Introductory applied machine learning (INFR10069)

These instructions are written for set up on DICE / a Unix environment.

Distance learners please note: We recommend two options for working remotely. We preference that you use your own machine:

1. Use your own machine! Conda installation should work fine on your own computer. You must still have a DICE account: when submitting assignments, you will need to copy work up to DICE and submit from there. Instructions will be given for this in each assignment
2. Use virtual dice - a virtual machine emulated on your own computer connected to the dice network. Please read here for installation instructions and more: http://computing.help.inf.ed.ac.uk/vdice

Windows users please note:

• After conda installation, all instructions are much the same
• please follow conda installation instructions on their website here
• to activate the iaml environment, note that you don't type source activate iaml but instead just activate iaml
• you can ignore section 3b (though you can google windows equivalents of all the Unix commands given)

Setting up on DICE or Unix

Within this course we will be using Python along with a few open-source libraries (packages). We will be using a virtual environment and package management tool called conda.

We're going to run the entire setup in the terminal. If you're on a DICE machine, click Applications in the top left, go to Utilities, then click Terminal.

In the below instructions, any text styled like this should be executed in the terminal. We are expecting you to enter these commands in by hand one-by-one. This is for pedagogical reasons, and to help detect new issues. Please read and heed any warnings and especially errors you may encounter. We are on standby in the labs to help if required.

1. Check your available space

Firstly, please note that your space on DICE is allocated dynamically. If you are having problems it may be because you were using new space faster than it could be allocated to you!

All DICE users registered for IAML will automatically be allocated 20GB extra space over their default space values. Please register for the course ASAP to get this space.

1. Check how much space you have on DICE. You will need at least 4.5GB.
   1. freespace
   2. If you don't have enough space, follow the instructions on this page

2. If you don't have it - Install conda

1. Check you don't already have conda installed!
   1. which conda
   2. if you already have it installed, skip ahead to Create an Environment
   3. It doesn't matter if you have miniconda2, miniconda3, anaconda2, or anaconda3 installed - they will all work for this course
2. If you don't have conda, download the latest version of miniconda2
   1. cd ~/Downloads (you can make a Downloads folder if you don't have one)
   2. Download the installer, depending on your system (you can check links here):
      • Linux: wget https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh
      • Mac: wget https://repo.continuum.io/miniconda/Miniconda2-latest-MacOSX-x86_64.sh or curl -LOk https://repo.continuum.io/miniconda/Miniconda2-latest-MacOSX-x86_64.sh
      • Or just simply download from the site
3. Install miniconda2 with default settings
   1. bash Miniconda2-latest-Linux-x86_64.sh
   2. Follow the prompt - type yes and hit enter to accept all default settings when asked
4. Close Terminal and reopen
5. Try executing conda -h. If it works, you can delete the installer rm ~/Downloads/Miniconda2-latest-Linux-x86_64.sh

3a. Create an environment for IAML

1. Update conda: conda update conda (at the time of writing, the latest version was 4.3.25, but you should be safe to use later versions)
2. Create the environment for the course. Call it iaml and install python 2: conda create --name iaml python=2

3b. Err...what's an environment?

An environment is a collection of packages of specific versions. You can have multiple environments and switch between them for different projects. Conda is a tool for managing both environments and the packages within each environment. Here's a quick intro:

1. Show a list of your environments: conda env list
2. Print $PATH, one of your system's environment variables, in the terminal: echo $PATH
   • $PATH is the list of directories your terminal can search to find anything you execute:
3. Print a list of python installations on your $PATH (the top one is the one that will get executed if you type python in the terminal): which python -a
4. Activate the new environment: source activate iaml
5. Show list of python installations on your system now: which python -a
6. Show your system $PATH again: echo $PATH
7. Deactivate the new environment: source deactivate
8. Observer how your $PATH has changed again: echo $PATH
9. Make an empty environment: conda create --name empty
10. You can clone environments; this is useful for backing up: conda create --name empty_bkp --clone empty
11. Make a python 3 environment with numpy already installed: conda create --name py3 python=3 numpy
12. conda env list
13. Activate py3: source activate py3
14. Show the installed packages: conda list
15. Switch environments: source deactivate; source activate empty
16. conda list to show packages (note that python and, crucially, pip are not installed)
17. Q: What python would get used now? which python A: the conda root environment installation of python i.e. not this environment's python.
18. Install numpy: conda install numpy
19. Q: What python would get used now? which python A: You may have clocked that conda installed a dependency of numpy (a python package)...python!
20. Let's delete these test environments:
    • source deactivate
    • conda env list
    • conda remove --name empty --all
    • conda remove --name empty_bkp --all
    • conda remove --name py3 --all
    • conda env list

4. Install all the packages for IAML

1. Activate the environment: source activate iaml
2. {May take 5 minutes} Install all required packages: conda install jupyter=1.0.0 matplotlib=2.0.2 pandas=0.20.3 numpy=1.13.1 scikit-learn=0.19.0 scipy=0.19.1 seaborn=0.8
   • Please note that normally we wouldn't specify the version numbers. Conda automatically downloads the most recent consistent set of packages. We specify versions here such that this course is consistent regardless of when you start (/recreate your environment!)
3. Get some space back: conda clean -a

IMPORTANT

Before starting any IAML work in a new terminal you must always activate the iaml conda environment using source activate iaml. If the environment is not activated, you will be using your base python with its own set of packages. If you are ever in any doubt of which python version is being used, execute which python.

