Data Science On WSL

Prerequisite: Windows 10 Build 14393 or later

This guide contains notes on setting up a development environment for data science in Windows 10's new Windows Subsystem for Linux (WSL) aka Bash on Ubuntu on Windows.

Bash and Python

(Optional) Install Cmder
Install WSL
Install Anaconda
Install X server for Windows
1. Configure X and apply fix for dbus
(Optional) Install Command Line power user features

Quick Start

After installing WSL and installing VcXsrv as your X server, run the following commands in Bash. This installs Anaconda, Oh My Zsh, and node. For more detailed information on these fixes, please see the rest of this guide.

For Windows 10 Anniversary Build 14393 users:

sudo apt-get update
sudo apt-get upgrade
wget https://repo.continuum.io/archive/Anaconda2-4.2.0-Linux-x86_64.sh
bash Anaconda2-4.2.0-Linux-x86_64.sh
conda install -c jzuhone zeromq=4.1.dev0
conda install matplotlib=1.5.1
sudo apt-get install zsh git libqtgui4 xserver-xorg-video-dummy
sh -c "$(curl -fsSL https://raw.github.com/robbyrussell/oh-my-zsh/master/tools/install.sh)"
echo "export DISPLAY=localhost:0.0" >> ~/.bashrc
echo "export KMP_AFFINITY=disabled" >> ~/.bashrc
echo "zsh" >> ~/.bashrc
sudo sed -i 's$<listen>.*</listen>$<listen>tcp:host=localhost,port=0</listen>$' /etc/dbus-1/session.conf
sudo apt-get install python-software-properties
curl -sL https://deb.nodesource.com/setup_6.x | sudo -E bash -
sudo apt-get install nodejs

For Windows 10 Insider Build 14986 users:

sudo apt-get update
sudo apt-get upgrade
wget https://repo.continuum.io/archive/Anaconda2-4.2.0-Linux-x86_64.sh
bash Anaconda2-4.2.0-Linux-x86_64.sh
conda install matplotlib=1.5.1
sudo apt-get install zsh git libqtgui4 xserver-xorg-video-dummy
sh -c "$(curl -fsSL https://raw.github.com/robbyrussell/oh-my-zsh/master/tools/install.sh)"
echo "export DISPLAY=localhost:0.0" >> ~/.bashrc
echo "zsh" >> ~/.bashrc
sudo apt-get install python-software-properties
curl -sL https://deb.nodesource.com/setup_6.x | sudo -E bash -
sudo apt-get install nodejs

Bash and Python

1. (Optional) Install Cmder

Cmder is an excellent console emulator for Windows offering many features beyond what cmd provides. To name just a few, you'll benefit from color themes, tabs, and git integration.

Go to cmder.net
Select Download full
Extract somewhere convenient
(Optional) Add the Cmder directory to your PATH variable to allow for quick launch from the Start Menu
1. Press Win
2. Type environment and select Edit the system environment variables
3. Click Environment Variables
4. Select Path entry under the System variables section and press Edit...
5. Click New to add your Cmder directory and press OK when done
Apply fix for Shift+Up hot key
1. Go to <Cmder Directory>\vendor\clink
2. Open clink_inputrc_base in Notepad
3. Find and replace all M-C-u with "\033`b". There should be three instances.
4. Save and restart Cmder. Now pressing Shift+Up will navigate up in your directories
Apply fix for up/down arrow keys for command history in Bash
1. Go to <Cmder Directory>\config
2. Open user-aliases.cmd in Notepad
3. Add this as a new line at the bottom:
```
bash=bash -cur_console:p
```

2. Install WSL

Follow Microsoft's installation guide here. Below are additional notes on WSL you should know about.

