R.A Fisher said in this essay that "it(Least Square Method) is obviously
inapplicable to fit frequency curve".
The following is a proof-of-concept demo that shows one CAN fit a frequency curve
using Least Square and it's "good to the eye", quantitativly good and fast.
The result is also compared with MLE which is proposed in this essay as well.
The only trick for using least square is that one has to scale the loss function
to help optimizer cope with numerical instability caused by probability density
functions.
On An Absolute Criterion for Fitting Frequency Curves http://www.jstor.org/stable/2246266
R.A Fisher said in this essay that "it(Least Square Method) is obviously inapplicable to fit frequency curve".
The following is a proof-of-concept demo that shows one CAN fit a frequency curve using Least Square and it's "good to the eye", quantitativly good and fast. The result is also compared with MLE which is proposed in this essay as well.
The only trick for using least square is that one has to scale the loss function to help optimizer cope with numerical instability caused by probability density functions.