lackel / dpn Goto Github PK

It seems self.predict_k(args, data) is not defined in the code (line numbers 32 and 33 in DPN.py) for args.cluster_num_factor > 1: (https://github.com/Lackel/DPN/blob/main/DPN.py#L33)

if args.cluster_num_factor > 1: self.num_labels = self.predict_k(args, data)

I tried to run the default code given for the "StackOverflow" dataset and the Clinc dataset. I am getting the following results:

got for Stackoverflow dataset: {'ACC_all': 75.6, 'ACC_known': 81.2, 'ACC_novel': 58.8}

Provided in paper for Stackoverflow dataset: {'ACC_all': 84.23, 'ACC_known': 85.29, 'ACC_novel': 81.07}

Please let me know how can I reproduce the results. I tried varying parameters also like 'predict_k'=2 or 3 but was unable to reproduce the results.

hi wenbin:
感谢贡献非常出色的工作，有一些疑问还麻烦解答一下，感谢：
1、

Hi ,thank you for your inspiring wok! I find that the code uses softmax function in https://github.com/Lackel/DPN/blob/main/DPN.py#L203. I don't know why the paramater dim is set 0.