HowtoSetupaPythonEnvironment

forMachineLearning

SettingupyourPythonenvironmentforMachineLearningcanbea

trickytask.Ifyou’veneversetupsomethinglikethatbefore,you

mightspendhoursfiddlingwithdifferentcommandstryingtogetthe

thingtowork.ButwejustwanttogetrighttotheML!

Inthistutorial,youwilllearnhowtosetupastablePythonMachine

Learningdevelopmentenvironment.You’llbeabletogetrightdown

intotheMLandneverhavetoworryaboutinstallingpackagesever

again.

(1)SetupPython3andPip

Thefirststepistoinstallpip,aPythonpackagemanager:

sudoapt-getinstallpython3-pip

Usingpip,we’llbeabletoinstallanyPythonpackagethat’sindexed

inthePythonPackageIndexwithasimplepipinstall

your_package.You’llseesoonhowweuseittosetupourvirtual

environmenttoo.

Next,we’llsetPython3tobethedefaultwhenrunningeitherthe

piporpythoncommandsfromcommandline.Thismakesusing

Python3easierandmoreconvenient.Ifwedidn’tdothis,thenifwe

wantedtousePython3,we’dhavetoremembertotypeoutpip3

andpython3everytime!

ToforcePython3tobethedefault,we’regoingtomodifythe

~/.bashrcfile.Fromthecommandline,executethefollowing

commandtoviewthatfile:

nano~/.bashrc

Scrollondowntothe#somemorelsaliasessectionandaddthe

followingline:

aliaspython='python3'

Savethefileandreloadyourchanges:

source~/.bashrc

Boom!Python3isnowyourdefaultPython!Youcanrunitwitha

simplepythonyour_programonthecommandline.

(2)Createavirtualenvironment

Nowwe’llsetupavirtualenvironment.Inthere,we’llinstallallof

thepythonpackagesthatweneedforMachineLearning.

Weusevirtualenvironmentsinordertoseparateourcodingsetups.

Imagineifatsomepointyouwantedtodo2differentprojectson

yourcomputer,whichrequireddifferentlibrariesofdifferent

versions.Havingthemallinthesameworkingenvironmentcanbe

messyandyou’lllikelyrunintotheproblemofconflictinglibrary

versions.YourMLcodeforproject1needsversion1.0ofnumpy,but

project2needsversion1.15.Yikes!

Avirtualenvironmentallowsustoisolateourworkingareastoavoid

thoseconflicts.

First,installtherelevantpackages:

sudopipinstallvirtualenvvirtualenvwrapper

Oncewehavevirtualenvandvirtualenvwrapperinstalled,we’llagain

needtoeditour~/.bashrcfile.Placethese3linesrightatthe

bottomandsaveit.

exportWORKON_HOME=$HOME/.virtualenvs

exportVIRTUALENVWRAPPER_PYTHON=/usr/bin/python3

source/usr/local/bin/virtualenvwrapper.sh

Savethefileandreloadyourchanges:

source~/.bashrc

Great!Nowwecanfinallycreateourvirtualenvironmentlikeso:

mkvirtualenvml

We’vejustcreatedavirtualenvironmentcalledml.Toenterit,do

this:

workonml

Nice!Anylibraryinstallationsthatyoudowhileinthemlvirtualenv

willbeisolatedinthereandneverconflictwithanyother

environments!Sowheneveryouwishtoruncodethatdependson

librariesinstalledinthemlenvironment,enterintofirstwiththe

workoncommandandthenrunyourcodeasnormal.

Ifyouneedtoexitthevirtualenv,runthiscommand:

deactivate

(3)InstallMachineLearning

libraries

NowwecaninstallourMLlibraries!We’llgowiththemost

commonlyusedones:

?numpy:foranyworkwithmatrices,especiallymathoperations

?scipy:scientificandtechnicalcomputing

?pandas:datahandling,manipulation,andanalysis

?matplotlib:datavisualisation

?scikitlearn:machinelearning

Here’sasimpletricktoinstallallofthoselibrariesinonequickshot!

Createarequirements.txtfileandlistallofthepackagesyouwish

toinstalllikeso:

numpy

scipy

pandas

matplotlib

scikit-learn

Oncethat’sdone,justexecutethiscommand:

pipinstall-rrequirements.txt

Voila!Pipwillgoaheadandinstallallofthepackageslistedinthefile

inonesho