This project involves feature representation of text data, implementation of perceptron algorithm, as well as gradient descent in Python.
This repository has the following structure
.
├── data # Contains reviews dataset
│ ├── truecased_reviews_dev.jsonl
│ ├── truecased_reviews_dev_one.jsonl
│ ├── truecased_reviews_test.jsonl
│ ├── truecased_reviews_train.jsonl
├── models
│ ├── _init_.py
│ ├── gd.py # Gradient descent algorithm
│ ├── perceptron.py # Perceptron algorithm
├── readers
│ ├── _init_.py
│ ├── reviews_dataset.py # Load reviews dataset function
├── _init_.py
├── sentiment_classifier.py # Performs sentiment analysis
├── utils.py
├── images
└── README.md
The main python file is sentiment_classifier.py, where the bulk of the code is implemented. There, we can load the dataset (function implemented in reviews_dataset.py), generate feature representations, create an instance of our perceptron algorithm implementation (with functions implemented in the perceptron.py file), as well as a gradient descent algorithm (with functions implemented in the gd.py file), using the squared loss and L2-norm regulariser.
