Python-pytorch Implementation of a dynamic memory network for a question answering system.
This Question Answering System is based on the paper "Ask Me Anything: Dynamic Memory Networks for Natural Language Processing" ( arXiv:1506.07285). The networks recieves facts as text input along with a question. The neural network architecture processes input facts and questions and forms episodic memories triggering an iterative attention process to form episodic memories and generate relevant answers.
- Pytorch (instructions)
- NumPy (instructions)
- OpenCV (instructions)
- Natural Language Toolkit (NLTK) (instructions)
- Pandas (instructions)
- Matplotlib (instructions)
Download the bAbI tasks (tasks_1-20_v1-2.tar.gz) from https://research.facebook.com/research/babi/.
The data preparation code can be found in the Data_prep.py
The model consists of 4 modules as shown in the diagram below.
The model architecture implemnted in pytorch can be found in the model.py