thuiar / adaptive-decision-boundary Goto Github PK

I config the bert_model as ./pretrained_models/uncased_L-12_H-768_A-12.
But I face the error as follow：
Model name './pretrained_models/uncased_L-12_H-768_A-12' was not found in model name list (bert-base-uncased, bert-large-uncased, bert-base-cased, bert-large-cased, bert-base-multilingual-uncased, bert-base-multilingual-cased, bert-base-chinese). We assumed './pretrained_models/uncased_L-12_H-768_A-12' was a path or url but couldn't find any file associated to this path or url.
Can u help me?

Hugging Face offers the latest library versions, but I'm curious as to why older versions of BERT, RoBERTa, etc., are being used. Could you explain the reason for this?

Can you provide the result of Figure 5?
The results of model (MSP, DOC, OpenMax, DeepUnk, ADB) under different settings (different labeled ratio and different scales of known intents).

Hi Hanlei,
Recently, I study many papers on open intent classification. Some questions make me confuse. Can you explain it to me? Thank you!

1. I notice that the results in Table 2 from [1] are macro f1-score of unknown intent detection. This is different form your ADB method. But i don't find a clear definition of unknown intent detection. And it seems not mean the macro f1-score of open class like Table 3 in ADB.
2. And what do you think the difference between open intent classification and out-of-domain (out-of-scope) intent detection?
3. The metric is confuse. AUROC, AUPR, and so on are used in out-of-domain [2]. I notice that the metric is used often is OOD paper. But it is merely used in open intent classification or unknown intent detection task. What your idea about this?

Finally, thank you a lot.

1. Deep Unknown Intent Detection with Margin Loss
2. Out-of-Domain Detection for Natural Language Understanding in Dialog Systems