想构建一个自定义报告 - 关于某个话题的最后两次回复之间的时间

社区 数据与报告
2

对于感兴趣的人，这里是似乎有效的方法：

-- [params]
-- null date :start_date
-- null date :end_date
-- null int_list :category_ids

WITH RankedPosts AS (
    SELECT
        p.topic_id,
        p.created_at,
        ROW_NUMBER() OVER (PARTITION BY p.topic_id ORDER BY p.post_number DESC) AS rank
    FROM
        posts p
    WHERE
        p.created_at BETWEEN :start_date AND :end_date
        AND EXISTS (
            SELECT 1
            FROM topics t
            WHERE t.id = p.topic_id
            AND t.category_id IN (:category_ids)
        )
),
FilteredPosts AS (
    SELECT
        topic_id,
        created_at,
        rank
    FROM
        RankedPosts
    WHERE
        rank <= 2
),
PostDifferences AS (
    SELECT
        topic_id,
        EXTRACT(days FROM (MAX(created_at) FILTER (WHERE rank = 1) - MAX(created_at) FILTER (WHERE rank = 2)))::numeric(9,2) AS days_difference
    FROM
        FilteredPosts
    GROUP BY
        topic_id
)
SELECT
    t.category_id,
    AVG(pd.days_difference) AS avg_days_difference,
    PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY pd.days_difference) AS median_days_difference
FROM
    topics t
JOIN
    PostDifferences pd ON t.id = pd.topic_id
GROUP BY
    t.category_id
ORDER BY
    avg_days_difference DESC

这应该可以给出创建时间戳之间平均值和中位数。更新时间戳似乎有问题（由于某种原因，当我只计算平均值时，我得到了一些负值）。

我得到了 AI 的帮助来创建这个查询，所以有可能存在错误——如果有人有关于如何改进它（或纠正任何错误）的建议，我将不胜感激。我得到的结果在对较小群体进行的检查以及日期范围增加导致趋势与我从更大样本数据中获得的预期一致的更改方面，似乎是合理的。