lucapug / bert_classification

example of NLP tasks fairly well done without the use of Big ChatGPT like models

Text Classification Sample Project

Overview

This project demonstrates a simple binary text classification task using a sample dataset. It includes scripts to generate sample data (very short recipes and job descriptions) and to train a binary classification model on this data.

Scripts

1. Data Generation (gen_test_data.py): This script contains functions to generate sample data, creating two types of short texts: recipes and job descriptions. These texts are then combined into a pandas DataFrame with corresponding labels.

2. Model Training and Prediction (split_train_test_sample.py): This script trains a binary classification model using the Hugging Face Transformers library. It handles splitting the data, tokenizing it, training the model, evaluating its performance, and predicting classes for new texts.

How to Use

1. Data Generation:

   • Run gen_test_data.py to create a DataFrame of sample texts and their labels, which is shuffled and ready for model training.

2. Model Training:

   • Configure the model checkpoint and training arguments in split_train_test_sample.py.
   • Run the script to train the model. It splits the data, tokenizes the texts, and trains the classification model.
   • It also evaluates the model on the test set and outputs a confusion matrix and accuracy score.
   • Predictions and softmax probabilities are saved to a CSV file.

3. Prediction:

   • Use the predict_text function to predict the class of new text samples with the trained model.

Dependencies

Install the required packages from the requirements.txt file:

pip install -r requirements.txt