查询替换map(bigint, array(bigint))列中的值

Question

c_tbl 表具有格式为 -map(varchar, array(int)) 的特征列 映射表有两列 id, s_id

特征中的行示例 -

row1
{
"0": [90],
"1":[80, "60", -87],
"2":[95, "67", 85]
}

row2
{
"0": [99],
"1":[82, "62", -107],
"2":[195, 167, -185]
}

映射表行示例

67, 1111
167, 2222

生成一个 presto SELECT 语句，生成两列 - features 和 features_updated

更新后将具有以下逻辑， 如果键 = 2 并且 如果特征数组中的值与映射表的 id 列匹配，则将其替换为 s_id，否则保持原样

例如 features_updated 将有以下输出 -

row1
{
"0": [90],
"1":[80, "60", -87],
"2":[95, "1111", 85]
}
row2
{
"0": [99],
"1":[82, "62", -107],
"2":[195, 2222, -185]
}

请注意 chtgpt 没有给我正确的 SQL 查询：P

我尝试过这种方法 -

    SELECT 
      features,
      map_agg(
        key,
        CASE 
          WHEN key = '2' AND m.s_id IS NOT NULL THEN CAST(m.s_id AS varchar)
          ELSE CAST(value AS varchar)
        END
      ) AS features_updated
    FROM c_tbl
    CROSS JOIN UNNEST(features) AS t(key, values)
    CROSS JOIN UNNEST(values) AS val(value)
    LEFT JOIN mapping m ON key = '2'
                              AND try(CAST(val.value AS integer)) = m.id
    GROUP BY features

Answer 1

使用转换函数迭代映射中每个键的数组。对于每个值v，它在映射表中查找s_id，其中id等于v。如果找到匹配，则将v替换为s_id；否则，当子查询返回 NULL 时，v 由 COALESCE 返回。

SELECT
     features
    , map_agg(KEY, CASE 
            WHEN KEY = '2'
                THEN transform(VALUES, v - > COALESCE (
                        (
                            SELECT s_id
                            FROM mapping
                            WHERE id = v
                            )
                        , v
                        ))
            ELSE VALUES
            END) AS features_updated
FROM c_tbl
CROSS JOIN UNNEST(features) AS t(KEY, VALUES)
GROUP BY features