Portfolio Assignment 2

Note: M3 - Group Assignment 2 Deadline: Monday 17th of Feb

Corgi Pups Neon Image

Introduction

This exercise is designed to deepen your understanding and skills in modern deep learning techniques with a focus on classification tasks. In this assignment, you are required to fine-tune a transformer model for one of the following classification tasks:

  • Binary Classification
  • Multi-Class Classification
  • Multi-Label Classification

You can choose any dataset (from 🤗 Hugging Face Datasets, Kaggle, or your own collection) that is suitable for the classification task you wish to address.

Task Description

Your assignment involves the following key steps:

  1. Dataset Selection and Preprocessing

    • Choose a dataset appropriate for your chosen classification task (binary, multi-class, or multi-label).
    • Preprocess the data as needed (e.g., tokenization, handling class imbalance, etc.).

  2. Model Selection and Fine-Tuning

    • Use a transformer model (e.g., BERT, RoBERTa, DistilBERT, etc.) available from Hugging Face.
    • Fine-tune the model on your selected dataset using libraries such as Simple Transformers or Hugging Face Transformers.
    • Document your training process, including hyperparameter choices, evaluation metrics, and any challenges encountered.

  3. Evaluation

    • Evaluate your fine-tuned model using appropriate metrics (e.g., accuracy, F1-score, ROC-AUC for binary; precision, recall, etc.).
    • Provide a brief analysis of the model’s performance on the test set.

  4. Deployment (Optional)

    • Optionally, develop a simple web application using Streamlit or Gradio that allows users to input text and see classification predictions.
    • This step is meant to demonstrate how your model can be deployed in a real-world scenario.

  5. Upload to Hugging Face Hub

    • Once fine-tuning is complete, upload your model to the Hugging Face Hub and provide a link to the repository.
    • Ensure that your model card includes information about the dataset, training parameters, evaluation metrics, and intended use.

Data

  • You may utilize datasets from 🤗 Hugging Face, Kaggle, or create your own.
  • Ensure that you cite your data source appropriately in your documentation.

Delivery

  1. GitHub Repository

    • Create a repository containing your code, notebooks, and documentation.
    • Include a README.md with a brief description of your assignment, your approach, and instructions on how to run your code.

  2. Colab or Jupyter Notebook

    • Save your working notebook in the repository.
    • Ensure that the notebook includes detailed explanations of each step, from data preprocessing to model evaluation.

  3. Group Work

    • You may work in groups of up to 3 members.
    • Each group member’s contribution should be briefly outlined in the README or the notebook.

  4. Technical Explainer Video

    • Record a short (~5 minutes) technical explainer video presenting your main ideas, methodology, and results.
    • You may use Panopto, OBS Studio, Loom, or any other screen-recording tool.
    • Include the video link in your submission.

  5. Submission

    • Send an email to Hamid (hamidb@business.aau.dk) with the link to your GitHub repository (and video) by the deadline.

Good luck and have fun exploring classification with transformer models in PyTorch and Hugging Face!