5. Get course material

You should now have all the required modules installed. Our next step is to make a new directory where we will keep all the lab notebooks, datasets and assignments. Within your terminal:

1. Navigate back to your home directory: cd
2. Make a new directory and navigate to it
   • mkdir iaml_2017
   • cd iaml_2017

Now you have two options:

1. We recommend that you directly download a .zip file from https://github.com/JamesOwers/iaml2017 which will contain everything you need and save it in the folder you have just created. You can do this from the terminal by typing:
   • wget https://github.com/JamesOwers/iaml2017/archive/master.zip
   • unzip master.zip
2. If and only if you are familiar and confident with using Git/GitHub, you can initialize a git directory, add the above repo as remote and pull everything into your local directory

IMPORTANT

Supporting and teaching git is not in scope for this course so please only do this if you are happy to google your own solutions!

6. Get started!!!

Once you have downloaded the material, you are now ready to start working with Jupyter notebooks. First you need to activate the software environment and then start a Jupyter Notebook session from within the folder where the material is stored. You will have to follow this procedure for all labs and assignments.

1. Activate the conda environment: source activate iaml
2. Enter the directory where you downloaded the course material: cd iaml_2017/iaml-master
3. Start a jupyter notebook
   • jupyter notebook
4. This should automatically open your browser
   • Click on 01_Lab_0_Introduction.ipynb to open it

Further Reading

• Conda getting started - 30 minute practical well worth going through https://conda.io/docs/user-guide/getting-started.html
• System Environment variables - https://en.wikipedia.org/wiki/Environment_variable
• Linux execution order - https://www.cyberciti.biz/tips/how-linux-or-unix-understand-which-program-to-run-part-i.html

Troubleshooting

I ran out of space when installing packages

Firstly, please note that your space on DICE is allocated dynamically. If you are having problems it may be because you were using new space faster than it could be allocated to you!

1. Check how much space you have on DICE. You will need at least 4.5GB.
   1. freespace
   2. If you don't have enough space, follow the instructions on this page
2. Try instaling packages individually and executing conda clean --all after each installation

Trashing your environment

If you install incorrect packages, or a package breaks for some reason, you can just delete your environment and start again. Execute conda remove --name iaml --all then install the package as described above.

Trashing your whole conda installation

This is fairly extreme but as a final resort can be done quickly and easily. Please note that you will lose all your environments if you do this, so check this will not affect you before proceeding...follow instructions here

Unzipping master.zip error

Check that you downloaded the zip correctly! An error like:

End-of-central-directory signature not found.  Either this file is not
  a zipfile, or it constitutes one disk of a multi-part archive.  In the
  latter case the central directory and zipfile comment will be found on
  the last disk(s) of this archive.

means that the file you've downloaded is likely incomplete. Try downloading from the GitHub repo directly by clicking the green button and downloading the zip.

wget: command not found

You do not have wget installed! Either install it, download from the GitHub repo directly by clicking the green button and download the zip, or try using another program like curl e.g. curl -LOk https://github.com/JamesOwers/iaml2017/archive/master.zip

conda: command not found or 'Conda never works in new terminal'

DICE issue: DICE has a different set of bash startup mechanism, and you may need to edit some different files yourself. Do the below with ~/.benv instead. See here for more info.

Unix solution: First try closing your terminal and reopening. If that doesn't fix, it's likely that, in the conda installation, you didn't allow conda to add the it's bin directory to your $PATH. Check your home directory for ~/.bashrc or ~/.bash_profile. You should have a line in one of those files that looks like this:

# added by Miniconda2 4.3.21 installer
export PATH="/afs/inf.ed.ac.uk/user/s12/s1234567/miniconda2/bin:$PATH"

If you don't, you can add it like this (changing the conda installation location line):

miniconda_location=/afs/inf.ed.ac.uk/user/s12/s1234567/miniconda2/bin
echo 'export PATH=:$PATH' >>~/.bash_profile

'source' is not recognized as an internal or external command, ...

You're on windows aren't you! Please see the note at the top of the file (you can omit source)

'which' is not recognized as an internal or external command, ...

You're on windows aren't you! Please see the note at the top of the file ( which = where on windows).

$PATH?

You're on windows aren't you! Please see the note at the top of the file ( echo $PATH == echo %PATH% on windows).

I can't find my conda environment....but I definitely created it

We have found that people also taking MLP and/or ANLP (other courses that use conda) have installed multiple versions of conda. To check whether you've done this, simply list your home directory:

ls ~

If you see multiple folders called anaconda or miniconda, e.g. anaconda3 and miniconda2, you have installed multiple versions of conda! Another way to check is to print your PATH or view your .bashrc / .benv:

echo $PATH
cat ~/.bashrc
cat ~/.benv  # if you're on DICE

This will show multiple conda directories.

You only need to use one installation of conda, and it doens't matter whether you use version 2 or 3 (there is no difference that will affect this course).

Simply recreate your environment(s) in one of the conda installations, and delete the other.

• https://conda.io/docs/user-guide/tasks/manage-environments.html#sharing-an-environment
• https://conda.io/docs/user-guide/install/linux.html#uninstalling-anaconda-or-miniconda