Multi-Task Learning for Emotion Recognition in Conversation

Getting started

On Colab

• (Make sure you allow Google Colab to access this private repo)
• Execute Runtime/Run All
• Customize training arguments in kwargs dictionary
  • Change do_train to False to only run inference and evaluation.

Condor or Locally

• Install conda
• Clone this project recursively (since data folder is a submodule) and change to the project's root directory
• conda env create -f environment.yaml to initialize conda environment
• Put the downloaded the pytorch_model.bin model file from Google Drive as pytorch_model.bin in outputs/D4

Condor

• condor_submit D4.cmd to run inference and evaluation
• condor_submit D4_full.cmd to run training and evaluation

Locally

• python src/pipeline.py to run inference and evaluation
• python src/pipeline.py --do_train to run training and evaluation
• You can find more flags via python3 src/pipeline.py -h