Adding SMAPE and MAPE loss functions in Machine Learning

[VedanshTyagi]

Feature Description

I would like to add SMAPE and MAPE loss functions to the loss_functions.py file in Machine Learning folder.

SMAPE

SMAPE (Symmetric Mean Absolute Percentage Error) measures forecast accuracy by comparing the absolute difference between predicted (𝐹_t) and actual (A_t) values to their average.
It is given by:
$$SMAPE=\frac{100%}{​n} \sum_{t=1}^{n} \frac{|Ft - At|}{(|At| + |Ft|)/2} $$

MAPE

MAPE (Mean Absolute Percentage Error) measures forecast accuracy by expressing the average absolute difference between predicted (𝐹_t) and actual (A_t) values as a percentage of the actual values.
It is given by:
$$SMAPE=\frac{100%}{​n} \sum_{t=1}^{n} |\frac{Ft - At}{At }| $$

Benefits

Machine learning models which predict continuous values like price of a house heavily use these loss functions to learn weights and biases.

[VedanshTyagi]

This will be my first open-source contribution and will also be a part of Hacktoberfest '25 . Please assign this issue to me.

[rajeev0521]

Hello,

I would like to be assigned this issue. I plan to implement SMAPE (Symmetric Mean Absolute Percentage Error) and MAPE (Mean Absolute Percentage Error) loss functions in the loss_functions.py file within the Machine Learning folder.

These metrics are commonly used in regression and forecasting tasks, and their inclusion will provide more flexibility for evaluating model performance.

Could you please assign this issue to me?

Thank you!

[cclauss]

@priya-sundaram-dev, please review this issue and the two PRs associated with it.

[priya-sundaram-dev]

@cclauss reviewed both PRs. #13355 (SMAPE) — approved: matches the issue formula, doctests verified (≈0.05702), CI green. #13357 (MAPE) — requested changes: math is right but CI is red and the doctests copy-paste SMAPE's function name (they call symmetric_mean_absolute_percentage_error instead of mean_absolute_percentage_error), plus a mislabeled formula and docstring indentation/line-length lint issues. Details on each PR. Note: both edit the same block of loss_functions.py, so the second to merge needs a trivial rebase.