Source Code Structure

python 3.7+, pytorch version: 1.5.1

• /data/ contains preprocessed dataset
• glove_embedding.py is the dataset prerprocessing script.
  • We have completed preprocessing, so you don't need to run it
• dataset.py contains the custom dataset iterator
• model.py contains the 4 model implementation:
  • 0 - NetAB - CNN+CNN
  • 1 - NetTrans - Transformer + Transformer
  • 2 - NetTransAB - CNN + Transformer
  • 3 - NetTransBA - Transformer + CNN
• main.py is the core training script.

Reproduction Instruction

• To run the script, you must first clone the code & data from the git, then create an empty "/ckpt/" folder

• Then you can reproduce all the result in the assignment with the following command:

python main.py --model 0 --lr 0.001
python main.py --model 1
python main.py --model 2
python main.py --model 3

• The default dataset is 'movie'. You can modify it by --dataset parameter, e.g.:

python main.py --model 1 --dataset restaurant

• You can choose among movie, laptop, restaurant.
• More options, please refer to the main.py.

===========================