Scribd
Upload
73 views

Data Mining Essen, Als 2: Data Mining in Prac, Ce, With Python

This document provides an overview of a course on social media mining using Python. It introduces Python as a good programming language for data mining due to its ease of use, efficiency, and wide range of available modules. It then discusses Scikit-Learn, a popular machine learning library for Python. The remainder of the document outlines how the course will demonstrate unsupervised and supervised machine learning techniques like PCA, k-means clustering, linear regression, and logistic regression using Python and Scikit-Learn. Code examples and Jupyter notebooks will be used to illustrate these techniques hands-on.

Uploaded by

anon_991097212
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
73 views

Data Mining Essen, Als 2: Data Mining in Prac, Ce, With Python

This document provides an overview of a course on social media mining using Python. It introduces Python as a good programming language for data mining due to its ease of use, efficiency, and wide range of available modules. It then discusses Scikit-Learn, a popular machine learning library for Python. The remainder of the document outlines how the course will demonstrate unsupervised and supervised machine learning techniques like PCA, k-means clustering, linear regression, and logistic regression using Python and Scikit-Learn. Code examples and Jupyter notebooks will be used to illustrate these techniques hands-on.

Uploaded by

anon_991097212
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

EECS 510: Social Media Mining

Spring 2015

Data  Mining  Essen,als  2:    


Data  Mining  in  Prac,ce, with  Python  

Rosanne  Liu  
[email protected]  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression  
•  Why  Python?  
 

What programming language


do you use for data mining?

Source from: http://www.kdnuggets.com/polls/index.html


How much is your salary as
analytics, data mining, data
science professionals?

Source from: http://www.kdnuggets.com/polls/index.html


Should data scientist / data
miners be responsible for
their predictions?

Source from: http://www.kdnuggets.com/polls/index.html


Why  Python?  

•  Why  Python?  
  Not
 Think  about  the  scien,st’s  needs:  
§  Get  data  (simulaGon,  experiment  control)  
§  Manipulate  and  process  data.  
§  Visualize  results...  to  understand  what  we  are  doing!  
§  Communicate  results:  produce  figures  for  reports  or  
publicaGons,  write  presentaGons.  
Why  Python?  

•  Why  Python?  
  Not
–  Easy  
•  Easy  to  learn,  easily  readable  
•  ScienGsts  first,  programmers  second  
–  Efficient  
•  Managing  memory  is  easy  –  if  you  just  don’t  care  
–  A  single  Language  for  everything  
•  Avoid  learning  a  new  soXware  for  each  new  problem  
More  to  Take  Away  

•  Free  distribuGon  from  hZp://www.python.org  


•  Known  for  it’s  “baZeries  included”  philosophy  
Similar  to  R,  Python  has  a  fantasGc  community  around  it  and,  
luckily  for  you,  this  community  can  write  
•  Two  popular  versions,  2.7  or  3.x  
•  A  single-­‐click  installer:  Enthought  Canopy  
•  Prepare  yourself  for  code  indentaGon  heaven  
All  the  Good  Modules  

•  numpy,  scipy:  basics  for  almost  everything  


•  Matplotlib,  a  Python  2D  plobng  library  hZp://matplotlib.org  
•  NLTK,  Natual  Language  Toolkit  hZp://www.nltk.org  
•  Pandas,  Python  Data  Analysis  Library  hZp://pandas.pydata.org  
•  mrjob,  route  to  wriGng  MapReduce  jobs  hZps://pythonhosted.org/mrjob/  
•  IPython,  InteracGve  console  with  IDE-­‐like  features  hZp://ipython.org  
•  Scikit-­‐Learn,  ML  resource  and  library  
hZp://scikit-­‐learn.org/dev/index.html  
•  Theano/Pylearn2,  deep  learning  
hZp://deeplearning.net/soXware/theano/  
hZp://deeplearning.net/soXware/pylearn2/  
•  More:  mlpy,  PyBrain,  Orange,  Scrapy,  …  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression  
The  Use  of  Python:  Simple  demos  
 
0  –  Python  Intro.ipynb  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression  
What  is  Scikit-­‐learn  

•  A  Python  Machine  Learning  Library  


•  Focused  on  modeling  data  
•  Developed  by  David  Cournapeau  as  a  Google  summer  of  code  
project  in  2007.    
•  First  public  release  (v0.1  beta)  published  in  late  January  2010.  
•  Now  has  more  than  30  acGve  contributors  and  has  had  paid  
sponsorship  from  INRIA,  Google,  Tinyclues  and  the  Python  
SoXware  FoundaGon.  
•  The  library  is  built  upon  the  SciPy  that  must  be  installed  
before  you  can  use  scikit-­‐learn.  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression  
The  use  of  Scikit-­‐Learn:  unsupervised  learning  demos  
 
PCA  Summary  

•  PCA  projects  to  axis  with  greatest  variance  


•  OXen  provides  good  first  insight  into  dataset  

 
 
•  IdenGfy  important  variables  in  projecGon  matrix  W:  
1  –  PCA.ipynb  
K-­‐Means  Algorithm  
2  –  k  means.ipynb  
Outline  

•  Why  Python?  
•  Intro  to  Python  
•  Intro  to  Scikit-­‐Learn  
•  Unsupervised  Learning  
–  Demo  on  PCA,  K-­‐Means  
•  Supervised  Learning  
–  Demo  on  Linear  Regression,  LogisGc  Regression,  kNN  
The  use  of  Scikit-­‐Learn:  supervised  learning  demos  
 
Linear  Regression  

1D 2D

To find w and b, minimize the error:


3  –  LinearRegression1.ipynb  
3  –  LinearRegression2.ipynb  
Logis,c  Regression  
Logis,c  Regression  
4  –  LogisGcRegression.ipynb  
Nonlinear  Problems  

•  LogisGc  regression  works  well  if  the  data  is  linearly  separable,  
but…  
K  Nearest  Neighbors  

•  ClassificaGon:  same  setup  as  logisGc  regression.    


•  ︎Very  simple  but  powerful  idea:  Do  as  your  neighbors  do.    
•  ︎For  a  new  point  x  look  at  the  nearest  (or  the  two  nearest  or  
three  nearest,  ...)  point(s)  in  the  training  data  for  a  label.    
•  ︎Usual  distance  measure:  Euclidean  distance    
Simple  Algorithm  

•  Pick  a  k,  for  example  k  =  3.  


•  Want  to  classify  new  example  x.  
•  Compute  di  =  d(xi,  x),  i.e.  d(xi,  x)  =  ||xi  −  x||.  
•  Sort  di  ,  take  k  smallest:  di0  ,  di1  ,  di2  .  
•  Assign  y  that  appears  most  oXen  among  yi0  ,  yi1  ,  yi2  .    
5  –  kNN.ipynb  

You might also like