded'] = test_df['category'].map(mean_target_train).fillna([2])

Drag options to blanks, or click blank then click option'

Amean

Bsum

Ccount

D0

Attempts:

3 left

💡 Hint

Common Mistakes

Calculating mean on the whole dataset causing leakage

Not filling missing values in test data causing NaNs

We use mean to calculate the average target in the training set. For categories in the test set not seen in training, we fill missing values with 0 to avoid errors.

4fill in blank

hard

Fill both blanks to create a target encoding function that fits on training data and transforms new data.

ML Python

def target_encode(train, test, cat_col, target_col):
    mean_target = train.groupby([1])[[2]].mean()
    train_encoded = train[cat_col].map(mean_target)
    test_encoded = test[cat_col].map(mean_target).fillna(mean_target.mean())
    return train_encoded, test_encoded

Drag options to blanks, or click blank then click option'

A'category'

B'target'

C'category_encoded'

D'target_encoded'

Attempts:

3 left

💡 Hint

Common Mistakes

Using encoded column names instead of original columns

Mixing up target and category column names

The function groups training data by the category column and calculates the mean of the target column. These are the inputs to groupby and the column to average.

5fill in blank

hard

Fill all three blanks to apply target encoding with smoothing to reduce noise for categories with few samples.

ML Python

def smooth_target_encode(train, test, cat_col, target_col, m=5):
    mean = train[target_col].mean()
    agg = train.groupby([1])[[2]].agg(['mean', 'count'])
    smooth = (agg['count'] * agg['mean'] + m * mean) / (agg['count'] + m)
    train_encoded = train[cat_col].map(smooth)
    test_encoded = test[cat_col].map(smooth).fillna(mean)
    return train_encoded, test_encoded

Drag options to blanks, or click blank then click option'

A'category'

B'target'

Attempts:

3 left

💡 Hint

Common Mistakes

Swapping category and target columns in groupby or agg

Not filling missing values in test data

The groupby is done on the category column ('category'), and the aggregation is on the target column ('target'). The smoothing formula uses counts and means to balance category averages with the global mean.