Fewshot-Learning-with-BERT icon indicating copy to clipboard operation
Fewshot-Learning-with-BERT copied to clipboard

We attempt to do few shot learning with BERT and prototypical network for Intent classification

Results 3 Fewshot-Learning-with-BERT issues
Sort by recently updated
recently updated
newest added

![image](https://user-images.githubusercontent.com/19158661/196652114-497f1076-7542-4223-af1a-7d61bd95de47.png) the first image say:generated batch data ![image](https://user-images.githubusercontent.com/19158661/196650789-81707370-6504-4095-bd77-2323fa4a9a30.png) the first image say:generated lable i think they dont matched!

Hi, I am interested in your work and have tried to run train.py. Is it the result with configuration of 1-shot? When I change "Nc" to 5, I get accuracy...

train.py程序中计算准确率使用的label是随机生成的。 个人认为应该调用模型生成的label与真实label进行比较,而非随机生成的label与真实label比较,该问题导致模型训练准确率一直在0.5左右 label = torch.arange(5).repeat(10).type(torch.LongTensor).cuda() correct += float(torch.sum(logits == label).item())