Recurrent neural networks (RNNs) have become increasingly popular in machine learning for handling sequential data. In this tutorial, we will cover background about the architecture, a toy training example, and a demo for evaluating a larger pre-trained model.
These materials are prepared for the Harvard-MIT Theoretical and Computational Neuroscience Journal Club and should be beginner-friendly!
The RNN multitask training repo is imported from https://github.com/benhuh/RNN_multitask.
Date: June 26, 2019
Authors: Jennifer Hu (MIT), Ben Huh (IBM), Peng Qian (MIT)
NOTE: If you are cloning this repository, make sure to use the flag --recurse-submodules or --recursive depending on your version of Git (see https://stackoverflow.com/a/4438292).
To run this tutorial, you will need to install Pytorch, numpy, and Jupyter. See the notebook demo.ipynb for more details.
To evaluate the pre-trained model, you will need to download the checkpoint from this link. Alternatively, you can run
wget https://www.dropbox.com/s/er9exdbwun4rex9/model_bnc.pt?dl=1Make sure to move the downloaded file to the materials folder.