On the "Create a UNIX user" step, if you want root access you can create a user with root as the name. This will create your user as a superuser and will preclude you from needing to use sudo for any commands.
If you mess up your WSL and wish to do a clean install, use the following commands in Cmder (not Bash):
```
 lxrun /uninstall /full
 lxrun /install
```
File permissions are handled separately by both Windows and Linux. See here for more information. You may want to run Cmder as administrator. If you want to always run Cmder as admin, you can do so by right-clicking its .exe file and selecting the option in its properties.
When in Windows, you can find the Linux file system at %AppData%\Local\lxss. When in Bash, you can find the Windows file system at /mnt/c.
- Despite this, file interoperability is not supported. Linux files have information stored in their NTFS Extended Attributes that Windows can't create. If you try to edit them in Windows, those attributes might get stripped and become unusable in Linux also. Very important read on what you can and cannot do here.

3. Install Anaconda

Anaconda is a widely used Python-based data science platform. It comes with a large selection of preinstalled packages as well as features to help you manage environments for multiple Python versions. Downloads are located at https://www.continuum.io/downloads.

Open Cmder and type bash

Choose if you want to use Python 2.7 or 3.5 with one of the following commands (Anaconda2 for 2.7, Anaconda3 for 3.5):

wget https://repo.continuum.io/archive/Anaconda2-4.2.0-Linux-x86_64.sh
wget https://repo.continuum.io/archive/Anaconda3-4.2.0-Linux-x86_64.sh

After the download completes, install via one of the following commands:

bash Anaconda2-4.2.0-Linux-x86_64.sh
bash Anaconda3-4.2.0-Linux-x86_64.sh

Fix Jupyter notebooks

In Windows 10 Build 14393, there is an issue with libzmq that is fixed in later Insider Builds. For now though, we will need to do the following command in bash:

conda install -c jzuhone zeromq=4.1.dev0

This patches zeromq to work with WSL. If this step is not done, the kernel will keep dying whenever you view a Jupyter notebook resulting in an inability to execute code cells. For more information about this issue, see here.

Also, if you see these messages when running jupyter notebook with a web browser installed:

libGL error: No matching fbConfigs or visuals found
libGL error: failed to load driver: swrast

You can try sudo apt-get install mesa-utils to see if that resolves it.

Fix MKL

In Windows 10 Build 14393's Ubuntu 14.04, using MKL optimized packages (NumPy, NumExpr, SciPy, Scikit-Learn) results in the following errors:

OMP: Error \#100: Fatal system error detected.
OMP: System error \#22: Invalid argument

This is caused by an issue with WSL's use of the Thread Affinity Interface. Luckily for us there are several possible workarounds/solutions.

Workaround #1 - Disable Thread Affinity

The community does not fully understand the implications of this workaround yet. However, the people using it have not reported any adverse side effects. If you run into issues consider using an alternative solution.

Open Bash

Type the following command to amend your .bashrc:

echo "export KMP_AFFINITY=disabled" >> ~/.bashrc

Workaround #2 - Disable MKL

MKL is relatively new and you may not necessarily need it. You can switch Anaconda to use non-MKL (OpenBLAS for Linux) package versions instead of the MKL ones. For more information on this, see here.

Open Bash

Use the following command:

conda install nomkl numpy scipy scikit-learn numexpr
conda remove mkl mkl-service

Note: This workaround resolves Anaconda's MKL usage, but there may still be other packages you use that relies on MKL (especially it's LAPACK and BLAS). If those packages do not have a non-MKL variant then this option won't work for you.

Solution #1 - Upgrade to Windows 10 Insider Build 14951 or above

With Build 14951, WSL installs Ubuntu 16.04 instead of 14.04. This version supports MKL.

Fix Verification

After using one of the options above, you can verify the fix with the following:

python -c 'import scipy; scipy.test()'

It should no longer generate the original OMP error messages.

Fix Matplotlib

At the time of writing, Anaconda2 v4.2.0 installs Matplotlib 1.5.3-np111py27_0 along with pyqt 5.6.0-py27_0 and qt 5.6.0-0. Unfortunately WSL seems to run into some bugs with qt5.6 that are going to be fixed in future release qt5.6.3.

