How does log of y and y_pred make a RMSPE to RMSE?

mlnoob · June 23, 2019, 4:00am

This is from the notes on lesson 6, but basically it says

Can someone show and explain to me the math that does this? This image is from https://github.com/hiromis/notes/blob/master/Lesson6.md

oneironaut · June 23, 2019, 7:16pm

The way Jeremy said it in the lecture seems to be wrong. One can see this with a simple example, by choosing to have only one pair of (y,y^) to compare (so n=1 in the equation above):

Maybe @jeremy can clarify this?

EDIT: Wait, mistake in the formula
EDIT^2: Fixed

joerg · July 1, 2019, 8:37am

I have the same question. The summands in RMSE_LOG can merely be rewritten as

which is still not the same as the RMSPE mathematically. The fact that log() turns quotients into differences and vice-versa does not seem to be enough in this case.

Brainkite · September 4, 2019, 4:12pm

Yes same question here.

pai095 · September 16, 2019, 3:23am

I have the same question as well.

Jasmine · October 18, 2019, 9:58am

Same question as well.

ashishp · March 20, 2020, 1:38pm

We need to find a function mapping for y and y_pred that will create new Y and Y_PRED s.t that the RMSPE (y,y_pred) = RMSE(Y,Y_PRED). This function comes out to be a log function using Taylor’s expansion. Please follow the link below for more: