About

Hello! This is a series of "workshops" designed to give a gentle introduction to deep learning. Each workshop is a Jupyter notebook that walks through some concepts and provides coding exercises along the way to test your knowledge. Learning about neural networks is not easy, so we include detailed diagrams and explanations to make learning as smooth as possible. We have also included answer keys to check your implementations. These will walk you through implementing a basic neural network from scratch (without an ML framework), and then learning the basics of using ML frameworks (PyTorch).

Feel free to work through these workshops alone, or come visit Agency tuesday meetings if you have questions or want help. We hold an open session where members can work through the workshops, and Agency officers will help with all questions, ranging from installation of the workshops to providing in-depth explanations of neural network topics. We may also walk through a particular worksheet depending on the level of interest on the day. It is very much a relaxed, but productive learning environment.

Installation

Simply fork the repository or download a zip of the code. It is important to keep all the workshops in the correct folder, otherwise images won't load properly.

Setting up a Python virtual environment

All of the required dependencies are listed in requirements.txt. This can also be used directly with pip to install all dependencies directly into your virtual environment.

If you are using Anaconda (which is recommended), you can create an environment directly using the provided yml file with the following command

conda env create -f workshops.yml

Other similar commands also exist to update an existing environment. These commands have been tested on MacOS Catalina, though they should also work for any linux-based or windows system. If you are having trouble and the internet is not helpful, please message us in the slack. Alternatively, you can also create a fresh environment and install the dependencies as follows

conda create -n workshops #replace "workshops" with desired name
conda activate workshops #activate your environment, use same name as before
conda install numpy ipykernel jupyter notebook 

Once you reach the pytorch section of workshops, you can install pytorch with the following command (can be reconfigured for your personal specs at https://pytorch.org/)

conda install pytorch torchvision torchaudio -c pytorch

You can deactivate your environment with

conda deactivate

Make sure to activate your environment before making changes to the installed packages, or before calling

jupyter notebook

in the source directory to open the workshops, and deactivate once you are done!

Installing and Preparing the MNIST Dataset

We have provided two scripts to install and prepare the MNIST dataset. From the source directory, navigate to the data folder with

cd data

Now, run the script to install the data from online

./get_data.sh

or

bash get_data.sh

You will need a working internet connection, and the linux functionalities wget and gzip. Make sure these are installed beforehand. If you are on MacOS, you can install wget via homebrew with

brew install wget

If the script is not working for you, you can alternatively go to Yann LeCun's Website and simply click on the 4 links at the top to download the dataset. Move these files into the data directory, and make sure to unzip them.

Once you have the data downloaded, we will reformat it. You can do this by running

python mnist_to_csv.py

This should add two csv files to your data directory. Now you're ready to write a neural network for this dataset!