python==3.8.16 torch==1.12.0+cu116 torch-geometric==2.3.0
python preprocess.py --data './data/iemocap/newdata.pkl' --dataset="iemocap"
python train.py --wf -10 --wp -10 --data './data/iemocap/newdata.pkl' --from_begin --device=cuda --epochs=80 --batch_size=20 --n_speakers 2
python prediction.py --data=./data/iemocap/newdata.pkl --device=cuda --epochs=1 --batch_size=20 --n_speakers 2
-|Dataset|Weighted F1(w) | Acc :-:|:-:|:-: Original|IEMOCAP|70.22% | 70.12%
python preprocess_roberta.py --data './data/meld/newdata.pkl' --dataset="meld"
python train.py --wf -10 --wp -10 --data './data/meld/newdata.pkl' --device=cuda --epoch 80 --from_begin --batch_size=20 --n_speakers 9
python prediction.py --data=./data/meld/newdata.pkl --device=cuda --epochs=1 --batch_size=20 --n_speakers 2
-|Dataset|Weighted F1(w) | Acc :-:|:-:|:-: Original|MELD|65.18% | 66.21%
The structure of our code is inspired by pytorch-DialogueGCN-mianzhang.