Difference in rmse values

[sheecegardezi]

Hi,
I wrote a implantation for rmse. I am getting different results when I compare it with sklearns implantation:

Simple example to reproduce:

    import numpy as np
    from sklearn.metrics import mean_squared_error


    def custom_rmse_function(y_pred, y_oos):
           return np.average(np.sqrt(np.square(np.subtract(y_oos, y_pred))))


    y_oos = [0.58, 0.55, 0.48, 0.52, 0.62]
    y_pred = [0.60, 0.59, 0.57, 0.60, 0.57]

    custom_rmse_function_results = custom_rmse_function(y_pred, y_oos)
    print("custom_rmse_function_results: ", custom_rmse_function_results)
    #custom_rmse_function_results:  0.05599999999999998

    sklearn_mean_squared_error_results = np.sqrt(mean_squared_error(y_oos, y_pred))
    print("sklearn_mean_squared_error_results: ", sklearn_mean_squared_error_results)
    #sklearn_mean_squared_error_results:  0.061644140029689744

Any guidance into why there is a discrepancy in results will be greatly appreciated.

[i-aki-y]

RMSE is the "square root of mean square errors".
However, your implementation calculates the "mean of square root of square errors".
These are mathematically different.
You can write like np.sqrt(np.average(np.square(np.subtract(y_oos, y_pred)))).

from sklearn.metrics import mean_squared_error

def fix_custom_rmse_function(y_pred, y_oos):
    return np.sqrt(np.average(np.square(np.subtract(y_oos, y_pred))))

y_oos = [0.58, 0.55, 0.48, 0.52, 0.62]
y_pred = [0.60, 0.59, 0.57, 0.60, 0.57]

print(fix_custom_rmse_function(y_oos, y_pred ))
print(np.sqrt(mean_squared_error(y_oos, y_pred)))
> 0.061644140029689744
> 0.061644140029689744

[sheecegardezi]

this makes sense thanks for the clarification.

[lesteve]

Thanks a lot @i-aki-y for your answer!

We enabled Github Discussions hoping that it would help getting scikit-learn users involved to answer other users questions. This is great to see it actually happening !