Repository for LING573 coursework.
Command Pipeline: python evaluate_reddit.py # to generate humor scores for reddit data python normalize.py # to generate normalized upvote scores for reddit data python run_correlation.py # to generate output & results
Spearman correlations and plot figures will be output to: ../results/D4/ Outputs of reddit data with predicted humor scores will be sent to: ../outputs/D4/
To Run: python FFNN.py
To re-extract pretrained embeddings: python FFNN.py --extract_pretrained
Pretrained embeddings are saved in the 'src/saved' directory. Pretrained BERT files are here (must be placed in the 'src' directory for use): https://drive.google.com/drive/folders/1n2QVlrKonIDcNHAD6PlzopQ4EALdcP-z?usp=sharing
The output of our hyperparameter testing is directed to the file locations: ../results/D3_scores.out & ../outputs/D3/{outputfile_name_for_hyperparameter_test}.out
Best model RMSE & hyperparameters will be printed to the bottom of ../results/D3_scores.out
To Run:
Install dependencies: pip install -r requirements.txt
To rerun training: python main.py --train
To run to and generate output with scores: python main.py
The output is directed to the console and file locations: ../results/D2_scores.out & ../outputs/D2/output.csv
Due to github;s limitations on file size, we are not able to host the pretrained files in this repo.
The pretrained files can be found hosted on our google drive here: https://drive.google.com/drive/folders/1n2QVlrKonIDcNHAD6PlzopQ4EALdcP-z?usp=sharing
Alternatively, to generate the files yourself, you can rerun the training prcodeure as outlined above.