Cloud Computing CS

Transcription

1 Cloud Computing CS Apache Mahout Feb 13, 2012 Shannon Quinn

2 MapReduce Review Scalable programming model Map phase Shuffle Reduce phase MapReduce Implementations Google Hadoop Map Phase Reduce Phase chunks C0 C1 C2 C3 mappers M0 M1 M2 M3 IO0 IO1 IO2 IO3 Shuffling Data Reducers R0 R1 FO0 FO1 Figure from lecture 6: MapReduce

3 MapReduce Review Scalable programming model Map phase Shuffle Reduce phase MapReduce Implementations Google Hadoop This is our focus! Map Phase Reduce Phase chunks C0 C1 C2 C3 mappers M0 M1 M2 M3 IO0 IO1 IO2 IO3 Shuffling Data Reducers R0 R1 FO0 FO1 Figure from lecture 6: MapReduce

4 Apache Mahout A scalable machine learning library

5 Apache Mahout A scalable machine learning library Built on Hadoop

6 Apache Mahout A scalable machine learning library Built on Hadoop Philosophy of Mahout (and Hadoop by proxy)

7 What does Mahout do?

8 Recommendation

9 Classification

10 Clustering

11 Other Mahout Algorithms Dimensionality Reduction Regression Evolutionary Algorithms

12 Mahout 1. Recommendation 2. Classification 3. Clustering

13 Recommendation Overview Help users find items they might like based on historical preferences

14 Recommendation Overview Mathematically

15 Recommendation Overview Alice Bob? 2 5 Peter *based on example by Sebastian Schelter

16 Recommendation Overview *based on example by Sebastian Schelter

17 Recommendation Overview Bob? *based on example by Sebastian Schelter

18 Recommendation Overview Bob *based on example by Sebastian Schelter

19 Recommendation in Mahout 1 st Map phase: process input *based on example by Sebastian Schelter

20 Recommendation in Mahout 1 st Map phase: process input *based on example by Sebastian Schelter 1 st Reduce phase: list by user

21 Recommendation in Mahout 2 nd Map phase: Emit co-occurred ratings *based on example by Sebastian Schelter

22 Recommendation in Mahout *based on example by Sebastian Schelter 2 nd Map phase: Emit co-occurred ratings 2 nd Reduce phase: Compute similarities

23 Mahout 1. Recommendation 2. Classification 3. Clustering

24 Classification Overview Assigning data to discrete categories

25 Classification Overview Assigning data to discrete categories Train a model on labeled data Spam Not spam

26 Classification Overview Spam? Not spam Assigning data to discrete categories Train a model on labeled data Run the model on new, unlabeled data

27 Naïve Bayes Example

28 Naïve Bayes Example Prob (token label) =

29 Naïve Bayes Example Not spam

30 Naïve Bayes Example Not spam President Obama s Nobel Prize Speech

31 Naïve Bayes Example Spam

32 Naïve Bayes Example Spam Spam content

33 Naïve Bayes Example

34 Naïve Bayes Example Order a trial Adobe chicken daily EAB-List new summer savings, welcome!

35 Naïve Bayes in Mahout Complex!

36 Naïve Bayes in Mahout Complex! Training 1. Read the features

37 Naïve Bayes in Mahout Complex! Training 1. Read the features 2. Calculate per-document statistics

38 Naïve Bayes in Mahout Complex! Training 1. Read the features 2. Calculate per-document statistics 3. Normalize across categories

39 Naïve Bayes in Mahout Complex! Training 1. Read the features 2. Calculate per-document statistics 3. Normalize across categories 4. Calculate normalizing factor of each label

40 Naïve Bayes in Mahout Complex! Training 1. Read the features 2. Calculate per-document statistics 3. Normalize across categories 4. Calculate normalizing factor of each label Testing Classification

41 Other Classification Algorithms Stochastic Gradient Descent

42 Other Classification Algorithms Stochastic Gradient Descent Support Vector Machines

43 Other Classification Algorithms Stochastic Gradient Descent Support Vector Machines Random Forests

44 Mahout 1. Recommendation 2. Classification 3. Clustering

45 Clustering Overview Grouping unstructured data

46 Clustering Overview Grouping unstructured data Small intra-cluster distance

47 Clustering Overview Grouping unstructured data Small intra-cluster distance Large inter-cluster distance

48 K-Means Clustering Example

49 K-Means Clustering Example

50 K-Means Clustering Example

51 K-Means Clustering Example

52 K-Means Clustering Example

53 K-Means Clustering Example

54 K-Means Clustering Example

55 K-Means Clustering Example

56 K-Means Clustering Example

57 K-Means Clustering Example Dogs Cats

58 K-Means Clustering in Mahout Map Phase chunks mappers C0 C1 C2 C3 M0 M1 M2 M3 + Reduce Phase IO0 IO1 IO2 IO3 Shuffling Data Reducers R0 R1 FO0 FO1 Figure from lecture 6: MapReduce

59 K-Means Clustering in Mahout Assume: # clusters <<< # points

60 K-Means Clustering in Mahout Assume: # clusters <<< # points M0 M1 M2 M3

61 K-Means Clustering in Mahout Assume: # clusters <<< # points M0 M1 M2 M3 <clusterid, observation> R0 R1

62 K-Means Clustering in Mahout Map phase: assign cluster IDs

63 K-Means Clustering in Mahout Map phase: assign cluster IDs Reduce phase: reset centroids

64 K-Means Clustering in Mahout Important notes --maxiter --convergencedelta method

65 Other Clustering Algorithms Latent Dirichlet Allocation Topic models

66 Other Clustering Algorithms Latent Dirichlet Allocation Topic models Fuzzy K-Means Points are assigned multiple clusters

67 Other Clustering Algorithms Latent Dirichlet Allocation Topic models Fuzzy K-Means Points are assigned multiple clusters Canopy clustering Fast approximations of clusters

68 Other Clustering Algorithms Latent Dirichlet Allocation Topic models Fuzzy K-Means Points are assigned multiple clusters Canopy clustering Fast approximations of clusters Spectral clustering Treat points as a graph

69 Other Clustering Algorithms Latent Dirichlet Allocation Topic models Fuzzy K-Means Points are assigned multiple clusters Canopy clustering Fast approximations of clusters Spectral clustering Treat points as a graph K-Means & Eigencuts

70 Mahout in Summary

71 Mahout in Summary Scalable library

72 Mahout in Summary Scalable library Three primary areas of focus

73 Mahout in Summary Scalable library Three primary areas of focus Other algorithms

74 Mahout in Summary Scalable library Three primary areas of focus Other algorithms All in your friendly neighborhood MapReduce

75 Mahout in Summary Scalable library Three primary areas of focus Other algorithms All in your friendly neighborhood MapReduce

76 Thank you!