You first need to add the Comet2020 common sense relations. You need to download the model and set the comet_model_path to it.
python atomic_augmentation/create_augmented_dataset.py --comet_model_path <path to comet model> --save_dataset_path <path to dictionary>
Then filter out unrelated common sense. we look for the common sense relations that are extracted from observation_2 but compatible with observation_1 and vice versa.
python atomic_augmentation/filtered_aug_dataset.py --input_dataset "/media/sdc/rohola_data/abduction_augmented_dataset --output_dataset "/media/sdc/rohola_data/selected_augmented_dataset
After creating the dataset you can run the training using, you need to set the dataset path created in the previous sections in abduction_augmented_config.json:
python nlg/abduction_atomic_augmented_training.py
Finally, after training you can generate abducgive explanations using: you need to check the checkpoint path in abduction_augmented_config.json:
python nlg/abductive_augmenetd_generation.py
dataset_name
/media/sdb4Tb/rohola_data/filtered_abduction_augmented_dataset
checkpoint
/media/sdb4Tb/rohola_data/abduction_aug_generation_checkpoint