Skip to content

Latest commit

 

History

History
144 lines (102 loc) · 5.05 KB

README.md

File metadata and controls

144 lines (102 loc) · 5.05 KB
Raw

Lingualytics : Indic analytics with codemix support

Lingualytics is a Python library for dealing with indic text.
Lingualytics is powered by powerful libraries like Pytorch, Transformers, Texthero, NLTK and Scikit-learn.

Checkout our demo video!
Lingualytics demo

train-demo

🌟 Features

  1. Preprocessing

    • Remove stopwords
    • Remove punctuations, with an option to add punctuations of your own language
    • Remove words less than a character limit

  2. Representation

    • Find n-grams from given text

  3. NLP

    • Classification using PyTorch
      • Train a classifier on your data to perform tasks like Sentiment Analysis
      • Evaluate the classifier with metrics like accuracy, f1 score, precision and recall
      • Use the trained tokenizer to tokenize text

🧠 Pretrained Models

Checkout some codemix friendly models that we have trained using Lingualytics

💾 Installation

Use the package manager pip to install lingualytics.

pip install lingualytics

🕹️ Usage

Preprocessing

from lingualytics.preprocessing import remove_lessthan, remove_punctuation, remove_stopwords
from lingualytics.stopwords import hi_stopwords,en_stopwords
from texthero.preprocessing import remove_digits
import pandas as pd
df = pd.read_csv(
   "https://github.com/lingualytics/py-lingualytics/raw/master/datasets/SAIL_2017/Processed_Data/Devanagari/validation.txt", header=None, sep='\t', names=['text','label']
)
# pd.set_option('display.max_colwidth', None)
df['clean_text'] = df['text'].pipe(remove_digits) \
                                    .pipe(remove_punctuation) \
                                    .pipe(remove_lessthan,length=3) \
                                    .pipe(remove_stopwords,stopwords=en_stopwords.union(hi_stopwords))
print(df)

Classification

Currently available datasets are

from lingualytics.learner import Learner

learner = Learner(model_type = 'bert',
                model_name = 'bert-base-multilingual-cased',
                dataset = 'SAIL_2017')
learner.fit()

Custom Dataset

The train data path should have 3 files

  • train.txt
  • validation.txt
  • test.txt

Any file should have the text and label in a line, separated by a tab. Then change the data_dir to the path of your custom dataset.

Find topmost n-grams

from lingualytics.representation import get_ngrams
import pandas as pd
df = pd.read_csv(
   "https://github.com/jbesomi/texthero/raw/master/dataset/bbcsport.csv"
)

ngrams = get_ngrams(df['text'],n=2)

print(ngrams[:10])

Documentation | API Reference

Documentation is a work in progress! Have a look at it here.

Development Roadmap

We plan to add the following functionality in the coming weeks:

  • Language Identification (LID)
  • POS Tagging (POS)
  • Named Entity Recognition (NER)
  • Sentiment Analysis (SA)
  • Question Answering (QA)
  • Natural Language Inference (NLI)
  • Topic Modelling(LDA)
  • Fuzzy text matching at scale
  • Word Sense Disambiguation, TF-IDF , Keyword Extraction
  • data distribution over different languages

👪 Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

⚖️ License

MIT

📚 References

  1. Khanuja, Simran, et al. "GLUECoS: An Evaluation Benchmark for Code-Switched NLP." arXiv preprint arXiv:2004.12376 (2020).