Problem

At Nimbus Analytics, analysts want a simple helper that recommends clear chart types for exploratory data analysis. Given metadata about dataset columns and optional analysis goals, write a function that returns the most appropriate chart recommendation for each request.

Formal Specification

Implement a function that takes:

• columns: a list of dictionaries, where each dictionary has:
  • name: string column name
  • type: one of "numeric", "categorical", "datetime", "boolean"
  • unique_count: non-negative integer
• requests: a list of dictionaries, where each dictionary has:
  • columns: list of 1 or 2 column names
  • goal: one of "distribution", "comparison", "relationship", "trend", "composition"

Return a list of chart names, one per request, using these rules:

• 1 numeric + distribution -> "histogram"
• 1 categorical/boolean + comparison/composition -> "bar"
• 1 datetime + trend -> "line"
• 2 numeric + relationship -> "scatter"
• 1 categorical + 1 numeric + comparison -> "bar"
• 1 datetime + 1 numeric + trend -> "line"
• 2 categorical + relationship/composition -> "stacked_bar"
• Otherwise -> "table"

If a request references an unknown column, return "invalid" for that request.

Examples

Example 1 Input: columns = [{"name":"age","type":"numeric","unique_count":40}], requests = [{"columns":["age"],"goal":"distribution"}] Output: ["histogram"] Explanation: A single numeric variable is best shown with a histogram for distribution.

Example 2 Input: columns = [{"name":"date","type":"datetime","unique_count":365},{"name":"sales","type":"numeric","unique_count":200}], requests = [{"columns":["date","sales"],"goal":"trend"}] Output: ["line"] Explanation: A time variable paired with a numeric measure is best shown as a line chart.

Constraints

• 1 <= len(columns) <= 10^4
• 1 <= len(requests) <= 10^4
• Each request contains 1 or 2 column names
• Column names are unique