​
Input

  • output (str): The generated text to be evaluated.
  • expectedOutput (str): The reference or ground truth text.

​
Output

  • Result (float): A distance score from 0 to infinity.

​
Interpretation

  • 0: The embeddings are identical.
  • Higher scores: The texts are more semantically different.
    Generally less sensitive to outliers than Euclidean distance.

​
Formula

Manhattan(x,y)=∑i∣xi−yi∣\mathrm{Manhattan}(x, y) = \sum_i |x_i - y_i| Where i indexes the dimensions of the vectors.
This is a distance metric. Lower scores indicate greater similarity.

​
How It Works

The evaluator computes embeddings for both texts and then sums the absolute differences for each corresponding dimension.

​
Use Cases

  • Scenarios where movement is constrained to a grid (e.g., city blocks)
  • Feature engineering in machine learning, as it can be more robust to outliers
  • High-dimensional spaces where it can be more effective than Euclidean distance