edit: nevermind I solved it
I am training an svm on a dataset with 5 classes using libsvm. I have a training set and a test set. I am using the easy.py script. The accuracy is a lot worse than the cv rate (accuracy 52%, cv rate 81%). What can be the reason for that difference?
Unlike this post, I have 200 features and 2000 examples
More info about libsvm https://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf easy.py is a script that automatically finds the best set of parameters doing grid search and computes the test accuracy