Join Our 5-Week ML/AI Engineer Interview Bootcamp 🚀 led by ML Tech Leads at FAANGs
Back to Questions
116. Vector normalization
easy
Generalsenior
Vector normalization makes embedding-based retrieval more reliable by ensuring similarity depends on direction, not magnitude. In this task, you’ll normalize a batch of embedding vectors so you can safely use cosine similarity later.
Requirements
Implement the function
python
Rules:
- Normalize each vector independently (each row gets its own L2 norm).
- Return a NumPy array.
- Do not use any prebuilt normalization utilities (e.g.,
sklearn.preprocessing.normalize). - Keep the output in the same order as the input.
- Handle the zero-vector case safely (e.g., avoid division by zero).
Example
python
Output:
python
Input Signature
| Argument | Type |
|---|---|
| embeddings | np.ndarray |
Output Signature
| Return Name | Type |
|---|---|
| value | np.ndarray |
Constraints
Return NumPy array.
Compute row-wise L2 norms (axis=1).
No sklearn/normalization utilities allowed.
Hint 1
Compute a per-row L2 norm: norms = np.sqrt(np.sum(embeddings**2, axis=1, keepdims=True)).
Hint 2
Normalize with broadcasting: embeddings / norms. Use keepdims=True in summation to ensure shapes match.
Hint 3
Handle division by zero (if any norm is 0) by setting that norm to 1 or using np.maximum.
Roles
ML Engineer
AI Engineer
Companies
GeneralLevels
senior
entry
Tags
numpy
l2-normalization
cosine-similarity
vectorization
13 people are solving this problem
Python EditorLn 1, Col 1