Timeline for Best regression model for a sigmoidal pattern
Current License: CC BY-SA 4.0
29 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 30, 2023 at 23:21 | comment | added | Martin - マーチン♦ | I'm sorry but I'll have to down vote this question: it, and all given answers are terribly misleading and show absolutely no respect for the chemistry involved. It you want to fit a couple of random points to a sigmoid function, then maths.se will be your friend. If you want to provide an actual service with your app, try to understand the chemistry involved and develop a model based on that. Polished brass isn't gold just because it shines similar in the right light. Whatever the purpose of this fit is, it looks right, while it is wrong. | |
| Aug 26, 2023 at 20:30 | history | edited | andselisk♦ | CC BY-SA 4.0 |
deleted 386 characters in body; edited title
|
| Aug 24, 2023 at 16:04 | answer | added | aaueer | timeline score: 4 | |
| Aug 24, 2023 at 11:11 | answer | added | Niels Holst | timeline score: 2 | |
| Aug 23, 2023 at 6:32 | comment | added | Poutnik | Ideally, the regression function would need only 2 parameters for strong-strong titration and 3 parameters for strong-weak titration. And 1 is already given, so 1 or 2 parameters. This way it would not be just math for math, but parameters would have direct chemical meaning. | |
| Aug 22, 2023 at 17:12 | answer | added | AccidentalTaylorExpansion | timeline score: 11 | |
| Aug 22, 2023 at 14:05 | comment | added | Ed V | The evolving consensus appears to be that the primary issue is that of fitting titration curve data to an arbitrary sigmoidal curve rather than to the relevant titration curve equations. An entirely secondary issue is that of over-fitting in whatever curve fitting procedure is employed. The most recent comment by @porphyrin nails it: “just fitting an arbitrary function misses the point and generates no understanding from the data”. (This question, unfortunately, appears to have been the X-Y problem.) | |
| Aug 22, 2023 at 7:38 | comment | added | porphyrin | @Buttonwood quite so, it is true that if you have two or more models to test against the data then you can statistically determine which is best and is the normal process, but my point is that just fitting an arbitrary function misses the point and generates no understanding from the data. | |
| Aug 21, 2023 at 22:39 | history | became hot network question | |||
| Aug 21, 2023 at 20:10 | comment | added | Buttonwood | @porphyrin One of the statistical tests in crystallography is Hamilton's $R$-factor ratio test (1964ActaCryst502). Put simply, passing from a higher space group symmetry to a lower (which is always possible), you model describing a fixed number of observations uses more and more parameters. Hamilton's test tells if the larger set of parameters offers a significantly improved model (i.e., better than just by mere addition of parameters). Is there a reliable test («Is model X for a small set of 20 points better than Y?») than $r^2$ hunting alone? | |
| Aug 21, 2023 at 19:44 | comment | added | porphyrin | You will see from the answers below that you can fit the data to various arbitrary functions given enough parameters, but this is meaningless. What you need is a model of your experiment and fit using this equation and so extract some parameters relating the the processes involved. You will need to use non-linear least squares with your own function, available in many computer packages such as Origin or Igor. (btw. What the arbitrary fitting functions do is, in effect, to smooth the data). | |
| Aug 21, 2023 at 18:32 | answer | added | ACR | timeline score: 6 | |
| Aug 21, 2023 at 18:28 | answer | added | Metal Storm | timeline score: 8 | |
| Aug 21, 2023 at 16:35 | history | edited | krirkrirk | CC BY-SA 4.0 |
added 106 characters in body
|
| Aug 21, 2023 at 15:56 | review | Close votes | |||
| Aug 26, 2023 at 3:03 | |||||
| Aug 21, 2023 at 15:53 | comment | added | Poutnik | That is what I advice. The least square method, not by some not related regression function, but with the true function with unknown yet parameters. | |
| Aug 21, 2023 at 15:52 | answer | added | Buttonwood | timeline score: 10 | |
| Aug 21, 2023 at 15:51 | comment | added | krirkrirk | @Poutnik the curve I'm looking for must fit the experimental points of the students, even if they've sent points that are falsy. The goal is to show the regression curve they did obtain with their data, even if it's a wrong one | |
| Aug 21, 2023 at 15:50 | history | edited | krirkrirk | CC BY-SA 4.0 |
added 468 characters in body
|
| Aug 21, 2023 at 15:47 | comment | added | Poutnik | You can as the function type is not dependent on experimental data, it depends only on the type of titration. It is a task for finding the minimum of the user function of sum of squares, changing function parameters. | |
| Aug 21, 2023 at 15:47 | comment | added | krirkrirk | @Mithoron thanks but not really, this question ask for a theorical equation whereas I'm looking for a regression curve that fit experimental data | |
| Aug 21, 2023 at 15:46 | comment | added | krirkrirk | @Poutnik I can't because those points are experimental results that students have sent to my app. | |
| Aug 21, 2023 at 15:46 | comment | added | krirkrirk | @porphyrin yes, a sigmoid function is exactly what I need. But what I'm having trouble with is finding the parameters a & b. I have to find a way to find these parameters automatically. | |
| Aug 21, 2023 at 15:38 | comment | added | Mithoron | chemistry.stackexchange.com/questions/117984/… | |
| Aug 21, 2023 at 15:37 | comment | added | Mithoron | Does this answer your question? Titration curve equation/function | |
| Aug 21, 2023 at 15:25 | comment | added | Poutnik | Why not to use the same function the curve theoretically follows? | |
| Aug 21, 2023 at 14:50 | comment | added | porphyrin | There are several function that you could use, look at the 'sigmoid' function in Wikipedia, and then look at the examples section. You will probably need to add a multiplier to fit the size of your data, and a parameter to make the curve sharper or shallower.,e.g a/(1+exp(-b*x)$ and vary a & b. | |
| S Aug 21, 2023 at 14:38 | review | First questions | |||
| Aug 21, 2023 at 16:20 | |||||
| S Aug 21, 2023 at 14:38 | history | asked | krirkrirk | CC BY-SA 4.0 |