question about the weight loss
Wangt-CN opened this issue · 2 comments
Thankyou for your great code!
Recently I am solving a problem about multilabel classify.And i have more lable(500) and is binary and one hot，but the 0 is more than 1, maybe about 50 times?
And i notice that you have use the weight loss and seem that the weight is relevent to the frequency?
Can you give me more details or ideas?
Thankyou best regards!
Thankyou for your great code!
Recently I am solving a problem about multilabel classify.And i have more lable(500) and is binary and one hot，but the 0 is more than 1, maybe about 50 times?
And i notice that you have use the weight loss and seem that the weight is relevent to the frequency?
Can you give me more details or ideas?
Thankyou best regards!
hello, i am doing a task of multilabel classification, may i have your wechat to chat with you about the problem of unbalanced data? Thank you, i am master candidate.
@Wangt-CN Sorry for replying more than a year later.
The idea was to give more weight to examples that are seen less otherwise in the extreme case say 99 true and 1 false, the network would have 100% accuracy by always predicting true.
In your case, since 2017/2018 there might have been several advances in anomaly detection techniques that should be even better than simple weight changes.