In deep learning, multi-category only converges on categories with more samples?

Why do I train the multi-category tag dataset, the network predicts the most categories of the category, the category with the highest probability of prediction is only the same as the number of all categories, and the probability of the sample with the largest number of predictions is the largest ?

If your data is imbalanced, you might want to use a WeightedRandomSampler for oversampling the minority classes or weight the criterion.
Let me know, if that’s the case for you.

thank you! when i use WeightedRandomSampler ,it works.