Running the code

The code uses Python 3.8.10 which is the Python 3 version in the NUS SOC Compute Cluster. It would probably be useful to use an environment, such as through virtualenv, to run this code. The code assumes that the LUN dataset's csv files are in a folder called Data at the same level as the src folder.

Install the required libraries using pip. You can either simply run

pip install -r requirements.txt

or install tensorflow-gpu, pandas, numpy, sklearn, scipy, and nltk, one by one with pip.

To train and evaluate a model encapsulated by a wrapper, run

python -m src.main --model_wrapper=[MODEL_WRAPPER_ID] --dataset=[DATASET_ID]

Model wrapper and dataset IDs can be found in src/model_wrappers/__init__.py and src/data/__init__.py respectively.

Some other optional parameters for running the code are listed below:

• --mode=[train | test | full] is used to choose whether to only train, only test, or do both.
• --save_filepath=[SAVE_FILEPATH] is used to specify where to save the model to.
• --load_filepath=[LOAD_FILEPATH] is used to specify where to load the model from.

Reproducing the results in the report

The model wrappers used have IDs nltk_rnn for the recurrent neural network and attention for the attention-based neural network. The more general Kaggle-BUPT-GAMMA dataset used to train and do an initial evaluation has the ID kaggle_bupt_gamma and while the Singaporean media dataset used to do the final evaluation has the ID singapore_test.

To train and evaluate the model on the Kaggle-BUPT-GAMMA dataset, run

python -m src.main --model_wrapper=[MODEL_WRAPPER_ID] --dataset=kaggle_bupt_gamma --save_filepath=[FILEPATH]

and to evaluate it again on the Singapore media dataset, run

python -m src.main --mode=test --model_wrapper=[MODEL_WRAPPER_ID] --dataset=singapore_test --load_filepath=[FILEPATH]

where MODEL_WRAPPER_ID are either nltk_rnn or attention and FILEPATH can be any suitable filepath to save and load the model.

For example,

python -m src.main --model_wrapper=nltk_rnn --dataset=kaggle_bupt_gamma --save_filepath=checkpoints/nltk_rnn/checkpoint

python -m src.main --mode=test --model_wrapper=nltk_rnn --dataset=singapore_test --load_filepath=checkpoints/nltk_rnn/checkpoint

or

python -m src.main --model_wrapper=attention --dataset=kaggle_bupt_gamma --save_filepath=checkpoints/attention/checkpoint

python -m src.main --mode=test --model_wrapper=attention --dataset=singapore_test --load_filepath=checkpoints/attention/checkpoint