JonahNoah / Cross_Stance

Cross-Target stance detection

Cross_Stance

Cross-Target stance detection

Requirements:

• python==3.7.10

• numpy==1.19.2

• matplotlib==3.5.2

• pandas==1.2.4

• seaborn==0.11.2

• tqdm==4.59.0

• scikit-learn==0.24.2

• pytorch==1.13.1 (Cuda 11.7)

• transformers==4.28.0

• faiss-gpu==1.7.1 (conda install -c conda-forge faiss-gpu==1.7.1) - only for JointCL

• transformers==4.28.0 - only for llm

• sentencepiece==0.1.98 - only for llm

• accelerate==0.18.0 - only for llm

• bitsandbytes==0.38.1 - only for llm

• protobuf==3.20.0 - only for llm

• llama-cpp-python==0.1.48 - only for llama_cpp

• sentence-transformers==2.2.2

Training a model:

General syntax:

python train_model.py -m train -c [config_file] -t [train_data] -v [valid_data] -p [test_data] -n [model_name] -e [early_stopping] -s [save_checkpoints]

Example:

python train_model.py -m train -c ../../config/BiCondBertLstm_example.txt -t ../../data/UStanceBR/v2/hold1topic_out/final_bo_train.csv -v ../../data/UStanceBR/v2/hold1topic_out/final_bo_valid.csv -n bo -e 5 -s 1

Contact:

Matheus Camasmie Pavan [e-mail]