This curriculum contains a lot of executable examples and labs that you would want to run. In order to do this, you need the ability to execute Python code in Jupyter Notebooks provided as part of this curriculum. You have several options for running the code:
To run the code locally on your computer, you would need to have some version of Python installed. I personally recommend installing miniconda - it is rather lightweight installation that supports conda
package manager for different Python virtual environments.
After you install miniconda, you need to clone the repository and create a virtual environment to be used for this course:
git clone http://github.com/microsoft/ai-for-beginners
cd ai-for-beginners
conda env create --name ai4beg --file .devcontainer/environment.yml
conda activate ai4beg
Probably the best way to use the curriculum is to open it in Visual Studio Code with Python Extension.
Note: Once you clone and open the directory in VS Code, it will automatically suggest you to install Python extensions. You would also have to install miniconda as described above.
Note: If VS Code suggests you to re-open the repository in container, you need to decline this to use local Python installation.
You can also use Jupyter environment right from the browser on your own computer. Actually, both classical Jupyter and Jupyer Hub provide quite convenient development environment with auto-completion, code highlighting, etc.
To start Jupyter locally, go to the directory of the course, and execute:
jupyter notebook
or
jupyterhub
You then can navigate to any of the .ipynb
files, open them and start working.
One alternative to Python installation would be to run the code in container. Since our repository contains special .devcontainer
folder that instructs how to build a container for this repo, VS Code would offer you to re-open the code in container. This will require Docker installation, and also would be more complex, so we recommend this to more experienced users.
If you do not want to install Python locally, and have access to some cloud resources - a good alternative would be to run the code in the cloud. There are several ways you can do this:
- Using GitHub Codespaces, which is a virtual environment created for you on GitHub, accessible through VS Code browser interface. If you have access to Codespaces, you can just click Code button in the repo, start a codespace, and get running in no time.
- Using Binder. Binder is a free computing resources provided in the cloud for people like you to test out some code on GitHub. There is a button at the front page to open the repository in Binder - this should quickly take you to the binder site, which will build underlying container and start Jupyter web interface for you seamlessly.
Note: To prevent misuse, Binder has access to some web resources blocked. This may prevent some of the code working, which fetches models and/or datasets from public Internet. You may need to find some workarounds. Also, compute resources provided by Binder are pretty basic, so training will be slow, especially in later more complex lessons.
Some of the later lessons in this curriculum would greatly benefit from GPU support, because otherwise training will be painfully slow. There are a few options you can follow, especially if you have access to the cloud either through Azure for Students, or through your institution:
- Create Data Science Virtual Machine and connect to it through Jupyter. You can then clone the repo right onto the machine, and start learning. NC-series VMs have GPU support.
Note: Some subscriptions, including Azure for Students, do not provide GPU support out of the box. You may need to request additional GPU cores through technical support request.
- Create Azure Machine Learning Workspace and then use Notebook feature there. This video shows how to clone a repository into Azure ML notebook and start using it.
You can also use Google Colab, which comes with some free GPU support, and upload Jupyter Notebooks there to execute them one-by-one.