Clean Product Codes — Solution & ExplanationProblems / Clean Product Codes / EditorialApproach

split_part(product_code, '-', n) extracts each dash-separated field: 1 for the category, 2 for the SKU, 3 for the size. It is 1-indexed and needs no regex.

The SKU field is still text after splitting — CAST(... AS INT) turns '00317' into 317, giving a properly typed numeric column.

Order by product_id.

In PySpark, F.split produces an array column; .getItem(i) (0-indexed) picks each element, and .cast('int') converts the SKU.Key InsightComposite codes are a classic denormalization: one string, three facts. split_part beats regex here because the structure is rigid (exactly two dashes). The subtle step is the cast — leaving sku as text would keep zero-padded artifacts like '00317' that break numeric comparisons and joins downstream.Hints

DuckDB's split_part(string, delimiter, n) returns the n-th field, 1-indexed — call it three times with n = 1, 2, 3.

After splitting, the SKU is still a string: CAST it to INT so zero-padded values like '00317' normalize to 317.

In PySpark, F.split(F.col('product_code'), '-') yields an array; .getItem(0/1/2) picks the fields (0-indexed) and .cast('int') converts the SKU.

Solve Clean Product Codes yourself →