Amazon · ML & AI Fundamentals

Apply Double ML with text-address features

TrueInterview

October 7, 2026 · 1 min read

Use Double Machine Learning (DML) to estimate the average treatment effect (ATE) of sending a first CSAT reminder, with address text included as an input. Address each of the following:

Define the outcome $Y$ , the treatment $T$ , and the feature set $X$ , and explain how you represent addresses (for example, geocoding plus neighborhood attributes, with an address text embedding).
State the orthogonalized moment you will use and how you will carry out sample splitting and K-fold cross-fitting.
Choose nuisance learners for $E[Y|X]$ and $E[T|X]$ (such as gradient boosting for $Y$ and calibrated logistic regression for $T$ ), set their hyperparameters, and explain how you avoid leakage from post-treatment variables.
Describe overlap/positivity diagnostics and how you would trim or reweight cases.
Explain how you would test sensitivity to unobserved confounding (such as Oster's $\delta$ or partial $R^2$ ) and how you would report subgroup effects for device and channel while controlling the false discovery rate.
Describe how you would validate the text features (ablation tests and SHAP consistency across folds) and how you would limit geographic privacy and fairness risks (for example, by excluding protected proxies and using coarse geohashes).

This item tests a candidate's skill in causal inference and Double Machine Learning for estimating average treatment effects with observational data, covering how address-derived text features are represented and validated, nuisance estimation, overlap diagnostics, sensitivity analysis, subgroup reporting, and geographic privacy and fairness concerns; it falls under Machine Learning and applied causal inference. It is typically used to assess both conceptual knowledge of orthogonalization, sample splitting, and identification assumptions, and practical ability in selecting and validating text/geospatial features, checking overlap/positivity, running sensitivity analyses, and controlling for multiple comparisons—a mix of conceptual and applied skill.

Amazon · ML & AI Fundamentals

Apply Double ML with text-address features

TrueInterview

October 7, 2026 · 1 min read

Use Double Machine Learning (DML) to estimate the average treatment effect (ATE) of sending a first CSAT reminder, with address text included as an input. Address each of the following:

Define the outcome $Y$ , the treatment $T$ , and the feature set $X$ , and explain how you represent addresses (for example, geocoding plus neighborhood attributes, with an address text embedding).
State the orthogonalized moment you will use and how you will carry out sample splitting and K-fold cross-fitting.
Choose nuisance learners for $E[Y|X]$ and $E[T|X]$ (such as gradient boosting for $Y$ and calibrated logistic regression for $T$ ), set their hyperparameters, and explain how you avoid leakage from post-treatment variables.
Describe overlap/positivity diagnostics and how you would trim or reweight cases.
Explain how you would test sensitivity to unobserved confounding (such as Oster's $\delta$ or partial $R^2$ ) and how you would report subgroup effects for device and channel while controlling the false discovery rate.
Describe how you would validate the text features (ablation tests and SHAP consistency across folds) and how you would limit geographic privacy and fairness risks (for example, by excluding protected proxies and using coarse geohashes).