Critique and Test Python Preprocessing Utilities Effectively

Scenario

Code review of Python preprocessing utilities (OutlierHandler and three Imputer classes)

Question

Summarize, at a high level, what the OutlierHandler class accomplishes. Why is keeping fit and transform as separate methods advantageous? Identify and critique any coding-style or import problems found in the three Imputer classes. Hand-write one critical unit test you would add for OutlierHandler.

Hints

Think scikit-learn patterns, PEP-8, avoiding "from numpy import *", assert expected vs. actual.

Implement a Fit/Transform Outlier Handler

In many preprocessing pipelines, a utility first learns statistics from training data (fit) and later applies them to any dataset (transform). Implement a simplified OutlierHandler.

Your function receives two lists:

• train: the data used to learn outlier thresholds
• values: the data to transform

Rules:

1. Ignore None values in train.
2. Sort the remaining values and compute quartiles using this definition:
   • If the number of values is odd, exclude the overall median when splitting into lower and upper halves.
   • Q1 is the median of the lower half.
   • Q3 is the median of the upper half.
   • The median of an even-length list is the average of the two middle values.
3. Compute IQR = Q3 - Q1.
4. Compute bounds:
   • lower = Q1 - 1.5 * IQR
   • upper = Q3 + 1.5 * IQR
5. Transform each element in values:
   • keep None as None
   • if a number is below lower, replace it with lower
   • if a number is above upper, replace it with upper
   • otherwise leave it unchanged

If train contains fewer than 2 non-None values, return values unchanged.

Return the transformed list.

Constraints

• 0 <= len(train), len(values) <= 100000
• Each element is either an integer, a float, or None
• Quartiles must be computed from train only, never from values

Examples

Example 1

Input: ([10, 12, 13, 14, 15, 16, 18, 19, 100], [5, 15, 30])
Output: [5, 15, 27.5]

Explanation: From the training data, Q1 = 12.5 and Q3 = 18.5, so IQR = 6. The bounds are 3.5 and 27.5. Only 30 is above the upper bound, so it is clipped to 27.5.

Example 2

Input: ([None, 1, 2, 2, 3, 4, 100], [None, -5, 8, 4])
Output: [None, -1.0, 7.0, 4]

Explanation: Ignoring None, the sorted training data is [1, 2, 2, 3, 4, 100]. Q1 = 2, Q3 = 4, so IQR = 2 and the bounds are -1 and 7. None stays None, -5 becomes -1, 8 becomes 7, and 4 stays unchanged.

Hints

1. Write a small helper function to compute the median of a sorted list, then reuse it for Q1 and Q3.
2. This is a fit/transform pattern: learn the clipping bounds from train once, then apply them to every element in values.