Label-flipping: Use EmailDataset.split and output fixes, refactor, use distance < alpha and not number of iterations, use non-integer q value, move to ECOS from SCS for performance reasons (bug seems fixed?), original q has the proper format
K-Insertion: Bound values in self.fvs
IRL: SimpleLearner -> TRIMLearner, set self.loss_threshold appropriately and only once, move from 50% to 40% of labels flipped during label-flipping attack
TRIM learner: Switch to ECOS from SCS for speed
Utils: Add np.ndarray support to logistic_loss
CVXPY: Force version 0.4-0.4.11
General: Use numpy functions instead of map whenever possible
EmailDataset.splitand output fixes, refactor, use distance < alpha and not number of iterations, use non-integer q value, move to ECOS from SCS for performance reasons (bug seems fixed?), original q has the proper formatself.fvsSimpleLearner->TRIMLearner, setself.loss_thresholdappropriately and only once, move from 50% to 40% of labels flipped during label-flipping attacknp.ndarraysupport tologistic_lossnumpyfunctions instead ofmapwhenever possible