Both of the anti-patterns have df.groupby('group_col')['agg_col'] ... , which we should be able to identify with the following check:

if isinstance(node.func.value, ast.Subscript) and 
              node.func.value.value.func.attr == 'group_by'

(AST chains can get crazy)

Do you see any cases where this pattern wouldn't work (i.e., other groupby anti-patterns that don't use slicing? or examples where slicing the groupby object would be acceptable?)

Turns out the AST syntax is actually more complex when the .agg() method is chained. But the fundamental pattern should still be the same -- we want to check for groupby() methods that also use slicing.

Unless there are cases where that syntax might be appropriate.

Unless there are cases where that syntax might be appropriate.

This is my fear. I am having second thoughts about this because groupby is EXTREMELY common in pandas code and I am afraid there are probably lots of cases I am not considering.

I am going to suggest that we should hold off on implementing this until we have a better survey of groupby in the wild. I will open an issue where we can collect examples.