Files
Data_engineer_test_task/sql/anomalies.sql
T

34 lines
1.4 KiB
SQL

WITH RECURSIVE calendar(day, last_day) AS (
SELECT MIN(order_date), MAX(order_date) FROM daily_category_revenue
HAVING COUNT(*) > 0
UNION ALL
SELECT date(day, '+1 day'), last_day FROM calendar WHERE day < last_day
),
categories AS (
SELECT DISTINCT category FROM daily_category_revenue
),
daily AS (
SELECT day, c.category, COALESCE(r.revenue_kopecks, 0) AS revenue_kopecks
FROM calendar CROSS JOIN categories AS c
LEFT JOIN daily_category_revenue AS r ON r.order_date = day AND r.category = c.category
),
-- Вычитаем общий для категории минимум в целых копейках до перехода AVG к REAL.
-- Такой сдвиг сохраняет отклонения и различия в копейку при большом базовом уровне.
shifted AS (
SELECT *, revenue_kopecks - MIN(revenue_kopecks) OVER (PARTITION BY category) AS amount
FROM daily
),
deviations AS (
SELECT *, amount - AVG(amount) OVER (PARTITION BY category) AS deviation
FROM shifted
),
stats AS (
SELECT *, COUNT(*) OVER (PARTITION BY category) AS days,
AVG(deviation * deviation) OVER (PARTITION BY category) AS variance
FROM deviations
)
SELECT day AS order_date, category, revenue_kopecks
FROM stats
WHERE days >= 7 AND variance > 0 AND deviation * deviation > 4 * variance
ORDER BY day, category;