Jasser-Chtourou/Language-Identification-Model

Language Detection Model

This repository contains a PyTorch-based model for language identification using multiple language detection methods. It combines predictions from various language detection libraries and models to determine the most probable language for a given text input.

Overview

The LanguageIdentificationModel integrates scores from several language detection methods:

LangDetect: Language detection using the langdetect library. LangID: Language identification with the langid library. Hugging Face: Language classification using the papluca/xlm-roberta-base-language-detection model from Hugging Face Transformers. FastText: Language prediction using the lid.176.bin model from FastText. These methods are integrated into a PyTorch neural network model, enabling accurate language identification across various text inputs.

---

language:

• en
• ar
• fr
• es
• pt
• ja
• it
• de
• ru
• zh metrics:
• accuracy
• code_eval library_name: transformers pipeline_tag: text-classification

Installation

Clone this repository: git clone https://github.com/jasserchtourou/Language_detection.git cd Language_detection

Install the required dependencies: pip install -r requirements.txt

Download the necessary models: Download the papluca/xlm-roberta-base-language-detection model from Hugging Face Transformers. Download the lid.176.bin model from FastText and place it in the repository.

Usage

Training the Model To train the model, customize the input_size, hidden_size, and output_size in train.py based on your data and run: python train.py

Using the Model

After training, you can use the model for language identification: from language_identification_model import LanguageIdentificationModel

model = LanguageIdentificationModel(input_size, hidden_size, output_size)
model.load_state_dict(torch.load('model_weights.bin'))
model.eval()
text = "Bonjour tout le monde"
language = model.predict(text)
print(f"The identified language is: {language}")

Contributing

Contributions are welcome! If you have any suggestions, improvements, or bug fixes, please submit a pull request or open an issue.

License

This project is licensed under the MIT License - see the LICENSE file for details.