Hi Heiko. The hours became days, but the problem couldn't be reproduced in a test code snippet. Any condition where the data files are incomplete is adequately detected by the Shogun error management mechanism.
SystemError: [ERROR] In file /home/ignacio/shogun/src/shogun/classifier/mkl/MKL.cpp line 208: MKLRegression::train_machine(): Number of training vectors (12917) does not match number of labels (4920) The thing is that I have the MKLRegression object embedded inside a greater object whose aim is to perform several experiments on the fly. The problem, for some unknown reason, for now, takes place right there. I configured my program for running uniquely one training and the Segmentation fault (core dumped) appears instead of the SystemError. Any way, I will validate the matching samples-labels separately. 2016-07-19 11:18 GMT+02:00 nacho arroyo <[email protected]>: > Of course... Give me some hours. > El 19/07/2016 10:52, "Heiko Strathmann" <[email protected]> > escribió: > >> Great to hear! >> Shogun should, however, not segfault in such cases, but rather tell you >> that the labels are missing. >> Can you reproduce this in a minimal example and post and issue on github? >> >> Thanks! >> H >> >> 2016-07-19 0:20 GMT+01:00 nacho arroyo <[email protected]>: >> >>> Hi Heiko, >>> >>> I had problems with my labels. After observing the files I have found >>> several of them were missing. I corrected this problem and now all is >>> working. >>> >>> Thank you very much for your support. >>> >>> 2016-07-14 17:41 GMT+02:00 Heiko Strathmann <[email protected]> >>> : >>> >>>> Hi! >>>> >>>> Can you share some more concrete information? >>>> E.g. the script to re-produce this problem? >>>> >>>> Did you try some standard KRR or SVR to see how the memory behaves? >>>> Should be ok as it is not THAT big. >>>> >>>> H >>>> >>>> 2016-07-14 13:24 GMT+01:00 nacho arroyo <[email protected]>: >>>> >>>>> Hi all, >>>>> >>>>> I have some problems trying to train MKLRegression with a dataset of >>>>> {13000 for train, 10000 for test} 20-dimensional real dense samples. >>>>> >>>>> The given error is: >>>>> >>>>> Segmentation fault (core dumped) >>>>> >>>>> When smaller a smaller dataset (e.g. of 30 20-dimensional real dense >>>>> samples), there are not problems. >>>>> >>>>> Some suggestions? in the case of memory issues, maybe there is memory >>>>> management Shogun routine/tutorial? >>>>> >>>>> Im using a 24-core 64Gb-ram machine with ubuntu 14.04 >>>>> I don't know how to know the Shogun version I have installed, but I >>>>> cloned from gitHub on Jan 12/2016. >>>>> >>>>> Thank you very much. >>>>> >>>>> -- >>>>> *Ignacio Arroyo-Fernández* >>>>> >>>> >>>> >>> >>> >>> -- >>> *Ignacio Arroyo-Fernández* >>> >> >> -- *Ignacio Arroyo-Fernández*
