第1回 jubatusハンズオン

39
1Jubatusハンズオン 2013 218() Jubatus Team

Upload: jubatusofficial

Post on 06-Dec-2014

11.372 views

Category:

Documents


2 download

DESCRIPTION

 

TRANSCRIPT

  • 1. 1 Jubatus 2013 218 Jubatus Team
  • 2. l (Yuya Unno)l Twitter: @unnonounol Preferred Infrastructurel l l 2
  • 3. Jubatusl l l OKl 3
  • 4. l l Jubatusl 4
  • 5. l l Jubatusl 5
  • 6. JubatusOSS l NTT SIC*Preferred Infrastructure l 201110OSS http://jubat.us/ 6* NTT SIC: NTT
  • 7. l l 7
  • 8. 1l xyl l or (classifier) or 8
  • 9. l l Jubatusl 9
  • 10. Jubatusl jubaclassiferl -f $ jubaclassifiercant start standalone mode withoutconfigpath specifiedusage: jubaclassifier [options] ...options:... []$ jubaclassifier -f /opt/jubatus/share/jubatus/example/config/classifier/pa1.json 10
  • 11. l l 11
  • 12. Jubatus Jubatus (jubaclassifier) Jubatus l Jubatusl l C++/Ruby/Python/Java 12
  • 13. https://github.com/jubatus/jubatus-examplel jubaclassifierl $ cd jubatus-example/gender/python$ ./gender.pyfemale 0.473417669535male 0.388551652431 female 2.79595327377male -2.36301612854 13
  • 14. 1.8 T 0.3 3.2 (+ 1.1 l 14
  • 15. 2.5 T 0.8 2.8 (+ 0.5 l l 15
  • 16. l sample.py#()client = jubatus.Classifier(host, port)train_data = [ ... ]client.train(name, train_data)test_data = [ ... ]results = client.classify(name, test_data)#() 16
  • 17. Jubatusl l #()client = jubatus.Classifier(host, port)train_data = [ ... ]client.train(name, train_data)test_data = [ ... ]results = client.classify(name, test_data)#() 17
  • 18. (train)l l client = jubatus.Classifier(host, port)train_data = [ (male, datum([(hair, short), ...), ...]client.train(name, train_data)test_data = [ ... ]results = client.classify(name, test_data) 18
  • 19. (classify)l l client = jubatus.Classifier(host, port)train_data = [ ... ]client.train(name, train_data)test_data = [ datum([(hair, short), ... ), ...]results = client.classify(name, test_data) 19
  • 20. datuml l l hairshorttopT shirtheight 1.81datum( [(hair, short), (top, T shirt),], [(height, 1.81)]) 20
  • 21. l l client = jubatus.Classifier(host, port)train_data = [ (male, datum([(hair, short), ...), ... # ]client.train(name, train_data)test_data = [ ... ]results = client.classify(name, test_data) 21
  • 22. l l client = jubatus.Classifier(host, port)train_data = [ (male (adult), datum([(hair,short), ...), ...]client.train(name, train_data)test_data = [ ... ]results = client.classify(name, test_data) 22
  • 23. l l Jubatusl 23
  • 24. { "converter" : { ... }, "parameter" : { "regularization_weight" : 1.0 }, "method" : "PA1"} 24
  • 25. { "converter" : { ... }, "parameter" : { ... }, "method" : AROW"}l method l PA1 AROW l 25
  • 26. { "converter" : { ... }, "parameter" : { "regularization_weight" : 10.0 }, "method" : "PA1"}l parameter l l 26
  • 27. { "converter" : { ... }, parameter : { ... }, "method" : "PA1"}l converter l 27
  • 28. (0, 1, 0, 2.5, -1, ) /SVM, LogReg, (1, 0.5, 0.1, -2, 3, ) PA, CW, ALOW, Nave Bayes (0, 1, 0, 1.5, 2, ) CNB, DT, RF, ANN, K-means, Spectral Clustering, MMC, LSI, LDA, GM, HMM, MRF, CRF, 28
  • 29. l l l l l l 29
  • 30. l l 2 1 1 IT 1 1 1 0.7 150 30 0
  • 31. { "hair": "short", "top": "T shirt", "bottom": "jeans", "height": 1.70 } hair=short 1.0 top=T shirt 1.0 bottom=jeans 1.0 height 1.70 31
  • 32. { "hair": "short", l "top": "T shirt", "bottom": "jeans", "height": 1.70 } l 1.0 hair=short 1.0 top=T shirt 1.0 bottom=jeans 1.0 height 1.70 32
  • 33. string_rules... "string_rules" : [ { "key" : "*, "type" : "str", "sample_weight" : "bin, "global_weight" : "bin" } ],...l key: * l type: str l sample_weight, global_weight: 1.0 33
  • 34. { "hair": "short", l "top": "T shirt", "bottom": "jeans", "height": 1.70}hair=short 1.0top=T shirt 1.0bottom=jeans 1.0height 1.70 34
  • 35. num_rules... num_rules" : [ { "key" : "*, "type" : num } ],...l key: * l type: num 35
  • 36. 1.0 1.0 2.0 1.0 1.0l 36
  • 37. ... "string_rules" : [ { "key" : "*, "type" : space", "sample_weight" : "bin, "global_weight" : "bin" } ],...l typespacel 37
  • 38. l l http://jubat.us/ja/ l l l http://groups.google.com/group/jubatusl l https://github.com/jubatus/jubatus 38
  • 39. l jubatus-example l l l http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets l news20l Enjoy! 39