This is a tutorial presented during a workshop at the Applied Machine Learning Days 2021.

It consists of 8 interactive Jupyter notebooks.

Notebooks

The links below open the notebooks in Google Colab. This way, you don't have to install anything on your computer and you get a GPU for free.

1. Basics
2. Data
3. Autograd
4. Optimization
5. Modules
6. CNN
7. Transfer Learning
8. Pretrained models for NLP

Log in using your Google account

Enable GPU support by changing the ‘runtime type’

Introduction to Google Colab

This notebook covers the basic features of Google Colab.

---

Further learning

Check out these others tutorials and courses if you are hungry to learn more:

• Official tutorials: pytorch.org/tutorials. There are many!
• Step-by-step tutorial: towardsdatascience.com
• PyTorch for deep learning (.py files): yunjey/pytorch-tutorial
• PyTorch for deep learning (notebooks): yandexdataschool/Practical_DL

---

Alternative setups

Binder

Binder is an alterantive way to run remote notebooks that does not require a Google account.

Running locally using Conda

If you want to run the notebooks locally, you can use conda. The following instructions should work on Linux/Mac OS, Windows might require slight adaptations.

Step 1: Install conda

If you have not installed it yet, you can download it from Anaconda (Python 3.6 version).

Verify that it is installed by running

conda -V

Make sure your conda installation is up-to-date:

conda update conda

Step 2: Download repository and install environment

Now clone the repository:

git clone https://github.com/theevann/amld-pytorch-workshop.git
cd amld-pytorch-workshop

The available conda environments can be listed using

conda env list

Let's now create a new environment called 'amld-pytorch'.

conda env create -f environments.yml

Step 3: Activate/Deactivate the environment

After the environment has been created, you can activate it by

source activate amld-pytorch

Now start the Jupyter notebook by running

jupyter notebook

The environment can similarly deactivated by

source deactivate