Problem

• When scoring a structured data model, there was a very large difference between scoring the entire dataset and scoring by stratification
• Calculation error

Cause

• One shape was a pandas Series and the other was a pandas DataFrame. This was the difference between a 2D array and a 1D array.

• When implementing SMAPE, the shapes of the ground truth (pandas Series) and prediction (pandas DataFrame) data were different -> (n,) and (n,1)

• When using NumPy in SMAPE, broadcasting automatically expanded the dimensions, resulting in an incorrect calculation

• When mixing 1D and 2D arrays, broadcasting caused unintended calculations

Solution

• Flatten the DataFrame values and proceed