Label vs Position Selection

The problem, from first principles

A DataFrame can be addressed in two different languages. By label: the row called "2024-03-01", the column called "price". By position: the third row, the second column. The two agree only when the index happens to be 0, 1, 2, ..., and the moment rows are filtered, sorted or re-indexed they stop agreeing. A bug that looks like "I selected the wrong rows" is nearly always a label used where a position was meant, or the reverse.

This question implements the selections you reach for every day, and each one is defined so you must know which language it speaks. They also settle two sharp edges: label slices include their end, position slices do not, and assigning into a selection can silently change nothing or change the original frame, so the safe pattern is to work on a copy.

From theory to code

Implement rows_by_label(df, start, stop), rows_by_position(df, start, stop), cell(df, row_label, column), numeric_columns(df) and with_value_where(df, mask, column, value). The signatures and docstrings are already in the editor.

Constraints

• df has a unique, sorted-or-unsorted index of any type and unique column names.
• rows_by_label(df, start, stop) returns the rows from label start to label stop inclusive of both ends, in the frame's own order. Both labels exist in the index.
• rows_by_position(df, start, stop) returns rows at positions start up to but not including stop, like a Python slice.
• cell(df, row_label, column) returns the single value at that row label and column name.
• numeric_columns(df) returns the names of the columns holding numbers (integers or floats, not booleans), in the frame's column order, as a list.
• with_value_where(df, mask, column, value) returns a new frame in which column is set to value on the rows where the boolean mask is true. df itself is never changed.

Hints