Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Apache Mahout: Driving the Yellow Elephant


Published on

  • Be the first to comment

Apache Mahout: Driving the Yellow Elephant

  1. 1. Apache Mahout – Driving the Yellow Elephant<br />Grant Ingersoll<br />TriHUG<br />
  2. 2. Anyone Here Use Machine Learning?<br />Any users of:<br />Google?<br />Search?<br />Priority Inbox?<br />Facebook?<br />Twitter?<br />LinkedIn?<br />
  3. 3. Topics<br />What is Machine Learning?<br />ML Use Cases<br />What is Mahout?<br />A Word on Scaling<br />What can I do with it right now?<br />Mahout and Hadoop: An Example<br />
  4. 4.<br />What is Machine Learning?<br />Google News<br />
  5. 5. Really it’s…<br />“Machine Learning is programming computers to optimize a performance criterion using example data or past experience”<br />Intro. To Machine Learning by E. Alpaydin<br />Subset of Artificial Intelligence<br />Lots of related fields:<br />Information Retrieval<br />Stats<br />Biology<br />Linear algebra<br />Many more<br />
  6. 6. Common Use Cases<br />Recommend friends/dates/products<br />Classify content into predefined groups<br />Find similar content based on object properties<br />Find associations/patterns in actions/behaviors<br />Identify key topics in large collections of text<br />Detect anomalies in machine output<br />Ranking search results<br />Others?<br />
  7. 7. Apache Mahout<br /><br />An Apache Software Foundation project to create scalable machine learning libraries under the Apache Software License<br /><br />Why Mahout?<br />Many Open Source ML libraries either:<br />Lack Community<br />Lack Documentation and Examples<br />Lack Scalability<br />Lack the Apache License ;-)<br />Or are research-oriented<br />
  8. 8. Who uses Mahout?<br /><br />
  9. 9. What does scalable mean?<br />Ted Dunning (Mahout committer):<br />As data grows linearly, either scale linearly in time or in machines<br />2X data requires 2X time or 2X machines (or less!)<br />Goal: Be as fast and efficient as possible given the intrinsic design of the algorithm<br />Some algorithms won’t scale to massive machine clusters<br />Others fit logically on a Map Reduce framework like Apache Hadoop<br />Still others will need different distributed programming models<br />Be pragmatic<br />
  10. 10. What Can I do with Mahout Right Now?<br />
  11. 11. Recommendations<br />Extensive framework for collaborative filtering<br />Recommenders<br />User based<br />Item based<br />Online and Offline support<br />Offline can utilize Hadoop<br />Many different Similarity measures<br />Cosine, LLR, Tanimoto, Pearson, others<br />It’s Valentine’s Day soon!<br />
  12. 12. Clustering<br />Document level<br />Group documents based on a notion of similarity<br />K-Means, Fuzzy K-Means, Dirichlet, Canopy, Mean-Shift<br />Distance Measures<br />Manhattan, Euclidean, other<br />Topic Modeling <br />Cluster words across documents to identify topics<br />Latent Dirichlet Allocation<br />
  13. 13. Categorization<br />Place new items into predefined categories:<br />Sports, politics, entertainment<br />Recommenders<br />Implementations<br />Naïve Bayes<br />Compl. Naïve Bayes<br />Decision Forests<br />Linear Regression<br /><ul><li>See Chapter 17 of Mahout in Action for Shop It To Me use case:
  14. 14.</li></li></ul><li>Freq. Pattern Mining<br />Identify frequently co-occurrent items<br />Useful for:<br />Query Recommendations<br />Apple -> iPhone, orange, OS X<br />Related product placement<br />Basket Analysis<br /><br />
  15. 15. Evolutionary<br />Map-Reduce ready fitness functions for genetic programming<br />Integration with Watchmaker<br /><br />Problems solved:<br />Traveling salesman<br />Class discovery<br />Many others<br />Caveat: Hasn’t received as much attention as others<br />
  16. 16. Other<br />Primitive Collections!<br />Math library<br />Vectors, Matrices, etc.<br />Noise Reduction via Singular Value Decomposition<br />Export from Lucene/Solr and other formats<br />
  17. 17. Mahout and Hadoop<br />Most Mahout implementations are built on Map-Reduce<br />Many also have sequential implementations<br />Linear Regression is blazingly fast without needing M/R<br />Let’s look at how K-Means is implemented in Mahout<br />
  18. 18. K-Means<br />Clustering Algorithm<br />Nicely parallelizable!<br /><br />
  19. 19. K-Means in Map-Reduce<br />Input:<br />Mahout Vectors representing the original content<br />Either:<br />A predefined set of initial centroids (Can be from Canopy)<br />--k – The number of clusters to produce<br />Iterate<br />Do the centroid calculation (more in a moment)<br />Clustering Step (optional)<br />Output<br />Centroids (as Mahout Vectors)<br />Points for each Centroid (if Clustering Step was taken)<br />
  20. 20. Map-Reduce Iteration<br />Each Iteration calculates the Centroids using:<br />KMeansMapper<br />KMeansCombiner<br />KMeansReducer<br />Clustering Step<br />Calculate the points for each Centroid using:<br />KMeansClusterMapper<br />
  21. 21. KMeansMapper<br />During Setup:<br />Load the initial Centroids (or the Centroids from the last iteration)<br />Map Phase<br />For each input<br />Calculate it’s distance from each Centroid and output the closest one<br />Distance Measures are pluggable<br />Manhattan, Euclidean, Squared Euclidean, Cosine, others<br />
  22. 22. KMeansReducer<br />Setup:<br />Load up clusters<br />Convergence information<br />Partial sums from KMeansCombiner (more in a moment)<br />Reduce Phase<br />Sum all the vectors in the cluster to produce a new Centroid<br />Check for Convergence<br />Output cluster<br />
  23. 23. KMeansCombiner<br />A Combiner is like a Map-side Reducer which helps save on IO<br />Just like KMeansReducer, but only produces partial sum of the cluster based on the data local to the Mapper<br />
  24. 24. KMeansClusterMapper<br />Some applications only care about what the Centroids are, so this step is optional<br />Setup:<br />Load up the clusters and the DistanceMeasure used<br />Map Phase<br />Calculate which Cluster the point belongs to<br />Output <ClusterId, Vector><br />
  25. 25. Summary<br />Machine learning is all over the web today<br />Mahout is about scalable machine learning<br />Mahout has functionality for many of today’s common machine learning tasks<br />Many Mahout implementations use Hadoop<br />KMeans clustering is an example of a machine learning algorithm in Mahout that is implemented using Map Reduce<br />
  26. 26. Resources<br /><br /><br />{user|dev}<br /><br /><br />
  27. 27. Resources<br />“Mahout in Action” by Owen, Anil, Dunning and Friedman<br /><br />“Introducing Apache Mahout” <br /><br />“Programming Collective Intelligence” by Toby Segaran<br />“Data Mining - Practical Machine Learning Tools and Techniques” by Ian H. Witten and Eibe Frank<br />
  28. 28. References<br />HAL:<br />Terminator:<br />Matrix:<br />Google News:<br /><br />Facebook:<br />Couple:<br />Beer and Diapers:<br /><br />DMOZ:<br />Shopping Cart:<br />