Redshift中的截斷平均值

Question

如何計算紅移數據庫中的Truncated Mean？我希望它運行在非常大的數據集上

Answer 1

Redshift包含常用的SQL統計函數，包括您需要的NTILE。

SELECT AVG(CASE WHEN quartile IN (2,3) THEN my_metric ELSE NULL END) central_mean 
     ,AVG(my_metric) mean 
FROM (SELECT my_metric, NTILE(4) OVER (ORDER BY cpu_usage) quartile 
     FROM (SELECT * FROM my_table LIMIT 1000) t) t 
; 

Answer 2

您可以獲得想要修剪的百分位數的閾值。然後過濾超出這些閾值邊界的度量值，最後可以計算平均值。

SELECT avg(your_metric) 
FROM (
    SELECT 
    your_metric, 
    PERCENTILE_DISC(0.1) -- 10% lower boundary 
    WITHIN GROUP (ORDER BY your_metric) OVER() AS lower_threshold, 
    PERCENTILE_DISC(0.9) -- 90% higher boundary 
    WITHIN GROUP (ORDER BY your_metric) OVER() AS higher_threshold 
    FROM your_table 

) t1 WHERE your_metric > lower_threshold AND your_metric < higher_threshold