Timeline for Making sense of Binominal GLM model
Current License: CC BY-SA 3.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 22, 2016 at 2:51 | vote | accept | R S | ||
| Apr 20, 2016 at 22:47 | answer | added | Josef | timeline score: 1 | |
| Apr 20, 2016 at 21:11 | comment | added | R S | Yes! Rescaling to [0,1] worked great. Please write it as an answer and I will accept it. | |
| Apr 20, 2016 at 20:28 | comment | added | Josef | One problem could be that a polynomial for x in range(50000) is very badly scaled and might mess up the optimization. Try rescaling x to for example [0, 5] interval or [-1, 1] and see if it helps. An alternative would be a low order spline. | |
| Apr 20, 2016 at 19:16 | comment | added | R S | The thing is that, with a polynomial fit to simply $s_i/n_i$, with a L2-loss I get very believable results (around where you would expect), so I don't understand why this be give worse results. I am not married to the polynomial fit idea, I just want to estimate a reasonable smooth function for $p_i$. | |
| Apr 20, 2016 at 19:07 | comment | added | Henry | Perhaps this is telling you that a polynomial fit is not a good idea | |
| Apr 20, 2016 at 18:51 | history | asked | R S | CC BY-SA 3.0 |