About the options chosen in the Train Vector Classifier (libsvm) module : - Input vector data: echantillons_global.shp (with 700 samples in 11 classes)
- Output model: model.txt - Input XML file : rstat.xml (this file has been generated with the module *Compute Image second order statistics*) - Field names for training features : meanB1, meanB2, meanB3, meanB4, meanB5, meanB6, MedB7 (these features have been computed with the module *Objects Radiometric Statistics*, which exist in Monteverdi but not in QGIS. So I have had to create an xml descriptor to use it in QGIS and it works fine with shape files.*)* - Field containing the class integer label for supervision: class - SVM Kernel : Gaussian radial basis - SVM model type : C support vector classification - Parameters otimizations : enabled - Probability estimation : enabled - Cost parameter C : 1 ==> *the classifier runs properly in 6.4.0 and with every options listed above*. But in 6.6.0, instead of displaying a contingency table and estimation performance for every classes, I get a log which does not stop (like in a loop) and without displaying any classification results. So I have to interrupt the process. The classifier seems to optimize infinitely. As for exemple, I get : Total nSV = 18 .* optimization finished, #iter = 124 nu = 0.013033 obj = -21.462956, rho = -0.347554 nSV = 18, nBSV = 0 Total nSV = 18 * optimization finished, #iter = 77 nu = 0.012001 obj = -19.762429, rho = -0.393102 nSV = 16, nBSV = 0 Total nSV = 16 *.* optimization finished, #iter = 130 nu = 0.010381 obj = -21.478759, rho = -0.334716 nSV = 19, nBSV = 0 * optimization finished, #iter = 37 nu = 0.004355 obj = -3.860645, rho = 0.055856 nSV = 14, nBSV = 0 Total nSV = 14 * optimization finished, #iter = 40 nu = 0.003965 obj = -3.516002, rho = 0.014556 nSV = 16, nBSV = 0 Total nSV = 16 *.* optimization finished, #iter = 52 nu = 0.004395 obj = -3.804882, rho = 0.069312 nSV = 18, nBSV = 0 Total nSV = 18 *.* optimization finished, #iter = 44 nu = 0.003462 obj = -3.069920, rho = 0.202671 nSV = 14, nBSV = 0 Total nSV = 14 * optimization finished, #iter = 36 nu = 0.004214 obj = -3.647664, rho = 0.004150 nSV = 17, nBSV = 0 Total nSV = 17 * optimization finished, #iter = 50 nu = 0.003556 obj = -3.904905, rho = 0.072482 nSV = 18, nBSV = 0 * optimization finished, #iter = 54 nu = 0.002071 obj = -3.454531, rho = -0.032030 nSV = 19, nBSV = 0 Total nSV = 19 * optimization finished, #iter = 44 nu = 0.002221 obj = -3.658435, rho = 0.021450 nSV = 17, nBSV = 0 Total nSV = 17 * optimization finished, #iter = 49 nu = 0.002228 obj = -3.715913, rho = 0.035005 nSV = 19, nBSV = 0 Total nSV = 19 * optimization finished, #iter = 50 nu = 0.001882 obj = -3.099503, rho = 0.123017 nSV = 18, nBSV = 0 Total nSV = 18 * optimization finished, #iter = 38 nu = 0.002078 obj = -3.422434, rho = 0.112471 nSV = 17, nBSV = 0 Total nSV = 17 * optimization finished, #iter = 58 nu = 0.001797 obj = -3.717713, rho = 0.034798 nSV = 20, nBSV = 0 * optimization finished, #iter = 20 nu = 0.002080 obj = -2.633699, rho = -0.082973 nSV = 7, nBSV = 0 Total nSV = 7 * optimization finished, #iter = 21 nu = 0.002527 obj = -3.148419, rho = -0.210446 nSV = 10, nBSV = 0 Total nSV = 10 * optimization finished, #iter = 22 nu = 0.002641 obj = -3.289533, rho = -0.234751 nSV = 9, nBSV = 0 Total nSV = 9 * optimization finished, #iter = 27 nu = 0.002652 obj = -3.303198, rho = -0.226272 nSV = 10, nBSV = 0 Total nSV = 10 * optimization finished, #iter = 22 nu = 0.002261 obj = -2.817021, rho = -0.256861 nSV = 10, nBSV = 0 Total nSV = 10 * optimization finished, #iter = 23 nu = 0.002114 obj = -3.303199, rho = -0.226222 nSV = 10, nBSV = 0 * optimization finished, #iter = 17 nu = 0.004061 obj = -4.543754, rho = -0.218781 nSV = 8, nBSV = 0 Total nSV = 8 .........AND CONTINUE OPTIMIZATION WITHOUT STOPPING..... *So I have some questions : * 1. Which new code has been added to this libsvm classifier (but I have not noticed any changes update on the website) in 6.6.0 ? 2. Do I have too many samples ? But once more, these samples have been checked and run fine with any vector classifier in 6.4.0. Hope it will help you. Awaiting any fixes, I will continue to use 6.4.0. Best regards, Etienne MAHE Le mercredi 4 juillet 2018 11:42:34 UTC+2, Etienne MAHE a écrit : > > Hi, > > Since OTB 6.6.0-Win64 release, the app Train Vector Classifier seems to > have a bug when running with libSVM classifier and especially when options > "Paramaters optimization" and "Probability estimation" are enabled. > > With previous version (6.4.0), the algorithm works properly (it displays > the confusion matrix and performance index) but since last version 6.6.0, > the optimization process runs infinitely without stopping and without > displaying performance index. I suspect an issue because if I disable those > options, the app properly works. > > But I really need those options to classify my orthotoplans though an > automated processings chain. Awaiting a fix, I continue to use OTB 6.4.0. > > > Sincerely, > Etienne. > -- -- Check the OTB FAQ at http://www.orfeo-toolbox.org/FAQ.html You received this message because you are subscribed to the Google Groups "otb-users" group. To post to this group, send email to otb-users@googlegroups.com To unsubscribe from this group, send email to otb-users+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/otb-users?hl=en --- You received this message because you are subscribed to the Google Groups "otb-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to otb-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
