Qube·Data Scientist·Take-home Assignment·Intermediate

Intermediate

May 2026

Summary

Qube gave me a take-home for a Data Scientist role, basically a leaf classification task with three numeric features and an unlabeled test set to predict on. The whole thing had to be done in a Jupyter notebook, which was fine, but the KNN-from-scratch requirement caught me a bit off guard.

Questions Asked (1)

Given a labeled dataset with three numeric features and a three-class target, implement a K-nearest neighbors classifier from scratch (no ML libraries for the prediction logic), tune the value of k, and use the final model to generate predictions on an unlabeled dataset.

Algorithms & Data StructuresTechnical Trade-offs

Author's notes

The 'from scratch' part is where I spent most of my time.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by outlining the KNN algorithm and the need for feature scaling, then describe how to implement distance calculation and majority voting from scratch. Explain how to tune k using cross-validation on the labeled data, and finally apply the tuned model to the unlabeled dataset.

Pro tip: Mention that you would standardize features before computing distances, as KNN is sensitive to scale, and discuss the trade-off between bias and variance when choosing k.

1. Data Preparation

Load the labeled dataset, separate features and target, and apply feature scaling (e.g., standardization) to ensure all features contribute equally to distance calculations.

2. Implement KNN from Scratch

Write a function to compute Euclidean distance between a test point and all training points, then select the k nearest neighbors and assign the majority class as the prediction.

3. Tune k with Cross-Validation

Use k-fold cross-validation on the labeled data to evaluate different values of k (e.g., 1 to 20) and select the one with the best average validation accuracy.

4. Train Final Model and Predict

Train the KNN model on the full labeled dataset using the optimal k, then generate predictions for the unlabeled dataset.

5. Evaluate and Discuss Trade-offs

If labels for the unlabeled data are available, evaluate performance; otherwise, discuss the impact of k on bias-variance and computational complexity.

Key Points to Mention

Feature scaling (standardization or normalization) is crucial for KNN due to its distance-based nature.
Choice of distance metric (e.g., Euclidean, Manhattan) and its implications.
Cross-validation for hyperparameter tuning to avoid overfitting.
Bias-variance trade-off: small k leads to low bias but high variance, large k leads to high bias but low variance.
Computational complexity: prediction time is O(n*d) for each test point, which can be expensive for large datasets.
Handling ties in majority voting (e.g., by choosing the class with the smallest average distance or random selection).

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.