Solution #1 - Downgrade to Matplotlib 1.5.1 and use Qt4

sudo apt-get install libqtgui4
conda install matplotlib=1.5.1

Alternate Option - Continue using Matplotlib 1.5.3, ignore Qt5 error

sudo apt-get install libqt5gui5

This is the error you will see when using Matplotlib. No clue what the full extent of implications may be.

QXcbShmImage: shmget() failed (38: Function not implemented) for size 1586872 (602x659)

Additional Notes

On Windows 10 Anniversary Build, from the X window, if you try to save a plot as an image, you may get the following:

QInotifyFileSystemWatcherEngine::addPaths: inotify_add_watch failed: Invalid argument
QFileSystemWatcher: failed to add paths: /home/josh

This is fixed in Windows 10 Insiders Build 14942. For more information see here.

4. Install X server for Windows

Linux uses the X Window System which uses a server-client model to display GUI applications. In order for us to view these applications on Windows, we will need a X server capable of receiving the communication coming from our Linux's X clients.

There are two main options we can choose from: VcXsrv and Xming. Choose whichever one you like. If you run into issues displaying applications in one, try the other. We will use VcXsrv in this guide since some users are reporting Xming to be a bit slower.

Download VcXsrv
Install and run VcXsrv
(Optional) Add VcXsrv as a service so you don't have to manually start it every time

Configure X and fix dbus

We will need to configure Linux to send X communication to where our Windows' X server is listening at.

Open Bash
Type the following command:
```
echo "export DISPLAY=localhost:0.0" >> ~/.bashrc
```
This is enough to launch X applications to our X server, but many applications will still perform poorly or crash. This happens because Windows 10 Build 14393 does not fully support Unix sockets yet. Later Insider Builds are said to have this fixed already as well.

Type the following to change dbus into using TCP instead of Unix sockets:

sudo sed -i 's$<listen>.*</listen>$<listen>tcp:host=localhost,port=0</listen>$' /etc/dbus-1/session.conf

To test if this is working we can install and run Sublime Text from Linux. You may need to restart your bash if it doesn't work right away.
```
sudo add-apt-repository ppa:webupd8team/sublime-text-3
sudo apt-get update
sudo apt-get install sublime-text-installer
subl
```
For other X GUI applications (like Firefox), you may also need LibGL.

sudo apt-get install xserver-xorg-video-dummy

5. (Optional) Install Command Line power user features

Bash is great, but there are quite a few annoying nuances to it like tab completion being case sensitive. To improve our productivity we can upgrade our CLI away from Bash to Zsh. To learn about some of the great Zsh features see this blog post. Those can then be further improved by using Oh My Zsh.

Install Oh My Zsh

Open Bash and install the prerequisites for Oh My Zsh with the following commands
```
sudo apt-get update
sudo apt-get install zsh git
```

Install Oh My Zsh with curl

sh -c "$(curl -fsSL https://raw.github.com/robbyrussell/oh-my-zsh/master/tools/install.sh)"

If you prefer wget you can use this command:

sh -c "$(wget https://raw.github.com/robbyrussell/oh-my-zsh/master/tools/install.sh -O -)"

(Optional) Configure your Bash to use Zsh as the default shell on startup
1. Open .bashrc in your choice of editor
```
vi ~/.bashrc
```
2. Add this as a new line at the very bottom
```
zsh
```
  Note: In the future if you end up further modifying your .bashrc, you will want to check that the zsh line remains at the bottom. This ensures your full .bashrc file is processed before starting Zsh.
3. Restart Bash and it should automatically start Zsh
(Optional) Configure your Oh My Zsh
1. Open .zshrc in your choice of editor
```
vi ~/.zshrc
```
2. To change the Zsh theme, change the ZSH_THEME line. I recommend the "agnoster" theme. More themes can be found here.
```
ZSH_THEME="agnoster"
```
3. To add additional plugins, scroll down to the plugins=(git) line. This line is space-delimited so you can add whichever plugins you want inside the parenthesis. The list of default plugins can be found here.
```
plugins=(git python sudo)
```

