Improve Data Preprocessing Speed - Regex in Python

Question

I use the following class in Python to preprocess a string before passing it to a machine learning classification model for predicting its sentiment.

I use regex for most of the transformation along with some libraries like emoji and tweet-preprocessor. The code works fine but I believe that it is slow.

Do you have any suggestions on how to improve its speed?

Example of usage:

string  = "I am very happy with @easyjet #happy customer 🙂. Second sentence"
preprocessor = TextPreprocessing()
result = preprocessor.text_preprocessor(string)

The result will be : ["i am very happy with happy smiling face", "second sentence", "i am very happy with happy smiling face second sentence"]

import re
import preprocessor as p   # this is the tweet-preprocessor library
import emoji
import os
import numpy as np
import pandas as pd

class TextPreprocessing:
    def __init__(self):
        p.set_options(p.OPT.MENTION, p.OPT.URL)

    # remove punctuation
    def _punctuation(self, val):
        val = re.sub(r'[^\w\s]', ' ', val)
        val = re.sub('_', ' ', val)
        return val

    #remove white spaces
    def _whitespace(self, val):
        return " ".join(val.split())

    #remove numbers
    def _removenumbers(self, val):
        val = re.sub('[0-9]+', '', val)
        return val

    #remove unicode
    def _remove_unicode(self, val):
        val = unidecode(val).encode("ascii")
        val = str(val, "ascii")
        return val

    #split string into sentenses
    def _split_to_sentences(self, body_text):
        sentences = re.split(
            r"(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s", body_text)
        return sentences

    # cleaning functions that combines all of the above functions
    def _clean_text(self, val):
        val = val.lower()
        val = self._removenumbers(val)
        val = p.clean(val)
        val = ' '.join(self._punctuation(emoji.demojize(val)).split())
        val = self._remove_unicode(val)
        val = self._whitespace(val)
        return val

    def text_preprocessor(self, body_text):
        body_text_df = pd.DataFrame({"body_text": body_text}, index=[1])
        sentence_split_df = body_text_df.copy()
        sentence_split_df["body_text"] = sentence_split_df["body_text"].apply(
            self._split_to_sentences)

        lst_col = "body_text"
        sentence_split_df = pd.DataFrame(
            {
                col: np.repeat(
                    sentence_split_df[col].values, sentence_split_df[lst_col].str.len(
                    )
                )
                for col in sentence_split_df.columns.drop(lst_col)
            }
        ).assign(**{lst_col: np.concatenate(sentence_split_df[lst_col].values)})[
            sentence_split_df.columns
        ]

        final_df = (
            pd.concat([sentence_split_df, body_text_df])
            .reset_index()
            .drop(columns=["index"])
        )

        final_df["body_text"] = final_df["body_text"].apply(self._clean_text)

        return final_df["body_text"]

This question might be relevant to all those Data Scientists who want to move their NLP models into production.

Answer 1

Since I cannot comment I will try to answer your question (to some extent):

1. You should clarify how to measure the execution time improvement. Use timeit and its repeat functionality for that:

import timeit
from functools import partial
...
if __name__ == "__main__":
    # http://25.io/toau/audio/sample.txt
    with open("sample.txt") as f:
        text = f.read()
        tp = TextPreprocessing()
        print(min(timeit.Timer(partial(tp.text_preprocessor, text)).repeat(repeat=10, number=1)))

You can also use timeit on specific methdos to check for bottlenecks.

2. Sadly I could not run your code sample due to the undefined np. in L58 and L64 so I cannot test my assumptions. Also you did not provide sample data.

3. Some general thoughts:

• Use re.compile() to compile all of your regular expressions
• If you do not need the modularity of _remove* you might consider combining these regular expressions
• .copy() operations are expensive try to get rid of them
• Also some of the _remove* methods have a smell. See the links for alternatives:

  • Remove zero width space unicode character from Python string

  • Removing numbers from string