garyyufei / promda Goto Github PK

Sorry to bother you, I'm very interested in your work and have read most of the code of the whole project. I have replicated your sentence classification experiment, but the results are not good. Did I do something wrong? The detailed results are as follows.

seed baseline-shot-10 PromDA-shot-10
18 | 76.66 | 78.14
45 | 62.66 | 69.3
76 | 75.56 | 69.96
256 | 70.51 | 73.37
512 | 68.15 | 69.69
Avg | 69.09 | 73.03

您好，您在论文中提到的是pretrain和fine-tune阶段都使用1e-3作为学习率，但是在复现的时候，使用1e-3来finetune的效果很差，而使用0.3的效果要好一些，请问您实验时的最优学习率是多少呢？
感谢。

I'm sorry to bother you. I'm very interested in your work, and I'm particularly interested in getting the source code of NRETM. Thank you very much.

Sorry to bother you .I have some question about running to script/run_few_shot_bert_prefix_sen_cls.sh at line 61,the train_SenLabel.py will raise "IndexError: list index out of range" at line 162.I check the detail and find that "test_data = SenLabelDataset(_C, _C.test_path, label_dict, tokenizer)" prints "Total instances 0"(But the line 154 "dev_data = SenLabelDataset(_C, _C.dev_path, label_dict, tokenizer)" works well,it prints "Total instances 960")

你好，您在论文中说采样的shot-K是指每种label都有K个instances，但是代码运行结果似乎是一共采样K个instances。应该是哪种方式呢？

Sorry to bother you, where is the NLU_training_dataset folder? Does that mean the datasets folder? So,what does the "subj" in "test_path:NLU_training_dataset/subj/test_whole.txt" in the path of "model_config/rt/bert_nlu.yml" mean? There is no "subj" in datasets folder.

Hi, I tried to set up the conda environment using the command mentioned in README. However, I am getting the following error.

conda create --name promda --file requirements.txt
Collecting package metadata (current_repodata.json): done
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed

PackagesNotFoundError: The following packages are not available from current channels:

  - scikit-learn==1.0=pypi_0
  - pillow==8.3.2=pypi_0
  - filelock==3.3.0=pypi_0
  - charset-normalizer==2.0.6=pypi_0
  - torchvision==0.10.1+cu111=pypi_0
  - regex==2021.10.8=pypi_0
  - portalocker==2.3.2=pypi_0
  - sacremoses==0.0.46=pypi_0
  - torch==1.9.1+cu111=pypi_0
  - numpy==1.21.2=pypi_0
  - zipp==3.6.0=pypi_0
  - urllib3==1.26.7=pypi_0
  - six==1.16.0=pypi_0
  - idna==3.2=pypi_0
  - tqdm==4.62.3=pypi_0
  - yacs==0.1.8=pypi_0
  - tokenizers==0.10.3=pypi_0
  - scipy==1.7.1=pypi_0
  - threadpoolctl==3.0.0=pypi_0
  - huggingface-hub==0.0.19=pypi_0
  - pyparsing==2.4.7=pypi_0
  - pyyaml==5.4.1=pypi_0
  - packaging==21.0=pypi_0
  - rake-nltk==1.0.6=pypi_0
  - typing-extensions==3.10.0.2=pypi_0
  - click==8.0.3=pypi_0
  - seqeval==1.2.2=pypi_0
  - transformers==4.10.2=pypi_0
  - sentencepiece==0.1.96=pypi_0
  - colorama==0.4.4=pypi_0
  - fast-bleu==0.0.89=pypi_0
  - tabulate==0.8.9=pypi_0
  - joblib==1.1.0=pypi_0
  - nltk==3.6.5=pypi_0
  - requests==2.26.0=pypi_0
  - torchaudio==0.9.1=pypi_0
  - sacrebleu==2.0.0=pypi_0
  - importlib-metadata==4.8.1=pypi_0

Current channels:

  - https://repo.anaconda.com/pkgs/main/linux-64
  - https://repo.anaconda.com/pkgs/main/noarch
  - https://repo.anaconda.com/pkgs/r/linux-64
  - https://repo.anaconda.com/pkgs/r/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

I have tried setting up on 2 different machines, and getting the same error.
Also tried resolving this error using some stackoverflow threads like https://stackoverflow.com/a/70775641/11730679 and https://stackoverflow.com/a/48509426/11730679. This resolved the issue to some extent, but I still get the following packages cannot be found:

fast-bleu=0.0.89
huggingface-hub==0.0.19
torch==1.9.1+cu111
torchvision==0.10.1+cu111
rake-nltk==1.0.6
torchaudio==0.9.1

Can you please provide a fix for the same.
Thanks