In Google BigQuery what is the benefit of ARRAY of STRUCT vs STRUCT of ARRAY? - sql

Question

I'm just getting to grips with ARRAYS and STRUCT in BigQuery
I'm wondering why one would choose this formatting
SELECT [STRUCT('Alice' AS col_1, 'Bob' AS col_2),STRUCT('Charlie' AS col_1, 'David' AS col_2)] AS names;
over formatting like this
SELECT STRUCT(['Alice','Charlie'] AS col_1, ['Bob','David'] AS col_2) AS names;
output 1
for both the output looks the same. What would be an example of why you would use one over the other? To me the first example makes more sense because I'd want Alice and Bob to be on the same record and it's more clear in the first example. However I've seen in Google's Vertex AI prediction output they use the second example. I.e. when outputting binary predictions they output as
SELECT STRUCT([0,1] AS class, [0.8,0.2] AS col_2) AS prediction;
instead of
SELECT [STRUCT(0 AS class, 0.8 AS prediction),STRUCT(1 AS class, 0.2 AS prediction)] AS names;
output 2
When is the right time to use each?

Answer 1

It should be based on what a row/entity represents.
In your first example, each element of the array could be a pair/team and you can add as many as you want. Alice and Bob are pair 1, Charlie and David are pair 2...
The schema would be:
names (repeated field - struct)
- col_1 (string)
- col_2 (string)
In your second example, you have an entity with two types of names. An example here would be a list of names you like and names that you don't like. So Alice and Charlie would be liked (col_1) and you can add as many as you want to this list, on the other side Bob and David would be names that you don't like (and you can add as many here as well).
Schema.
names.col_1 (repeated field - string)
names.col_2 (repeated field - string)
In JSON format:
Example 1: [{"col_1":"Alice", "col_2":"Bob"}, {"col_1":"Charlie", "col_2":"David"}]
Example 2: {"col_1":['Alice', 'Charlie'], "col_2":["Bob", "David"]}