You should install OpenPrompt https://github.com/thunlp/OpenPrompt
First, Predict initial pseudo-label for the target domain from source domain data.
Second, Adjust initial target domain data pseudo-label, divide target domain data and iterative training model.
Then, Adjust all target domain data pseudo-label and voted invariant label.
Final, Average the experimental results three times.
Note that the file paths should be changed according to the running environment.
example shell scripts:
- python fewshot.py --result_file ./output_fewshot.txt --dataset newsgroups1 --template_id 0 --seed 144 --shot 20 --verbalizer manual
- python adj_data.py
- python div_data.py
- python itera_model.py
- python adj_label.py
- python voted_label.py
- python final_test.py