Neural_Network_Project

Introduction

This is our implementation of classical RNN networks, including Vanilla, GRU and LSTM models, versus the novel architecture proposed by Michael Rotman and Lior Wolf named Shuffling RNN.

Data

Classification Dataset

Our dataset is the IMDB datasaet. IMDB dataset having 50K movie reviews for natural language processing or Text analytics. This dataset is provided with a set of 50,000 highly polar movie reviews.

Regression Dataset

The dataset used in the study consists of a multivariate time series related to wheather conditions. Specifically, the time series consisted of four variables: humidity, pressure, temperature, and wind speed.

Libraries

• Pythorch
• Scikit Learn
• Pandas
• Matplotlib
• Numpy
• Torchtext

Usage

Our project is divided into 2 differents part, the simple Python project and the Jupiter Notebook project.

Python Project

Before to start install all the libraries listed before.

To start the program you have to be in the one of the 2 folders in the project, tiping cd ./Classification or cd ./Regression. In the folder simply type python main.py or python main.py.

Arguments

The main functions take various arguments. To use test the training of one neural network -t, to reload a saved neural network -r and to train all the neural networks to plot our experiments -i.

Test Aguments

If type -t you can use more arguments:

• -m for the model you want to train, followed by the initial letter of the possible models: rlsg
• -s to save the model
• -l to select the learning rate
• -p to plot the loss
• -e to select the number of epochs
• -b to select the batch size
• -d to choose the directory of the dataset
• -h to select the hidden size

##Jupiter Notebook Project

To start the program you just have to have Jupyter Notebook installed on your system. If you don't have Jupyter Notebook installed or use sites like Google Colab. Then you have you execute the Notebook_Neural_Network_Project.ipynb file.

Jupiter Notebook Structure

At the start of the Notebook there is an introduction about the method we used and the imported libraries.

The Notebook is divided in 2 parts, the classification and the regression one. Every of these 2 sections is divided in sections that are the following:

• Utils functions we used
• The RNNs used for the test and the implementation of the LSTM
• The functions to run the tests with paramiters
• The tests

Results

The results of our experiments are saved in the Results folder.