Machine Learning Techniques

A brief introduction to machine learning and machine learning resources for Rubyists.

Transcript

1. Machine Learning Techniques
   Ben Curtis / ben@honeybadger.io / @stympy
   

   View Slide

2. Making Sense of a Bunch of Data
   Ben Curtis / ben@honeybadger.io / @stympy
   

   View Slide

3. View Slide

4. Logging
   It's big, it's heavy, it's wood
   http://www.stockvault.net/photo/102553/cutdown
   

   View Slide

5. { Graylog2
   Apache Kafka
   

   View Slide

6. Clustering
   So happy together
   http://www.stockvault.net/photo/115439/yellow-flowers
   

   View Slide

7. Group anything that can be plotted on an X-Y chart
   K-Means Clustering
   7
   

   View Slide

8. K-Means Live Demo!
   8
   http://en.wikipedia.org/wiki/K-means
   

   View Slide

9. http://en.wikipedia.org/wiki/K-means
   K-Means Live Demo!
   

   View Slide

10. http://en.wikipedia.org/wiki/K-means
    K-Means Live Demo!
    

    View Slide

11. http://en.wikipedia.org/wiki/K-means
    K-Means Live Demo!
    

    View Slide

12. View Slide

13. View Slide

14. View Slide

15. reddavis/K-Means
    1
    require  'k_means'  
       
    data  =  [[1,1],  [1,2],  [1,1],  [1000,  1000],  [500,  500]]  
    kmeans  =  KMeans.new(data,  :centroids  =>  2)  
    kmeans.inspect    #  Use  kmeans.view  to  get  hold  of  the  un-­‐inspected  array  
    =>  [[3,  4],  [0,  1,  2]]
    

    View Slide

16. Deciding
    …and that has made all the difference
    http://www.flickr.com/photos/47051377@N00/4033866900/
    

    View Slide

17. igrigorik/decisiontree
    require  'decisiontree'  
    !
    attributes  =  ['Temperature']  
    training  =  [  
       [36.6,  'healthy'],  
       [37,  'sick'],  
       [38,  'sick'],  
       [36.7,  'healthy'],  
       [40,  'sick'],  
       [50,  'really  sick'],  
    ]  
    !
    #  Instantiate  the  tree,  and  train  it  based  on  the  data  (set  default  to  '1')  
    dec_tree  =  DecisionTree::ID3Tree.new(attributes,  training,  'sick',  :continuous)  
    dec_tree.train  
    !
    decision  =  dec_tree.predict([37,  'sick'])  
    puts  "Predicted:  #{decision}  ...  True  decision:  #{test.last}";  
    !
    #  =>  Predicted:  sick  ...  True  decision:  sick  
    

    View Slide

18. A shallow magnitude 2.7 earthquake aftershock was reported Monday morning
    four miles from Westwood, according to the U.S. Geological Survey. The temblor
    occurred at 7:23 a.m. Pacific time at a depth of 4.3 miles.
    A magnitude 4.4 earthquake was reported at 6.25 a.m. and was felt over a large
    swath of Southern California.
    According to the USGS, the epicenter of the aftershock was five miles from Beverly
    Hills, six miles from Santa Monica and six miles from West Hollywood.
    In the last 10 days, there has been one earthquake of magnitude 3.0 or greater
    centered nearby.
    This information comes from the USGS Earthquake Notification Service and this
    post was created by an algorithm written by the author.
    http://lat.ms/1lTIGqa
    

    View Slide

19. Classifying
    Everyone starts as a Level 1 human with no class
    

    View Slide

20. Good/Bad? Spam/Not-spam?
    Bayesian Classifiers
    2
    

    View Slide

21. cardmagic/classifier
    2
    require  'classifier'  
    !
    b  =  Classifier::Bayes.new  'Interesting',  'Uninteresting'  
    b.train_interesting  "here  are  some  good  words.  I  hope  you  love  them"  
    b.train_uninteresting  "here  are  some  bad  words,  I  hate  you"  
    b.classify  "I  hate  bad  words  and  you"  #  returns  'Uninteresting'
    

    View Slide

22. Analyze text to identify the topic
    Latent Semantic Indexing
    2
    

    View Slide

23. cardmagic/classifier
    2
    require  'classifier'  
       
    lsi  =  Classifier::LSI.new  
    strings  =  [  ["This  text  deals  with  dogs.  Dogs.",  :dog],  
                       ["This  text  involves  dogs  too.  Dogs!  ",  :dog],  
                       ["This  text  revolves  around  cats.  Cats.",  :cat],  
                       ["This  text  also  involves  cats.  Cats!",  :cat],  
                       ["This  text  involves  birds.  Birds.",:bird  ]]  
    strings.each  {|x|  lsi.add_item  x.first,  x.last}  
       
    lsi.search("dog",  3)  
    #  returns  =>  ["This  text  deals  with  dogs.  Dogs.",  "This  text  involves  dogs  too.  
    Dogs!  ",    
    #                          "This  text  also  involves  cats.  Cats!"]  
       
    lsi.find_related(strings[2],  2)  
    #  returns  =>  ["This  text  revolves  around  cats.  Cats.",  "This  text  also  involves  
    cats.  Cats!"]  
       
    lsi.classify  "This  text  is  also  about  dogs!"  
    #  returns  =>  :dog
    

    View Slide

24. Recommending
    If I Knew You Were Comin' I'd've Baked a Cake
    http://www.flickr.com/photos/48973657@N00/4556156477/
    

    View Slide

25. How similar are sets of data?
    Jaccard Index
    

    View Slide

26. francois/jaccard
    2
    a  =  ["likes:jeans",  "likes:blue"]  
    b  =  ["likes:jeans",  "likes:apples",  "likes:red"]  
    c  =  ["likes:apples",  "likes:red"]  
       
    #  Determines  how  similar  a  pair  of  sets  are  
    Jaccard.coefficient(a,  b)  
    #=>  0.25  
       
    Jaccard.coefficient(a,  c)  
    #=>  0.0  
       
    Jaccard.coefficient(b,  c)  
    #=>  0.6666666666666666  
       
    #  According  to  the  input  data,  b  and  c  have  the  most  similar  likes.
    

    View Slide

27. davidcelis/recommendable
    2
    class  User  
       recommends  :movies,  :books,  :minerals,  :other_things  
       
       #  ...  
    end  
       
    >>  user.like(movie)  
    =>  true  
    >>  user.liked_movies  
    =>  [#]  
    >>  user.recommended_movies  
    =>  [#,  ...]
    

    View Slide

28. Ted Dunning
    & Ellen Friedman
    Innovations in Recommendation
    Practical Machine
    Learning
    http://www.mapr.com/practical-machine-learning
    

    View Slide

29. Machine Learning Techniques
    Ben Curtis / ben@honeybadger.io / @stympy
    

    View Slide