Реализация qcut в SQL для создания новых столбцов - PullRequest
Новая
       12

Реализация qcut в SQL для создания новых столбцов

0 голосов
/ 18 мая 2018

Я занимаюсь анализом Recency-Frequency-Monetary и, хотя у меня есть модель, работающая на Python, я пытаюсь реализовать ее на SQL, потому что производственный код состоит в основном из PHP (Oracle 12c fwiw или также может быть выполнен в postgres),В Интернете я видел пример, в котором добавлялись новые столбцы с использованием ntile, однако я хочу использовать эквивалент квантиля (например, разделение ребер корзины недавности до 0,0,74,321 для 5 квантилей).Я начал идти по маршруту ниже, но есть ли более аккуратный путь? 

SELECT
   percentile_disc(0.2) within group (order by recency) as recency_1_quin,
   percentile_disc(0.4) within group (order by recency) as recency_2_quin,
   percentile_disc(0.6) within group (order by recency) as recency_3_quin,
   percentile_disc(0.8) within group (order by recency) as recency_4_quin,
   percentile_disc(0.2) within group (order by monetary_value) as monetary_1_quin,
   percentile_disc(0.4) within group (order by monetary_value) as monetary_2_quin,
   percentile_disc(0.6) within group (order by monetary_value) as monetary_3_quin,
   percentile_disc(0.8) within group (order by monetary_value) as monetary_4_quin,
   percentile_disc(0.2) within group (order by frequency) as frequency_1_quin,
   percentile_disc(0.4) within group (order by frequency) as frequency_2_quin,
   percentile_disc(0.6) within group (order by frequency) as monetary_3_quin,
   percentile_disc(0.8) within group (order by frequency) as monetary_4_quin
FROM RFM;

Редактировать: С помощью, это то, где я нахожусь, но застрял в группе внутри CTE.[42000] [978] ORA-00978: функция вложенной группы без GROUP BY 

RFM AS (
SELECT SRC_USER_ID,
  COUNT(distinct PICKUP_DATE) -1 as frequency,
  (MAX(PICKUP_DATE) - MIN(PICKUP_DATE)) as recency,
  (TO_DATE ('2018/05/12', 'yyyy/mm/dd') - MIN(PICKUP_DATE)) as T,
  SUM(PRICE_TOTAL) AS monetary_value
FROM TRANSACTIONS
group by SRC_USER_ID
ORDER BY frequency DESC),
  MAX_VALUES AS (
      select sum(max(recency) + 0.0000000001) as max_recency,
              sum(max(frequency) + 0.00000001) as max_frequency,
             sum(max(monetary_value) + 0.000000001) as max_monetary
      FROM RFM
  )
SELECT
    SRC_USER_ID,
    recency,
    frequency,
    monetary_value,
  WIDTH_BUCKET(recency, 0, max_recency, 5) "recency_quantile",
  WIDTH_BUCKET(frequency, 0, max_frequency, 5) "frequency_quantile",
  WIDTH_BUCKET(monetary_value, 0, max_monetary, 5) "monetary_quantile"
FROM RFM, MAX_VALUES

1 Ответ

0 голосов
/ 18 мая 2018

width_bucket () не дал то, что я искал, поэтому я получил следующее: 

RFM AS (
SELECT SRC_USER_ID,
  COUNT(distinct PICKUP_DATE) -1 as frequency,
  (MAX(PICKUP_DATE) - MIN(PICKUP_DATE)) as recency,
  (TO_DATE ('2018/05/12', 'yyyy/mm/dd') - MIN(PICKUP_DATE)) as T,
  SUM(PRICE_TOTAL) AS monetary_value
FROM TRANSACTIONS
group by SRC_USER_ID
ORDER BY frequency DESC),
QUINTILES AS (SELECT
   percentile_disc(0.2) within group (order by recency) as recency_1_quin,
   percentile_disc(0.4) within group (order by recency) as recency_2_quin,
   percentile_disc(0.6) within group (order by recency) as recency_3_quin,
   percentile_disc(0.8) within group (order by recency) as recency_4_quin,
   percentile_disc(0.2) within group (order by monetary_value) as monetary_value_1_quin,
   percentile_disc(0.4) within group (order by monetary_value) as monetary_value_2_quin,
   percentile_disc(0.6) within group (order by monetary_value) as monetary_value_3_quin,
   percentile_disc(0.8) within group (order by monetary_value) as monetary_value_4_quin,
   percentile_disc(0.2) within group (order by frequency) as frequency_1_quin,
   percentile_disc(0.4) within group (order by frequency) as frequency_2_quin,
   percentile_disc(0.6) within group (order by frequency) as frequency_3_quin,
   percentile_disc(0.8) within group (order by frequency) as frequency_4_quin
FROM RFM)
SELECT
    SRC_USER_ID,
    recency,
    frequency,
    monetary_value,
  (CASE WHEN recency <= recency_1_quin THEN 1
   WHEN recency > recency_1_quin and recency <= recency_2_quin THEN 2
   WHEN recency > recency_2_quin and recency <= recency_3_quin THEN 3
   WHEN recency > recency_3_quin and recency <= recency_4_quin THEN 4
   ELSE 5 END) "recency_quantile",
       (CASE WHEN frequency <= frequency_1_quin THEN 1
   WHEN frequency > frequency_1_quin and frequency <= frequency_2_quin THEN 2
   WHEN frequency > frequency_2_quin and frequency <= frequency_3_quin THEN 3
   WHEN frequency > frequency_3_quin and frequency <= frequency_4_quin THEN 4
   ELSE 5 END) "frequency_quantile",
     (CASE WHEN monetary_value <= monetary_value_1_quin THEN 1
   WHEN monetary_value > monetary_value_1_quin and monetary_value <= monetary_value_2_quin THEN 2
   WHEN monetary_value > monetary_value_2_quin and monetary_value <= monetary_value_3_quin THEN 3
   WHEN monetary_value > monetary_value_3_quin and monetary_value <= monetary_value_4_quin THEN 4
   ELSE 5 END) "monetary_value_quantile"

FROM RFM, QUINTILES
ORDER BY recency DESC, frequency DESC, monetary_value DESC;
...