← Virtu Interview Insights

Virtu·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
Jun 2026

Summary

Virtu data scientist interview focused on a pretty involved ML system design problem around basketball shot prediction. One question but it had a lot of moving parts, and there was a tricky feature selection sub-question tucked in at the end that I almost glossed over.

Questions Asked (1)

Q1

Design an ML system to predict whether a basketball shot will go in or not, using only information available at the moment of release. Walk through problem formulation, data collection, features, modeling, loss function, evaluation, and feature selection. And specifically: if a feature fails a univariate significance test against the target, is it safe to drop it?

System DesignTechnical Trade-offsData Modeling
Author's notes

The main design part I felt okay about.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer as a complete ML system design, starting with problem formulation (binary classification with strict temporal constraints), then data collection, feature engineering, modeling, loss, evaluation, and feature selection. Explicitly address the univariate significance test question by explaining why it's unsafe to drop a feature solely based on that test, citing interaction effects and multivariate importance.

Pro tip: Emphasize that the 'moment of release' constraint means you must avoid any post-release data leakage, and that univariate tests can miss features that are only predictive in combination with others—use multivariate feature importance and ablation studies instead.

1. Problem Formulation

Define the task as binary classification (shot made/missed) with a strict temporal constraint: only use features available at or before the moment of release. Clarify the prediction point and the target variable.

2. Data Collection & Features

Gather tracking data (player pose, ball trajectory, shot location, defender distance, game context) from sources like SportVU. Engineer features such as release angle, velocity, player height, distance to hoop, and defensive pressure.

3. Modeling & Loss Function

Choose a model (e.g., gradient boosting or logistic regression) and use binary cross-entropy loss. Consider class imbalance and calibration if probabilities are needed.

4. Evaluation & Feature Selection

Evaluate with AUC, log loss, and calibration. For feature selection, use multivariate methods (e.g., permutation importance, SHAP) and ablation studies, not just univariate tests.

5. Address the Univariate Test Question

Explain that a feature failing a univariate significance test is not safe to drop because it may have interaction effects or be important in the presence of other features. Use multivariate analysis to decide.

Key Points to Mention

  • Temporal constraint: only pre-release information to avoid leakage
  • Feature engineering: release angle, velocity, player pose, defender distance, shot distance
  • Model choice: gradient boosting (XGBoost/LightGBM) or logistic regression with regularization
  • Loss function: binary cross-entropy, with potential for focal loss if imbalanced
  • Evaluation metrics: AUC, log loss, calibration, and business-relevant metrics
  • Univariate tests are insufficient for feature selection due to interactions; use multivariate importance and ablation

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