This repository contains the PyTorch implementation and the data of the paper: CLIPMulti. AAA, BBB, CCC, DDD, EEE. Computer Speech & Language.[paper].

• Submit paper
• Prepare code and push

This codebase was developed and tested with the following settings:

scikit-learn==1.2.1
numpy==1.24.2
pytorch==2.0.1
torchvision==0.15.2
tqdm==4.64.1
clip==1.0
transformers==4.27.1
regex==2022.10.31
ftfy==6.1.1
pillow==9.4.0

CLIP:

pip install git+https://github.com/openai/CLIP.git

CLIPMulti

run main.py:

python main.py --dataset [dataset] --combination [combination] --prompt [prompt] --num_of_word [number_of word] --topk [number_of_topk] --test

[dataset] in ['topic', 'emotion', 'situation', 'agnews', 'snips', 'trec', 'subj']

[combination] in ['NSC', 'NWSC', 'MF']
dataset [subj] only use ['MF']

[prompt] in [0, 1, 2, 3]
prompt_0 is text written by humans based on labels
other is Match-CLIPMulti generation

[number_of word] in [30, 40, 50]
prompt_1 have [30, 40, 50], prompt_2 and prompt_3 only have [30]

[number_of_topk] in [1, 2, 3]

Match-CLIPMulti

run label2text_main.py:

python label2text_main.py --prompt [prompt] --num_of_word [number_of word] --topk [number_of_topk]

[prompt] in [1, 2, 3]

[number_of word] in [30, 40, 50]
prompt_1 have [30, 40, 50], prompt_2 and prompt_3 only have [30]

[number_of_topk] in [1, 2, 3]