-
Notifications
You must be signed in to change notification settings - Fork 0
Ideas
-
triplet loss: "We sample a triplet in which the first two items are of the same class and the last is of another class. Then we create a loss such that we want the two items of the same class to be closer to each other than to the third item. FaceNet uses the difference in euclidean distance along with a preset margin. There are other formulations of triplet loss you can find as well. The type of triplet loss, as well as the number of dimensions in the embeddings, can be played around with." (Nastika Vidwan - https://www.kaggle.com/c/whale-categorization-playground/discussion/47419) see: https://www.kaggle.com/bibuml/beating-cvxtz-very-good-code-0-38-to-0-42 for possible improvement on the triplet loss.
-
Penalized classification imposes an additional cost on the model for making classification mistakes on the minority class during training. These penalties can bias the model to pay more attention to the minority class. (we might wanna be careful with this approach as the new_whale class is an important class as the test set is larger than the training set.
-
resample the very unbalanced data set into multiple balanced data sets.