Apache Mahout Architecture Overview

Mahout is an open source machine learning java library from Apache Software Foundation, and therefore platform independent, that provides a fertile framework and collection of patterns and ready-made component for testing and deploying new large-scale algorithms.
With these slides we aims at providing a deeper understanding of its architecture.

1. MAHOUT
Dalla Palma Stefano
Open source Machine Learning Java library
A PA C H E
PartI:Overview

2. Search
Text mining
Information Retrieval

3. Search
Text mining
Information Retrieval

4. Search
Text mining
Information Retrieval
Collaborative
filtering

5. Search
Text mining
Information Retrieval
Collaborative
filtering

6. MahoutMachine Learning maintechniques
and
architectureoverview

7. Recommender
Data Store
User
PreferenceItem
PreferenceItem
Recommender
Neighborhood
Correlation
Preference
Inferrer
DataModel
User
User

8. Classification
(Naïve Bayes)
Training
examples
Training
algorithm
ModelNew examples Decisions
Predictorsand
target variables
ClassificationSystem
Copy
Estimated
target
variable
Predicted
variables
only
Model

9. Clustering

10. Clustering

11. Clustering
(k-means)
Mapoutput = <centroid_id, data_point>
S = Shuffleand Sorting
M1
Split 1
Split 2
Split n-1
Split n
…Input
M2
Mn
R1
R2
Rk
S
New Centroids
file on Distr.
Cache
Centroidsfile on
Distr. Cache
Mapphase Reduce phase

13. Samsara
Apache Flink

14. Flink

15. MAHOUT
Open source Machine Learning Java library
A PA C H E
PartII:Architecture

38. MAHOUT
Open source Machine Learning Java library
A PA C H E
PartIII:Conclusions

44. Stablestcomponents

45. Mostinstablecomponents

51. Questions?
Dalla Palma Stefano
A PA C H E M A H O U T