Pytorch-Tutorial

Pytorch tutorial with jupyter notebook

Installing prerequisites

• pip install jupyter
• pip install tqdm
• torch and torchvision. Follow instructions in https://pytorch.org/get-started/locally/.

How to run this tutorial

1. Clone this repository to any folder.
2. Open your favorite shell, and cd to that folder.
3. Run jupyter notebook
4. Shift+Enter through each notebook and cell.

Table of Contents

1. Tensors and Dynamic Graphs
   • Basic Operations involving torch.Tensor
   • Introduction to the dynamic computational graphs of torch.Autograd
2. Modules and Custom Models
   • Adding predefined modules from torch.nn to torch.nn.Sequential
   • Creating custom models by inheriting torch.nn.Module
3. Training and Evaluating a Model
   • General training setup for cpu or cuda
   • Managing and loading predefined/custom datasets
   • Experiencing the conventions of training and evaluating (actually train a CNN on MNIST to accuracy 96%)
4. Saving and Loading Model States
   • Using predefined models in torchvision.models
   • Saving and loading model/optimizer parameters