Deep Learning Teaching Examples (PyTorch)

Deep Learning examples used for teaching within the Department of Computer Science at Durham University (UK) by Dr. Amir Atapour-Abarghouei.

The material is presented as part of the "Deep Learning" lecture series at Durham University.

All material here has been tested with PyTorch 1.12 and Python 3.9.

---

Running the Code:

• You may download each file as needed.
• You can also download the entire repository as follows:

git clone https://github.com/atapour/dl-pytorch
cd dl-pytorch

In this repository, you will find directories that contain examples that demonstrate different features of PyTorch Programming and Deep Learning in general. In the directories, you can find:

• .py file - python code for the example
• .ipynb file - Jupyter notebook for the example

• You can simply run each Python file by running:

python <example file name>.py

• You can run the notebooks using Jupyter.
• Note that it is recommended that you run the scripts (especially those that train a neural network) on GPU hardware. If you do not have access to a GPU locally, you can use free services like Google Colaboratory using the following steps:

Running the Code in Google Colaboratory

• Navigate to - https://colab.research.google.com
• Sign in with your Google account.

Using Google Colab Directly from Github

• Select File -> Upload Notebook... -> Github
• Paste the URL of the notebook on GitHub. For instance, for the notebook in this repository that covers PyTorch Tensors, you can use this URL: https://github.com/atapour/dl-pytorch/blob/main/1.Tensors/1.PyTorch_Programming_Tensors.ipynb. You can change the URL depending on the notebook you would like to run.

Uploading the Notebook from the Local Copy

• Select File -> Upload Notebook...
• Drag and drop or browse to select the notebook you wish to use (e.g., 1.Tensors/PyTorch_Programming_Tensors.ipynb).

Important Note

• If a program is specifically written to use a GPU, make sure you enable the use of a GPU in Google Colab.

• Select Runtime -> Change runtime type -> GPU

• Alternatively, you can change the first code cell of the notebook to use a CPU to run the code by including device = torch.device('cpu').

Contents

In this repository, you can find the following examples:

- 0. Setup

This directory contains examples (```PyTorch_Programming_Setup.py``) that demonstrate how a simple PyTorch environment can be setup andhow visdom works.

Video: https://youtu.be/k-VpBk81k-U

- 1. Tensors:

This directory contains examples (PyTorch_Programming_Tensors.py and PyTorch_Programming_Tensors.ipynb) that demonstrate the functionalities of Tensors in PyTorch.

Video: https://youtu.be/enShn2dhlPo

- 2. Datasets:

This directory contains the dataset "AckBinks: A Star Wars Dataset", which is used to demonstrate how PyTorch handles datasets. It also contains examples (PyTorch_Programming_Datasets.py and PyTorch_Programming_Datasets.ipynb) that show how PyTorch deals with datasets and what tools are available to process data.

Video: https://youtu.be/UIk0MgOsa6c

- 3. Backpropagation:

This directory contains examples (PyTorch_Programming_Backpropagation.py and PyTorch_Programming_Backpropagation.ipynb) that demonstrate how PyTorch enables backpropagation.

Video: https://youtu.be/mLc78Vcqv-g

- 4. Classifier:

This directory contains examples (PyTorch_Programming_Classifier.py and PyTorch_Programming_Classifier.ipynb) that provide an example of training a complete classifier using a simple neural network.

Video: https://youtu.be/Yvvm3w3jLfg

---