Author |
Message |
Dongsheng Bu (dbu)
New member Username: dbu
Post Number: 5 Registered: 6-2006
| Posted on Monday, September 18, 2006 - 4:14 pm: | |
Pedro-Other important nature (spectroscopy) of derivatives is to remove baseline drift (1st derivative), and remove linear baseline slope change (2nd derivative). Noise could be enlarged in the derivatives, however, several derivative methods such as Savitzky-Golay, and Gap-Segment can provide smoothing function at the same time, which can take care measurement noise issue. Lidia-I have same interest as yours, and wait for comments on option, and application from whoever wrote the program. I tried polypls, spl_pls before, though. My suggestion is: you may try multiple regression models with classification results if your data is multivariate spectroscopic data. dbu |
Pedro Castro Nunes Fiolhais (pedro_fiolhais)
New member Username: pedro_fiolhais
Post Number: 4 Registered: 9-2006
| Posted on Friday, September 15, 2006 - 6:44 am: | |
derivatives are used to maximize diferences in the spectra. somtimes you can�t see differences in the original spectra, but then you derivate it and see it. this is due to the nature of the derivation (1st dev=velocity and 2nd=acelaration). So with derivatves you can feel the spectra tendency. the problem of derivating is that you not only maximize diferences, but also the noise! so you have to find your own equilibrium. |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 51 Registered: 9-2001
| Posted on Friday, September 15, 2006 - 4:48 am: | |
Lidia - well of course, "it depends" is always a general truth about the application of NIR. NIR is a very empirical analytical method. The intended purpose of using derivatives is to remove, or at least reduce, the effect of changes in optical scattering between different samples. That tends to be fairly effective for visually comparing spectra with each other. As to whether it improves calibration performance, there are different opinions as to its effectiveness. But you can always try it and see if it helps you with your particular data from your application. |
Lidia Esteve (veiva)
New member Username: veiva
Post Number: 3 Registered: 9-2006
| Posted on Friday, September 15, 2006 - 3:07 am: | |
Hello again, I am checking one of your papers (Howard) about the derivatives, and I wanted to ask a basic question (sure it's very common but I am a very beginner in NIR): which advantatges do derivatives offer compared with other pretreatment methods? Or should we say that it always depends on the data set and the "situation"? |
Lidia Esteve (veiva)
New member Username: veiva
Post Number: 2 Registered: 9-2006
| Posted on Thursday, September 14, 2006 - 9:14 pm: | |
Thank you Howard, In the tool box documentation it seems that the 'default' options are between brakets. I think it uses globalPCR if you don't choose any other option. Anyway, I will try to contact someone from the company to be sure. Thanks again! |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 50 Registered: 9-2001
| Posted on Thursday, September 14, 2006 - 3:28 pm: | |
Lidia - this isn't a standard MATLAB function. Therefore you need to ask whoever wrote the program, how it behaves under those conditions. Howard \o/ /_\ |
Lidia Esteve (veiva)
New member Username: veiva
Post Number: 1 Registered: 9-2006
| Posted on Thursday, September 14, 2006 - 2:23 pm: | |
Hello everybody, It's my first experience with matlab, and i am trying to develop a LWR model since PLS with unscrambler did not give nice results because of clustering. The function lwrpred offers the possibility to choose the regression algorythm, but... If i do not choose one, which one does matlab use? |