加权随机选择

Question

请。我有两个 table 最常见的名字和姓氏。每个 table 基本上有两个字段：

表

CREATE TABLE "common_first_name" (
    "first_name" text PRIMARY KEY, --The text representing the name
    "ratio" numeric NOT NULL, -- the % of how many times it occurs compared to the other names.     
    "inserted_at" timestamp WITH time zone DEFAULT timezone('utc'::text, now()) NOT NULL,
    "updated_at" timestamp WITH time zone DEFAULT timezone('utc'::text, now()) NOT NULL
);

CREATE TABLE "common_last_name" (
    "last_name" text PRIMARY KEY, --The text representing the name
    "ratio" numeric NOT NULL, -- the % of how many times it occurs compared to the other names.     
    "inserted_at" timestamp WITH time zone DEFAULT timezone('utc'::text, now()) NOT NULL,
    "updated_at" timestamp WITH time zone DEFAULT timezone('utc'::text, now()) NOT NULL
);

P.S：TOP 1 名称仅出现约 1.8% 的时间。 table 各有 1000 行。

函数（伪，未就绪）

CREATE OR REPLACE FUNCTION create_sample_data(p_number_of_records INT)
    RETURNS VOID
    AS $$
DECLARE
    SUM_OF_WEIGHTS CONSTANT INT := 100;
BEGIN

    FOR i IN 1..coalesce(p_number_of_records, 0) LOOP
      --Get the random first and last name but taking in consideration their probability (RATIO)round(random()*SUM_OF_WEIGHTS); 
      --create_person (random_first_name || ' ' || random_last_name);
    END LOOP;
END
$$
LANGUAGE plpgsql VOLATILE;

P.S.: 每个名字的所有比率的总和（每个 table）总和为 100%.

我想运行一个函数 N 次并得到一个名字和一个姓氏来创建示例数据...两个 table 各有 1000 行。

样本大小可以是从 1000 个全名到 1000000 个名字的任何地方，所以如果有一个“快速”的方法来执行这个随机加权函数，那就更好了。

在 PL/PGSQL 中有关于如何操作的建议吗？

我在 SUPABASE.IO 上使用 PG 13.3。

谢谢

Answer 1

鉴于输入数据集较小，在纯 SQL 中执行此操作很简单。使用 CTE 为每个 common_FOO_name 表中的每一行构建下限和上限列，然后使用 generate_series() 生成随机数集。将所有内容连接在一起，并使用边界之间的随机值作为 WHERE 子句。

with first_names_weighted as (
  select first_name,
    sum(ratio) over (order by first_name) - ratio as lower_bound,
    sum(ratio) over (order by first_name) as upper_bound
  from common_first_name
),
last_names_weighted as (
  select last_name,
    sum(ratio) over (order by last_name) - ratio as lower_bound,
    sum(ratio) over (order by last_name) as upper_bound
  from common_last_name
),
randoms as (
  select random() * (select sum(ratio) from common_first_name) as f_random,
         random() * (select sum(ratio) from common_last_name) as l_random
  from generate_series(1, 32)
)
select r, first_name, last_name
from randoms r
cross join first_names_weighted f
cross join last_names_weighted l
where f.lower_bound <= r.f_random and r.f_random <= f.upper_bound
  and l.lower_bound <= r.l_random and r.l_random <= l.upper_bound;

更改传递给 generate_series() 的值以控制要生成的名称数量。如果它是一个函数很重要，您可以只使用 LANGAUGE SQL 函数定义来参数化该数字：

https://www.db-fiddle.com/f/mmGQRhCP2W1yfhZTm1yXu5/3

加权随机选择

Weighted Random Selection

postgresql

algorithm

plpgsql

probability-distribution

supabase