mootezsaad / csci6908-a1-pt2 Goto Github PK

The following repository contains the replication package to train an LSTM model for next word generation conditioned on literary genres.

• data: contains training, test and validation data in form of pickle files. It also contains vocab dictionary and the genres mapping. They can be downloaded from the following Google Drive folder: https://drive.google.com/drive/folders/1SaBaq1KmAvj9K-4a9vvsdxzp8iOniRYL?usp=sharing
• logs: Contains Tensorboard logs from my experiment.
• model2: This package contains the model implementation, data classes and utility functions for I/O operations and training.
• checkpoint: Contains the model's weights after training.
• Data_Preparation.ipynb: Jupyter notebook used to generate the data provided in the Google Drive folder.
• train_attn.py Main script that contains the logging, hyperparemeters values, and training/testing loops.
• M2Inference.ipynb: Jupyter notebook used for inference (text generation).

(Option) First step is to generate the data files. This an optional step as training files are already provided.
To train the model, execute the the following:

python train_attn.py