Limit max_exemplars and choose them per random

[fwirthm]

Hi quantombone,

I'm courrently working with your framework. I have longer training times than in your examples, because I changed the features (to CN-HOG features). To make the training faster I would like to limit the maximum number of trained exemplars - which is possible by setting "stream_params.stream_max_ex" (for example to 100). But if I do this, the framework takes only the first 100 images and initializes 100 exemplars on this basis - now my question: Is it possible to tell the framework to select the 100 images per random out of the trainings images? In my case this is important, because my example images have filenames with a specific structure, so that images with similar names are more likely to each other...

Thank's a lot and kind regards, Florian