COM 3031 Computational Intelligence Coursework

This repository contain the code implementation for COM 3031 Computational Intelligence assignment. Using Pytorch for implementation of a deep neural network for image classification task on CIFAR-10 dataset and compare a result of a model trained on different types of optimizers that are

• Gradient based optimizaton (gradient descent)
• Metaheuristic optimization (GA, PSO and Memetic)

Metaheuristic optimizer was implemented using Deap library.

Grade: 93/100.

Report : Link to report folder

Memeber of a groups

1. Wish Suharitdamrong

2. Taimoor Rizwan

3. Ionut Boston

Installation

There are two ways of installing package using conda

1.Create virtual conda environment from environment.yml

2.Use conda and pip to install a pakages

1.Create Virtual Environment from environment.yml

# Create virtual environment from .yml file
conda env create -f environment.yml

# activate virtual environment
conda activate ci 

2.Create Virtual Environment

Use Anaconda virtual environment to install all dependencies

# create virtual environment
conda create -n ci 

# activate virtual environment
conda activate ci 

Use anaconda to install Pytorch

For MacOs (only CPU)

conda install pytorch torchvision torchaudio -c pytorch

For Linux (Both GPU and CPU)

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

Use pip to install other packages

pip install -r requirement.txt

Directory

Pretrain weight directory

├── ckpt # checkpoint of a pretrain model
│   └── CIFAR-10_GD_SGD.pth

Run

Simple training

python train_gd.py # Use gradient descent to train whole network
python train_ga.py # Use genetic algorithms to train whole network
python train_pso.py # Use particle swarm optimization to train whole network

Two stage training

python train_hybrid_*.py # All the fiels start with train_hybrid are two stage training
python train_meme.py # Train two stage with memetic algorithms
python train_nsga.py # Train two stage with NSGA II algorithms

Visualization

Visualize result in jupyter notebook

jupyter notebook plot.ipynb

Visualization the results in Tensorboard

Tensordboard --logdir=../CI_logs