Looks great. I have a question, this is not over fitting??? 🤔
@shajidmughal3386
14 күн бұрын
If his validation loss is not going up, this may not be overfitting
@akaalis_csc
13 күн бұрын
Hello @elecmar_mx So glad to see first ever comment in my channel other than people I know of, My intention of this fourier series replacing any regression curve is , Fourier series can mould into any curve . dont have the constraints of the initial shape that we started with (like a "line" or "parabola" or "exponetial") In case the users feel (infinite series of 100) is overfitting they can stop at any point (like at 5 or 50 or 100 or any) in the infinite series to lessen the overfitting , as you can see in the video the series started from 1(not aligned at all) to 100(exact alignement) sine and cosine s, we can choose any middle point between 1 to infinity , all that convinience provided by fourier by just stopping with number of waves we choose To be honest , I am in the learning phase of machine learning concepts , I was always fascinated with fourier concepts , so this was just a fun hobby , there could be some mistakes
@landsgevaer
7 күн бұрын
It will surely overfit, generally speaking. Adding extra non-colinear regressors gives smaller residuals. A polynomial of arbitrarily high degree would also get closer and closer to the data. What are "good" regressors really depends on the data. A model may be better at fitting some problems, but will always be poorer in some other problems. Check out the "no free lunch" theorem.
Пікірлер: 4