paranormal-or-skeptic-ISI-public/main.py

from nltk.util import pr
import pandas as pd
import numpy as np
import torch
from gensim import downloader
from nltk.tokenize import word_tokenize
import csv

BATCH_SIZE = 5


class NeuralNetworkModel(torch.nn.Module):

    def __init__(self):
        dim = 200
        super(NeuralNetworkModel, self).__init__()
        self.fc1 = torch.nn.Linear(dim, 500)
        self.fc2 = torch.nn.Linear(500, 1)

    def forward(self, x):
        x = self.fc1(x)
        x = torch.relu(x)
        x = self.fc2(x)
        x = torch.sigmoid(x)
        return x


def read_data():
    x_labels = (pd.read_csv('in-header.tsv', sep='\t')).columns
    y_labels = (pd.read_csv('out-header.tsv', sep='\t')).columns

    x_train = pd.read_table('train/in.tsv', error_bad_lines=False,
                            header=None, quoting=csv.QUOTE_NONE, names=x_labels)
    y_train = pd.read_table('train/expected.tsv', error_bad_lines=False,
                            header=None, quoting=csv.QUOTE_NONE, names=y_labels)
    x_dev = pd.read_table('dev-0/in.tsv', error_bad_lines=False,
                          header=None, quoting=csv.QUOTE_NONE, names=x_labels)
    x_test = pd.read_table('test-A/in.tsv', error_bad_lines=False,
                           header=None, quoting=csv.QUOTE_NONE, names=x_labels)

    # remove some rows for faster development
    remove_n = 200000
    drop_indices = np.random.choice(x_train.index, remove_n, replace=False)
    x_train = x_train.drop(drop_indices)
    y_train = y_train.drop(drop_indices)

    return x_labels, y_labels, x_train, y_train, x_dev, x_test


def process_data(x_labels, y_labels, x_train, y_train, x_dev, x_test):
    x_train = x_train[x_labels[0]].str.lower()
    x_dev = x_dev[x_labels[0]].str.lower()
    x_test = x_test[x_labels[0]].str.lower()
    y_train = y_train[y_labels[0]]

    x_train = [word_tokenize(x) for x in x_train]
    x_dev = [word_tokenize(x) for x in x_dev]
    x_test = [word_tokenize(x) for x in x_test]

    w2v = downloader.load('glove-wiki-gigaword-200')

    x_train = [np.mean([w2v[w] for w in d if w in w2v] or [
        np.zeros(200)], axis=0) for d in x_train]
    x_dev = [np.mean([w2v[w] for w in d if w in w2v]
                     or [np.zeros(200)], axis=0) for d in x_dev]
    x_test = [np.mean([w2v[w] for w in d if w in w2v]
                      or [np.zeros(200)], axis=0) for d in x_test]

    return x_train, y_train, x_dev, x_test


def predict(model, x_data, out_path):
    y_out = []
    model.eval()
    with torch.no_grad():
        for i in range(0, len(x_data), BATCH_SIZE):

            x = x_data[i:i+BATCH_SIZE]
            x = torch.tensor(x)
            pred = nn_model(x.float())

            y_pred = (pred > 0.5)
            y_out.extend(y_pred)

    y_data = np.asarray(y_out, dtype=np.int32)
    pd.DataFrame(y_data).to_csv(out_path, sep='\t', index=False,  header=False)


if __name__ == "__main__":

    x_labels, y_labels, x_train, y_train, x_dev, x_test = read_data()

    x_train, y_train, x_dev, x_test = process_data(
        x_labels, y_labels, x_train, y_train, x_dev, x_test)

    nn_model = NeuralNetworkModel()
    criterion = torch.nn.BCELoss()
    optimizer = torch.optim.SGD(nn_model.parameters(), lr=0.1)

    for epoch in range(5):
        nn_model.train()

        for i in range(0, y_train.shape[0], BATCH_SIZE):
            X = x_train[i:i+BATCH_SIZE]
            X = torch.tensor(X)
            Y = y_train[i:i+BATCH_SIZE]
            Y = torch.tensor(Y.astype(np.float32).to_numpy()).reshape(-1, 1)

            Y_predictions = nn_model(X.float())

            loss = criterion(Y_predictions, Y)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

    predict(nn_model, x_dev, 'dev-0/out.tsv')
    predict(nn_model, x_test, 'test-A/out.tsv')