Neural Networks Project Collection

Overview

This repository contains implementations of various neural network projects, each utilizing a different type of neural network architecture. The projects cover a range of tasks, from image generation to text classification, demonstrating the versatility of neural networks in solving diverse problems.

---

Project List

1. Variational Autoencoder (VAE): Traffic Sign Generation

• Task: Generate synthetic images of traffic signs.
• Dataset: Provided as trafic_32.zip (similar structure to ImageFolder).
• Output: 1,000 generated images saved as a PyTorch tensor.
• Evaluation: Frechet Inception Distance (FID) metric.

2. Long Short-Term Memory (LSTM): Composer Prediction

• Task: Predict the composer of classical music based on chord sequences.
• Dataset: Provided in train.pkl (training data) and test_no_target.pkl (test data).
• Classes: Bach, Beethoven, Debussy, Scarlatti, Victoria.
• Special Requirements:
  • Use padding or PackedSequence for variable-length sequences.
  • Output predictions in a CSV file without headers or indices.

3. Feedforward Neural Network (FNN): Housing Price Classification

• Task: Classify properties as cheap, average, or expensive.
• Dataset: Training data in train_data.csv and test data in test_data.csv.
• Output: CSV with predicted class for each test instance (0: cheap, 1: average, 2: expensive).
• Special Notes:
  • Handle class imbalance.
  • Ensure predictions file is correctly formatted with no extra rows or columns.

4. Convolutional Neural Network (CNN): Image Classification

• Task: Classify images into 50 classes of objects and animals.
• Dataset: Training set with subfolders per class and a flat test set.
• Output: CSV with filenames and corresponding predicted classes.
• Special Notes:
  • Use custom architectures (do not use pretrained models).
  • Include two columns in the predictions CSV: filename and predicted class index.

5. BERT: Hotel Review Classification

• Task: Predict star ratings (0 to 4) based on hotel reviews.
• Dataset: Provided training set.
• Output: CSV with predicted ratings for the test set.
• Special Notes:
  • Tackle this as a multi-class classification problem.
  • Ensure predictions match the order of test samples.

---