| Author | Message | ||
     Mariel elayne monrroy (mariel) New member Username: mariel Post Number: 1 Registered: 8-2008 |
DEAR: Do I need know if PLS spreads to overestimate the models ? and in the PCR case like is ? thank you | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 32 Registered: 3-2004 |
Forrest Stout PLS stand good in case of quantitative prediction only. Qualitative prediction , I have used Pattern recognition with well preprocessed spectral data . I got good result. Now I am working on the method of selection of training set through PCA and the components are marked as quality one two etc.. and applying PLS instead of Pattern reognition . The results are encouraging one. | ||
| Forrest Stout (forrest) Intermediate Member Username: forrest Post Number: 18 Registered: 7-2006 |
When you say "[m]ore over extraction of information from the spectral data for multi component ,Instrument engineers prefer PLS," are you referring to quantitative predictions or qualitative analysis of the components detected spectroscopically? For the later, does PLS hold any benefits over principal component analysis (PCA)? ...I have never actually used PLS for the such qualitative analysis. | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 28 Registered: 3-2004 |
Forrest Stout, your commands are welcomed. You are aware that the model depend upon the data preprocessing . As a instrument men I have observed PLS is time optimized one and implementation in embedded system is a straight forward method. More over extraction of information from the spectral data for multi component ,Instrument engineers prefer PLS. | ||
| Forrest Stout (forrest) Intermediate Member Username: forrest Post Number: 17 Registered: 7-2006 |
Venkynir, did you mean to give those three points to PCR, instead of PLS? As has been stated, I think that PLS and PCR are very comparable. (Note: PLS and PCR factors are in a different space and cannot be directly compared. E.g. a 2 factor PLS model has a different degree of freedom than a 2 factor PCR model.) However, in the realm of splitting hairs, I would say that PCR is: 1) A more compact program (using Matlab). 2) Gives more useful component analysis On the issue of speed, it all depends on the programs employed. I've used PLS programs that run faster than PCR and many that run slower than PCR. | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 27 Registered: 3-2004 |
Dear Christina, Tony has rightly pointed out the application of PLS and PCR.From the publication and work experience , if your are developing a smart sensor system around NIRS ,it is preferable PLS. 1.Implementation of algorithms is compact 2.Takes optimum time for execution and implementation is more handy. 3.Through PLS we can extract /analysis multiple component's (e.g gasoline products BTEX) | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 30 Registered: 9-2001 |
In theory, PLS should give better accuracy than PCR, for the same number of factors (very important condition!!) I do not find it at all surprising to learn that this improvement in accuracy would come (in at least some cases) at the expense of precision. \o/ /_\ | ||
| Forrest Stout (forrest) New member Username: forrest Post Number: 3 Registered: 7-2006 |
In my experience PCR yields lower accuracy but higher precision than PLS. Although, I agree with Tony that there really isn't much prediction quality difference and PLS is far more difficult to understand. | ||
| Tony Davies (Td) Moderator Username: Td Post Number: 123 Registered: 1-2001 |
Dear Christina, As far as the final results are concerned there is not much to choose between PLS and PCR. This does not prevent us having very strongly held views about their relative merits! I would much prefer to have programs that do PCR by the optimum method (many do not) and not have PLS but the quality of PLS software is better. The disadvatages of PLS is that it is more complex to understand and to do it properly requires more samples. The main advantage of PLS is that the majority of people believe that it is the best way of using NIR data and if you use the software correctly (and if NIR analysis is applicable) you will get a useful result. I cannot remember hearing of an example which could only be successfully analysed by either PLS or PCR but not the other. If you are using PCR successfully do not bother with PLS but if you are just starting then it is probably better to go for PLS. Best wishes, Tony | ||
| MPDC (Unregistered Guest) Unregistered guest |
In short, PLS concentrates more of the correlations between spectrum and reference into the first principal components. For the long story, you really should read a book. In most situations, PLS will do just fine. | ||
| Christina Timofeyenko (Ctimofeyenko) New member Username: Ctimofeyenko Post Number: 11 Registered: 7-2005 |
What are the advantages and disadvantages of using PLS instead of PCR? |