| Author | Message | |||
     Jos� Ram�n Cuesta (jrcuesta) Member Username: jrcuesta Post Number: 15 Registered: 11-2009 |
The selection for the samples in the cross validation are based in the order of the samples in the CAL file, so except if you use the method "leave one out" (available in the last 4.3 Win ISI version) the calculations are diferent so you can see small changes. | |||
| Jos� Ram�n Cuesta (jrcuesta) Member Username: jrcuesta Post Number: 14 Registered: 11-2009 |
You get diferent statistics in the equation because of the cross validation groups are different in bougth. | |||
| KongFanli (kongfanli) New member Username: kongfanli Post Number: 4 Registered: 10-2011 |
And the number of outlier elimination passes I chose 0 thanks again Yours Kong Fanli | |||
| KongFanli (kongfanli) New member Username: kongfanli Post Number: 3 Registered: 10-2011 |
Dear all Thank you very much for your answers. It is true when you develop a equation, you can calculate the GH by your idea, but if it is a new sample you do not know the score , can you calculate the H ? And now I have a new problem, the same data, same math treatment, same scatter, the only difference is the data order (first equation, Series of data following one another in time .second equation I sort the order by reference value ), I got difference equations, although the difference is very small. Best wishes... | |||
| Ian Michael (admin) Board Administrator Username: admin Post Number: 33 Registered: 1-2006 |
Looks like a problem with modern filenames! Either download the .unk and rename to .xslx or if my uploading of an .xls works use that below!
| |||
| Pierre Dardenne (dardenne) Senior Member Username: dardenne Post Number: 69 Registered: 3-2002 |
Hi, something was wrong with the file. New attempt
| |||
| Pierre Dardenne (dardenne) Senior Member Username: dardenne Post Number: 68 Registered: 3-2002 |
Hi, With ISI, the GH is the sum of the squares of the standardized scores divided by the number of PC (or the average of all the sums the scores). Basically a square root would have to be applied to have an �actual� distance. Pierre Calculation in the excel file.
| |||
| Peter Tillmann (tillmann) Advanced Member Username: tillmann Post Number: 25 Registered: 11-2001 |
Dear KongFanli, the Hs are scores and they are calculated according to the textbooks for PLS mathematics (e.g. Martens and Martens, Multivariate calibrations or any other textbook). Whether you do it manually (manus (lat.): hand) or use a PC may be up to you. I have my personal preferences. Be aware that the calculation of Hs depend on the software implementation. I.e. the Hs from ISI differ from the Ms in Grams, while they have the same scientific background and the same purpose. Usually it will be something like normalisation or squares that cause the differrences. Using the text book algorithms you will find the same samples to be unique/outliers. Yours Peter Tillmann | |||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 460 Registered: 9-2001 |
Dear KongFanli - yes, in principle, anything that can be calculated by a computer can be calculated manually. It will, however, take a lot longer. These are matrix calculations, and generally require an amount of computation proportional to the square of the number of wavelengths used. Therefore, unless you're dealing with very narrow wavelength ranges, the computations are not practical to do without a computer. If you want to verify that the computations are being done correctly, there are several approaches: 1) Narrow down the wavelength range to the point where you can do the calculations manually. Don't be concerned about optimizing the calibration, just see if you can reproduce the computer's results, good or bad. 2) Write your own program to do that calculation. 3) Find some other program that does the same calculation and apply it to your data. 3) There are some standard data sets on the NIST website; http://www.itl.nist.gov/div898/strd/ with various statistics calculated and certified correct. See if they do the calculation you want, then you can download tha data set and run it through the program you want to check \o/ /_\ | |||
| KongFanli (kongfanli) New member Username: kongfanli Post Number: 2 Registered: 10-2011 |
Dear all I am using the FOSS NIR. The WinISI and ISIscan will give you a GH&NH value automatically when you predict a sample, I konw the values means the distance between the sample and the equation,but i want to konw can we calulate the value by manually? thans Best wishes |
xlsx file