Join Our 5-Week ML/AI Engineer Interview Bootcamp 🚀 led by ML Tech Leads at FAANGs
Back to Questions
246. Cumulative sum features
easy
Generalsenior
Build cumulative-sum (running total) features, a simple but useful feature-engineering trick for time-ordered data. You’ll compute per-group running totals so each row can use information from earlier rows only.
Requirements
Implement the function
python
Rules:
- Treat
valuesas already sorted in time order (do not sort). - The cumulative sum should reset when the group id changes (i.e., compute cumsum separately per unique group).
- Return a NumPy array of floats with the same length as
values. - Don’t use pandas; use only NumPy and built-in Python libraries.
- Aim for an efficient approach (single pass is fine).
Example
python
Output:
python
Input Signature
| Argument | Type |
|---|---|
| groups | np.ndarray |
| values | np.ndarray |
Output Signature
| Return Name | Type |
|---|---|
| value | np.ndarray |
Constraints
No pandas; only Python/NumPy.
Input order is time order; do not sort.
Single pass O(n) expected.
Hint 1
Use a dictionary/map: group_id -> running_sum while scanning left-to-right.
Hint 2
For each index i, read previous sum via totals.get(group, 0.0), add values[i], store back, and write to output.
Hint 3
Single-pass pattern: for i, (v,g) in enumerate(zip(values, groups)): update totals[g] and set out[i] = totals[g].
Roles
ML Engineer
AI Engineer
Companies
GeneralLevels
senior
entry
Tags
prefix-sum
hashmap
feature-engineering
time-series
20 people are solving this problem
Python EditorLn 1, Col 1