Wuwei Lan

Thanks! I think this was caused by PyTorch version difference. In my previous experiment, both were OK. Now it should be list instead of tuple.

Sorry about the unclear ReadMe, I used Python 2.7 for this project. You may need do other modifications for Python 3 .

Please check think link: https://github.com/lanwuwei/SPM_toolkit/blob/master/PWIM/main_sts.py Actually STS is converted into classification task. All these tasks are very similar, so I skip some files in order to reduce redundancy.

Thanks for your suggestions :) I am shifting from python 2 to python 3 currently and making this kind of inconsistent problem. I will pay attention to it in the...

Hi, the config file just defines some file paths. Actually, I didn't use the information in config file, you can delete this line.

Sorry about the inconvenience, I just updated the config file: https://github.com/lanwuwei/SPM_toolkit/blob/master/SSE/config.py Best, Wuwei From: funnytestingcase Reply-To: lanwuwei/SPM_toolkit Date: Thursday, January 7, 2021 at 8:12 AM To: lanwuwei/SPM_toolkit Cc: Wuwei Lan...

Hi Shuailong, When you specify the embedding path and name, torchtext will automatically download data from GloVe website: https://nlp.stanford.edu/projects/glove/ , which is saved into your local machine. Best, Wuwei

just run this command: `wv_dict, wv_arr, wv_size = load_word_vectors('/your/local/path/', 'glove.840B', 300)` It will download glove.840B.300d.zip Note: the torchtext version in my code is 0.1.1

Hi caoxu, I tried your second answer in other code: https://github.com/lanwuwei/Subword-PWIM/blob/6e9ef11fa404c4ea6e350f8e75f7d1055869d8d9/model.py#L436 In my PyTorch version 0.4.0, both are OK. You can modify the code based on your version.

`parser.add_argument('--deep_CNN', type=bool, default=True, help='use 19 layer CNN or not')`