2

I am using Mean Absolute Value(MAE) and Mean Squared Error(MSE) as my evaluation parameters for the regression model. I am trying to predict the future values for time series model and have used CNN fro the same. However, my CNN model is giving very low values for MSE compared to MAE. On such example is

MAE:0.04978915070122473, MSE:0.004155844765967494

I know for sure that MAE has to be always less than MSE since we are squaring the coefficients in MSE.

Rohan Akut
  • 23
  • 4

1 Answers1

2

You're forgetting about squaring small numbers. Here is an example where the MSE would be smaller:

import numpy as np
label = np.array([0.392, 0.342, 0.344, 0.534])
predict = np.array([0.394, 0.340, 0.347, 0.525])

MSE = np.average((label - predict)**2)
MAE = np.average(np.abs(label - predict))

print(MSE)
print(MAE)

>>>2.4500000000000043e-05
>>>0.0040000000000000036

label *= 1000
predict *=  1000

MSE = np.average((label - predict)**2)
MAE = np.average(np.abs(label - predict))

print(MSE)
print(MAE)

>>>24.5
>>>4.0