(Optional) Disable beeping in the terminal:

echo "setopt NO_BEEP" >> ~/.zshrc
sudo sh -c 'echo "set bell-style none" >> /etc/inputrc'
echo "set visualbell" >> ~/.vimrc

Install Powerline compatible fonts

Many Zsh themes use custom characters to display symbols like git branches. To have these displayed properly in Cmder you will need to use a Powerline compatible font.

Choose a font from this GitHub repo. Font samples can be seen here. I recommend "DejaVu Sans Mono".
Download and install the font's .ttf files into Windows
Open Cmder
Press Win+Alt+P to bring up the Settings screen
Change Main console font to your Powerline font.
Press Save settings

Configure Cmder startup tasks

From Cmder's Settings screen, go to the Startup --> Tasks tab
Press +
Name the group Bash on Ubuntu
(Optional) Check the box Default task for new console

Fill in the following for Task parameters:

-icon %USERPROFILE%\AppData\Local\lxss\bash.ico

Fill in the following for the task command:
```
%windir%\system32\bash.exe
```
(Optional) Go to Startup tab and change Specified named task to {Bash on Ubuntu}

Node and Web Dev

1. Install node

With how dominant node.js is in web development these days, it is important to get this working on WSL.

Open Bash and add the official Node.js PPA

sudo apt-get install python-software-properties
curl -sL https://deb.nodesource.com/setup_6.x | sudo -E bash -

Install node
```
sudo apt-get install nodejs
```
Test node and npm versions
```
node -v
npm -v
```

(Optional) nvm

If you need multiple versions of node, use these node version manager (nvm) instructions.

Open Bash and install the prerequisites for nvm with the following commands

sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install libssl-dev

Install nvm with curl

curl -o- https://raw.githubusercontent.com/creationix/nvm/v0.32.1/install.sh | bash

If you prefer wget you can use this command:

wget -qO- https://raw.githubusercontent.com/creationix/nvm/v0.32.1/install.sh | bash

If using Zsh, you'll need to add the command to your .zshrc

export NVM_DIR="/home/josh/.nvm"
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"  # This loads nvm

Replace josh with your own username.

Install and use latest node LTS
```
nvm install --lts
nvm use --lts
```

2. Known Issues/Limitations

Not everything in the web development world works properly with WSL yet. The following is an incomplete list of some items I've personally ran into.

Item	Version	Issue
nvm	any	Slow shell loading especially when a default node version is set. See discussion here. Not related to WSL per se.
FileSystem watch	any	Any node watch features will fail because inotify is not implemented on Build 14393. Fixed in Build 14942.
Meteor	1.4.2.3	Unable to establish connections to MongoDB when starting projects
docpress	0.7.1	Unable to build static site. Issue discussion found here.

joshpeng/Data-Science-On-WSL

Data Science On WSL

Table of Contents

Quick Start

Bash and Python

Node and Web Dev

Quick Start

Bash and Python

1. (Optional) Install Cmder

2. Install WSL

3. Install Anaconda

Fix Jupyter notebooks

Fix MKL

Workaround #1 - Disable Thread Affinity

Workaround #2 - Disable MKL

Solution #1 - Upgrade to Windows 10 Insider Build 14951 or above

Fix Verification

Fix Matplotlib

Solution #1 - Downgrade to Matplotlib 1.5.1 and use Qt4

Alternate Option - Continue using Matplotlib 1.5.3, ignore Qt5 error

Additional Notes

4. Install X server for Windows

Configure X and fix dbus

5. (Optional) Install Command Line power user features

Install Oh My Zsh

Install Powerline compatible fonts

Configure Cmder startup tasks

Node and Web Dev

1. Install node

(Optional) nvm

2. Known Issues/Limitations