|
# Language Identifier |
|
--- |
|
|
|
### OVERVIEW |
|
This project is a Flask web application that identifies the language of input text. It uses a machine learning model trained on text data to make predictions. The user inputs text into a form on the web app, and the app returns the predicted language. |
|
|
|
### SPECIFICATIONS |
|
- The data used for training is taken from Kaggle. It has 22 different languages. |
|
- The text in the dataset has tokenization, non alphanumeric characters removal and vectorization applied to it. |
|
- The model used for training has 4 layers with 27M params which is enough for getting high accuracy. Complex architectures couldn’t be used because of not sufficient GPUs. |
|
- Techniques like early stopping, learning rate decay and weight decay are used while training to get the most accurate results. |
|
- The metrics used for evaluation is accuracy, 97.89% of which is achieved. |
|
- I usually use Pytorch but this time I used Tensorflow because converting tokens into tensors crashed the GPU constantly. |
|
- The project uses Flask, a lightweight web framework for Python, to create the web application. |
|
- The input text is preprocessed before being fed into the model for prediction. |
|
|
|
### USAGE |
|
```python |
|
def predict_language(text, model, cv, le): |
|
cleaned_text = clean_text(text) |
|
text_vectorized = cv.transform([cleaned_text]) |
|
prediction = model.predict(text_vectorized) |
|
predicted_label = le.inverse_transform([np.argmax(prediction)])[0] # Get the first element of the list |
|
return predicted_label |
|
sentence = 'random text' |
|
predicted_label = predict_language(sentence, model, cv, le) |
|
print(predicted_label) |
|
|