Uber Data Scientist Interview Questions and Answers 2026

WITH cohort AS (
    SELECT
        rider_id,
        DATE_TRUNC('week', MIN(request_ts))  AS cohort_week
    FROM trips
    WHERE status = 'completed'
    GROUP BY rider_id
    HAVING DATE_TRUNC('week', MIN(request_ts))
           = DATE_TRUNC('week', '2026-01-01'::date)   -- Week 1 of target month
),
weeks AS (
    SELECT generate_series(1, 4) AS week_num
),
activity AS (
    SELECT
        t.rider_id,
        (DATE_TRUNC('week', t.request_ts) - c.cohort_week) / 7  AS weeks_since_first
    FROM trips t
    JOIN cohort c USING (rider_id)
    WHERE t.status = 'completed'
      AND t.request_ts >= c.cohort_week + INTERVAL '7 days'
    GROUP BY t.rider_id, c.cohort_week, DATE_TRUNC('week', t.request_ts)
)
SELECT
    w.week_num,
    COUNT(DISTINCT a.rider_id)                                AS retained_riders,
    COUNT(DISTINCT c.rider_id)                                AS cohort_size,
    ROUND(COUNT(DISTINCT a.rider_id)::numeric
          / COUNT(DISTINCT c.rider_id) * 100, 1)             AS retention_pct
FROM weeks w
CROSS JOIN cohort c
LEFT JOIN activity a
    ON a.rider_id = c.rider_id
   AND a.weeks_since_first = w.week_num
GROUP BY w.week_num
ORDER BY w.week_num;

WITH status_sessions AS (
    SELECT
        driver_id,
        city,
        status,
        event_ts                                                      AS start_ts,
        LEAD(event_ts) OVER (PARTITION BY driver_id ORDER BY event_ts) AS end_ts
    FROM driver_status_log
),
online_hours AS (
    SELECT
        city,
        DATE_TRUNC('week', start_ts)                                   AS week_start,
        SUM(EXTRACT(EPOCH FROM (end_ts - start_ts)) / 3600)            AS total_online_hrs
    FROM status_sessions
    WHERE status IN ('available', 'on_trip')
      AND end_ts IS NOT NULL
    GROUP BY city, DATE_TRUNC('week', start_ts)
),
earnings AS (
    SELECT
        city,
        DATE_TRUNC('week', end_ts)  AS week_start,
        SUM(driver_earnings_usd)    AS total_earnings
    FROM trips
    WHERE status = 'completed'
    GROUP BY city, DATE_TRUNC('week', end_ts)
),
weekly AS (
    SELECT
        e.city,
        e.week_start,
        e.total_earnings / NULLIF(o.total_online_hrs, 0) AS earnings_per_hr
    FROM earnings e
    JOIN online_hours o USING (city, week_start)
)
SELECT
    curr.city,
    curr.earnings_per_hr                                        AS curr_week_eph,
    prev.earnings_per_hr                                        AS prev_week_eph,
    ROUND((curr.earnings_per_hr - prev.earnings_per_hr)
          / NULLIF(prev.earnings_per_hr, 0) * 100, 1)          AS pct_change,
    CASE WHEN curr.earnings_per_hr < prev.earnings_per_hr * 0.9
         THEN 'FLAG' ELSE 'OK' END                             AS status
FROM weekly curr
JOIN weekly prev
    ON curr.city = prev.city
   AND curr.week_start = prev.week_start + INTERVAL '7 days'
WHERE curr.week_start = DATE_TRUNC('week', CURRENT_DATE - INTERVAL '7 days')
ORDER BY pct_change;

WITH peak_requests AS (
    SELECT
        r.request_id,
        r.zone_id,
        r.request_ts::date                  AS req_date,
        CASE
            WHEN EXTRACT(HOUR FROM r.request_ts) BETWEEN 7 AND 8  THEN 'morning'
            WHEN EXTRACT(HOUR FROM r.request_ts) BETWEEN 17 AND 19 THEN 'evening'
        END                                 AS peak_slot,
        m.match_ts
    FROM ride_requests r
    LEFT JOIN trip_matches m
        ON m.request_id = r.request_id
       AND m.match_ts - r.request_ts <= INTERVAL '5 minutes'
    WHERE r.request_ts >= CURRENT_DATE - INTERVAL '30 days'
      AND EXTRACT(DOW FROM r.request_ts) BETWEEN 1 AND 5   -- Mon–Fri
      AND (
          EXTRACT(HOUR FROM r.request_ts) BETWEEN 7 AND 8
       OR EXTRACT(HOUR FROM r.request_ts) BETWEEN 17 AND 19
      )
),
daily_zone AS (
    SELECT
        zone_id,
        req_date,
        peak_slot,
        COUNT(*)                                                  AS total_requests,
        SUM(CASE WHEN match_ts IS NULL THEN 1 ELSE 0 END)        AS unfulfilled,
        ROUND(SUM(CASE WHEN match_ts IS NULL THEN 1.0 ELSE 0 END)
              / COUNT(*) * 100, 1)                               AS unfulfilled_rate
    FROM peak_requests
    GROUP BY zone_id, req_date, peak_slot
)
SELECT
    zone_id,
    peak_slot,
    COUNT(*) FILTER (WHERE unfulfilled_rate > 15)    AS high_miss_days,
    ROUND(AVG(unfulfilled_rate), 1)                  AS avg_miss_rate
FROM daily_zone
GROUP BY zone_id, peak_slot
HAVING COUNT(*) FILTER (WHERE unfulfilled_rate > 15) > 3
ORDER BY high_miss_days DESC, avg_miss_rate DESC;

import pandas as pd
import numpy as np
from scipy import stats

# df columns: rider_id, group ('control'/'treatment'),
#             trips_pre (pre-experiment avg trips/day),
#             trips_post (experiment avg trips/day)

def apply_cuped(df, metric_col='trips_post', covariate_col='trips_pre'):
    """Apply CUPED variance reduction to an experiment metric."""
    X = df[covariate_col].values
    Y = df[metric_col].values

    # Estimate theta via OLS: Cov(Y, X) / Var(X)
    theta = np.cov(Y, X, ddof=1)[0, 1] / np.var(X, ddof=1)

    # Compute CUPED-adjusted metric
    df = df.copy()
    df['Y_cuped'] = Y - theta * (X - X.mean())
    return df, theta

df, theta = apply_cuped(df)
print(f"Theta: {theta:.4f}")

# Variance reduction
var_original = df['trips_post'].var()
var_cuped    = df['Y_cuped'].var()
print(f"Variance reduction: {(1 - var_cuped/var_original)*100:.1f}%")

# t-test on CUPED metric
control   = df[df['group'] == 'control']['Y_cuped']
treatment = df[df['group'] == 'treatment']['Y_cuped']
t_stat, p_val = stats.ttest_ind(treatment, control)
print(f"t-stat: {t_stat:.4f}, p-value: {p_val:.4f}")

# Effect size
effect = treatment.mean() - control.mean()
print(f"ATE (CUPED): {effect:.4f} trips/day")