assignment2.py

import argparse
import numpy as np

from utils import load_dataset
from model.model_utils import bag_of_words_matrix, labels_matrix
from model.ffnn import NeuralNetwork
from helper import batch_train, minibatch_train

DATA_PATH = './data/dataset.csv'


def main():
    parser = argparse.ArgumentParser(
        description='Train feedforward neural network'
    )

    parser.add_argument(
        '--minibatch', dest='minibatch',
        help='Train feedforward neural network with mini-batch gradient descent/SGD',
        action='store_true'
    )

    parser.add_argument(
        '--train', dest='train',
        help='Turn on this flag when you are ready to train the model with backpropagation.',
        action='store_true'
    )

    args = parser.parse_args()

    sentences, intent, unique_intent = load_dataset(DATA_PATH)

    ############################ STUDENT SOLUTION ####################
    # YOUR CODE HERE
    #     TODO:
    #         1) Convert the sentences and intent to matrices using
    #         `bag_of_words_matrix()` and `labels_matrix()`.
    #         2) Initiallize the model Class with the appropriate parameters
    
    # Convert sentences to bag-of-words matrix
    X = bag_of_words_matrix(sentences)
    
    # Convert intent labels to matrix
    Y = labels_matrix((intent, unique_intent))
    
    # Initialize the model
    input_size = X.shape[0]  # vocabulary size
    hidden_size = 150        
    num_classes = Y.shape[0] # number of unique intents
    model = NeuralNetwork(input_size, hidden_size, num_classes)
    ##################################################################

    if not args.minibatch:
        print("Training FFNN using batch gradient descent...")
        batch_train(X, Y, model, train_flag=args.train)
    else:
        print("Training FFNN using mini-batch gradient descent...")
        minibatch_train(X, Y, model, train_flag=args.train)


if __name__ == "__main__":
    main()