convert.py

import csv
import random

from io import StringIO

import pandas as pd

from bs4 import BeautifulSoup


def remap_polarity(row):
    if row["label"] < 2:
        row["label"] = 0
    else:
        row["label"] = 1
    return row


def run():
    rows = []

    with open("training.1600000.processed.noemoticon.csv", "rb") as file:
        lines = [line[:-1] for line in file.readlines()]

    for line in lines:
        try:
            data = StringIO(line.decode("utf-8"))
        except UnicodeDecodeError:
            data = StringIO(line.decode("latin-1", errors="ignore"))

        reader = csv.reader(data, delimiter=",")
        for row in reader:
            bs = BeautifulSoup(row[5], "lxml")
            obj = {"label": int(row[0]), "text": bs.get_text()}
            rows.append(obj)

    df = pd.DataFrame(rows)
    df.to_csv("complete.csv", index=False, encoding="utf-8")

    polarity_rows = [remap_polarity(row) for row in rows if row["label"] != 2]
    positive_rows = [row for row in polarity_rows if row["label"] == 1]
    negative_rows = [row for row in polarity_rows if row["label"] == 0]
    min_size = min(len(positive_rows), len(negative_rows))
    polarity_rows = positive_rows[:min_size] + negative_rows[:min_size]

    random.shuffle(polarity_rows)

    df = pd.DataFrame(polarity_rows)
    df.to_csv("polarity.csv", index=False, encoding="utf8")


if __name__ == "__main__":
    run()