Search code examples
mathdeep-learningkerasmse

What is wrong with Root Mean Square Error?

I don't understand what is wrong with my rMSE implementation. I'm training my model using MSE as loss function and same for metrics. After training, I use evaluate function for evaluate my model in test set, and then predict function to get values. Then I apply rMSE. My code is:

obs= model.compile(loss='mse', optimizer=keras.optimizers.Adam(lr=0.001),metrics=['mse'])
.......
test_eval = model.evaluate(X_test, Y_test, verbose=1)
print('Test loss (MSE):', test_eval[0])
predicted= model.predict(X_test, verbose=0)
rMSE = np.sqrt(pow(np.mean(predited- Y_test), 2))
print(rMSE)

And I had this results:

Test loss (MSE): 12.0075311661
2.90274470011

But square of 12.0075311661 isn't 2.90274470011. So, what is wrong?

Solution

  • Elementwise square the differences before finding the mean. You want to find the average of the squared difference, not the square of the average distance.