Adaptive Temporal Convolutional Neural Network for Language models

Overview

• This repository contains the experiments that were proposed in the paper (AdaTCN)
• In this repository, a modified Temporal Convolutional Network (TCN) is presented, where the selection criteria of dilated convolutions that are used in TCN are adaptive.
• our code is trained and evaluated in a word-level language model using Penn TreeBank (PTB) and WikiText-02 datasets.
• Experiments are coded in PyTorch

Requirements

• you can install requirements by

pip3 install -r requirements.txt 

##Usage

• To run the code in a word-level scenario For PTB dataset, please use the following code:

python3 /PTB_word_level/text_penn.py

For WT2 dataset, please use the following code

python3 /PTB_word_level/text_wt2.py

-to use different hyperparameters, please use the args as presented in text_penn.py and text_wt2.py