vijaydwivedi75 / emotion-gnn

Repository for CE7455 2019/2020

INSTALLATION

1. Setup Conda

# Conda installation

# For Linux
curl -o ~/miniconda.sh -O https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh

# For OSX
curl -o ~/miniconda.sh -O https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh

chmod +x ~/miniconda.sh    
./miniconda.sh  

source ~/.bashrc          # For Linux
source ~/.bash_profile    # For OSX

2. Setup Python environment for CPU

# Clone GitHub repo
git clone https://github.com/vijaydwivedi75/emotion-gnn.git
cd emotion-gnn

# Install python environment
conda env create -f env_cpu.yml   

# Activate environment
conda activate emotion_gnn

Then install Pytorch Geometric using the following commannds.

python -m pip install torch-scatter==latest+cpu -f https://pytorch-geometric.com/whl/torch-1.4.0.html  
python -m pip install torch-sparse==latest+cpu -f https://pytorch-geometric.com/whl/torch-1.4.0.html  
python -m pip install torch-cluster==latest+cpu -f https://pytorch-geometric.com/whl/torch-1.4.0.html  
python -m pip install torch-spline-conv==latest+cpu -f https://pytorch-geometric.com/whl/torch-1.4.0.html  
python -m pip install torch-geometric  

3. Reproduce results of the DialogueGCN paper in terminal

python pytorchgeometric/train.py --graph-model --no-cuda

4. Fine-tune BERT-base model on IEMOCAP dataset

python fine_tune_bert.py

5. Generate the text embedding and export to pickle file

python generate_text_embedding.py

6. Change line #40 in data/iemocap.py to load the newly generated pickle file, eg:

self.testVid = pickle.load(open('data/IEMOCAP_features_BERT_FT.pkl', 'rb'), encoding='latin1')

7. Run specific 4-layer GNN model on the newly constructed graphs

Note: For now, sequential encoding is not being used

python main_emotion_node_classification.py --config configs/emotion_node_classification_MLP_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_MLP_GATED_IEMOCAP.json   
python main_emotion_node_classification.py --config configs/emotion_node_classification_GCN_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_GAT_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_GraphSage_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_GIN_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_MoNet_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_GatedGCN_IEMOCAP.json  
python main_emotion_node_classification.py --config configs/emotion_node_classification_GatedGCN_E_IEMOCAP.json