homework 3

Homework 3

This homework has two parts:

1. Hyperparameter Tuning with W&B — use Weights & Biases to systematically calibrate your model’s hyperparameters and identify the best configuration.
2. TF Binding Prediction — implement the TF Binding Prediction Challenge using the starter notebook as your starting point.

Part 1: Hyperparameter Calibration with Weights & Biases

• Implement hyperparameter tuning using Weights & Biases (wandb)
• Choose one of the following problems to apply this to:
  • TF binding prediction
  • Simple DNA scoring
  • Cubic function with MLP
• Reference example: TF Binding with wandb
• Document the hyperparameters you tuned and their impact on model performance

Part 2: Model Deployment and Testing

• Upload your complete notebook containing:

  • Model architecture
  • Training code
  • Evaluation metrics
  • Prediction code

• Save your model weights using state dict (not the full model object):

  torch.save(model.state_dict(), 'model_weights.pt')

• Submit a self-contained predict.py script that:

  • Defines your model architecture (copy the class definition from your notebook)
  • Loads your saved weights via model.load_state_dict(torch.load('model_weights.pt', ...))
  • Accepts a sequences data path as argument and writes predictions to a file
  • Can be run as: python predict.py --data <path_to_sequences> --out <output_file>
  • See the starter notebook for a skeleton

Submission Requirements

1. Submit your Jupyter notebook with all code and documentation
2. Submit model_weights.pt (state dict) and predict.py
3. Document your hyperparameter tuning process and results
4. Provide clear instructions for running the prediction script

Evaluation Criteria

• Proper implementation of hyperparameter tuning
• Model performance on test set
• Code organization and documentation
• Reproducibility of results

© HakyImLab and Listed Authors - CC BY 4.0 License