Assignment_1 - Deep Representation Learning.

Author: Sebastian Alejandro Velasco Dimate

CIPHAR10 dataset was used to train 2 models:

• Conv2d -> 3 Conv2x2 and 2 FCL.
• ResNet -> ResNet18 architecture with basic block of 2 Conv3x3.

Both architectures were trained usinng pytorch Distributed Data Parallel (DPP)

• conv2d-DPP.py : Conv2d architecture with DataParallel
• resnet-18-DPP.py: ResNet18 architecture with DataParallel

Wandb was used to log the total accuracy and loss metrics, as well as local acuracy, precision, recall and F1 for each CIPHAR10 class.

Prerequeisites - Deep Representation Learning.

• Conda
• Mamba
• PyTorch
• NVIDIA GPU (gloo)

Installation

• create virtual environment:

  mamba create -n assign01

• Launch virtual environment:

  conda activate assign01

• Isntall dependencies>

  pip install -r requirements.txt

Run the models

• Execute the follwing commands:

  nohup python3 conv2d-DPP.py > conv2d-DPP.txt &

  nohup python3 resnet-18-DPP.py > resnet-18-DPP.txt &