Deep Learning Course Practical Assessment

This repository contains solutions for two practical challenges from the Deep Learning course assessment.

Overview

• Challenge 1: Image Classification A practical assessment on building, training, and evaluating deep learning models for image classification tasks. Kaggle link: https://www.kaggle.com/competitions/unipd-deep-learning-2025-challenge-1

• Challenge 2: Time-Series Modeling A practical assessment on building, training, and evaluating deep learning models for sequential (time-series) data. Kaggle link: https://www.kaggle.com/competitions/unipd-deep-learning-2025-challenge-2

Repository Structure

├── ensembleCNN.ipynb  # Notebook with solution for Challenge 1
├── weatherLSTM.ipynb           # Notebook with solution for Challenge 2
├── images_results/                                # Screenshots of model results (notebooks are inaccessible)
│   ├── challenge1_results.png
│   └── challenge2_results.png
└── README.md                              # This file

Notebooks

Note: For confidentiality reasons, the original challenges are not publicly accessible. Instead, this repository includes snapshots of key outputs.

• challenge1_image_classification.ipynb

  • Data loading and preprocessing
  • CNN architecture design and training logs
  • Evaluation: accuracy, confusion matrix, sample predictions

• challenge2_time_series.ipynb

  • Data preparation and windowing for sequence modeling
  • RNN/LSTM model design and training history
  • Evaluation: loss curves, forecast vs. ground truth plots

Results

13th place out of 135 participants

Challenge 1: Image Classification

• Final test accuracy: 0.81% the higher the best
• Best-performing model: Custom CNN with data augmentation and ensemble method of different models

Challenge 2: Time-Series Modeling

1th place out of 92 participants

• Forecast RMSE on test set: 0.94 the lower the best
• Best-performing model: LSTM with multiple type of regularization