# Installation
The following contains everything you need to know to install Python and all relevant packages on your computer, access online computing resources and run jupyter notebooks. Please read it carefully before approaching us with questions.
We describe one way of doing things that works. Sometimes there are alternative ways (e.g. using a different shell, using a graphical interface instead of a shell, ...). If you know what you are doing, feel free to do things differently but be aware that we do not provide help if you run into problems in that case.
## In JupyterHub
JupyterHub lets you access a jupyter notebook online. All packages you need are already installed. You can start writing code right away.
The only donwside of JupyterHub is that the computing resources, disk space and memory are very modest. Your laptop or google colab notebooks might be faster and give you more space (i.e. you can work with larger models or have to clear the cache less often).
You can start JupyterHub from ecampus. To do so, navigate to the course on ecampus and click on the JupyterHub button
Click on "start"
Click on "start my server" and wait. This can take a while. Do not reload the page.
You can click on the top tile ("Notebook - Python 3") to start a fresh notebook.
You can also drag and drop a notebook from your computer to the left sidebar and open it
by double-clicking on it.
One the notebook is open, you can add your python code:
When you are done, you can download the notebook and use it on a different computer:
## On your computer
Your own computer is usually the most convenient place to run code and in most cases it is more powerful than what you get on JupyterHub. However, you have to install Python there and create and activate a conda environment.
### Get Anaconda
First, go to the [anaconda webpage](https://www.anaconda.com/products/distribution) and download the installer.
Next, follow the [installation instructions](https://docs.anaconda.com/anaconda/install/index.html) for your operating system. Go with the recommended defaults in all cases.
If you want, you can watch the video for your operating system below but please ignore everything that uses the anaconda navigator and always use the latest versions of anaconda in case the videos are outdated.
### Create the environment
Download the environment file for your operating system. You can choose where you save it.
```{eval-rst}
:download:`windows <_static/environment_windows.yml>`
:download:`mac and linux <_static/environment.yml>`
```
Open a Terminal (on windows, use the Anaconda Prompt that comes with Anaconda) in the folder in which you stored the environment file.
Run the command
```bash
conda env create -f environment.yml
```
If your environment file is called differently (e.g. `environment_windows.yml`) you need to adjust the command (e.g. to `conda env create -f environment_windows.yml`).
This will take a while and the terminal output you get differs across different computers.
You only have to do this once for the entire course.
The following video shows this process on windows but it is the same on all other operating systems:
### Opening a notebook
There are many ways to open a notebook. Some do not involve a terminal. Use them if you know what you are doing, but do what we write below if you are new:
Open a terminal (on windows use the Anaconda Prompt that comes with Anaconda) in the folder in which your notebook is stored or in the folder in which you want to store a new notebook. This can be any folder and it does not need to contain the environment file from the pervios step.
Run the following commands:
```bash
conda activate dl_intro
```
`dl_intro` is the name of the environment you created in the previous step. The environment name is not the same as the name of the `.yml` file.
```bash
jupyter notebook
```
This will open a tab with the notebook in your browser. Moreover, it will print cryptic messages to your terminal. You do not need to understand them but you cannot close the terminal. Otherwise your notebook shuts down as well.
On the start screen you can open existing notebooks or create new ones.
Once you opened a notebook, it is the same as on jupyterhub.
## On Google Colab
Google colab is a platform by google that allows you to run notebooks online, without setting anything up on your computer. The main advantage of google colab is that it provides free GPUs, which you can use to train your models faster. Using google colab for the class is strictly optional.
To get started, log into your google account and visit: [https://colab.research.google.com/](https://colab.research.google.com/).
Click on upload and browse and select the notebook you previously downloaded to your computer.
The result should look like this
Almost all packages we will need in this class are pre-installed on colab. The exception are the huggingface libraries. Since there is no good conda integration for google colab, we have to install these packages from within the notebook. To do so, add a new cell at the beginning and paste the following code snippet:
```
import os
IS_ON_COLAB = bool(os.getenv("COLAB_RELEASE_TAG"))
if IS_ON_COLAB:
!pip install transformers tokenizers datasets sentencepiece huggingface_hub[cli]
```
In the notebook it should look like this:
Press shift-enter to run the installation. The output should look like this:
Note that while this cell does check if the notebook is run on colab, you should remove this cell for notebooks that are not run on colab. You can ommit the pip installation if you do not need the hugginface libraries for a particular notebook.