Machine learning a zpracovn dat pomoc Microsoft Azure

Bc. Luk Beran

Diplomov prce 2015

ABSTRAKT

Tato diplomov prce se zabv monostmi Machine Learning v Microsoft Azure. V teore-

tick sti prce je nahldnuto do historie strojovho uen v Microsoftu, popsny jsou prak-

tick pklady vyuit strojovho uen a soust jsou i dv ppadov studie vyuit Azure

Machine Learning v praxi, z nich jedna popisuje inteligentn zen univerzitn budovy.

V praktick sti prce jsou nzorn ukzny monosti vyuit Azure Machine Learning na

pedpovdi hodnocen film.

Klov slova: strojov uen, azure, microsoft, zpracovn dat

ABSTRACT

This Masters thesis deals with the possibilities of Machine Learning in Microsoft Azure. In

the theoretical part of the thesis is looked into the history of machine learning in Microsoft,

described are specific examples of using machine learning and included are two case studies

of the use of Azure Machine Learning in practice, one of which describes the intelligent

management of a university building. In the practical part of the thesis are clearly presented

how to use Azure Machine Learning predictions on movie ratings.

Keywords: machine learning, azure, microsoft, data processing

Dkuji vedouc m diplomov prce pan doc. Ing. Zuzan Komnkov Oplatkov, Ph.D. za

velmi cenn pipomnky na konzultacch a veden m prce. Dle dkuji spolenosti Micro-

soft za poskytnut bezplatnho pstupu do Microsoft Azure, jmenovit pak pan

Daniele Likov a panu Brandonu Beckovi za vstcnost a poskytnut informace. A samo-

zejm dkuji i sv rodin a ptelkyni za trplivost a podporu.

Prohlauji, e odevzdan verze bakalsk/diplomov prce a verze elektronick nahran do

IS/STAG jsou toton.

http://www.utb.cz/fai/struktura/zuzana-kominkova-oplatkova

OBSAH

VOD .................................................................................................................................... 8

I. TEORETICK ST ............................................................................................... 9

1 STROJOV UEN ................................................................................................. 10

2 VUDYPTOMN DOPORUOVN .............................................................. 12

3 MICROSOFT AZURE ............................................................................................ 16

4 SLUBY MICROSOFT AZURE ........................................................................... 17

4.1 ACTIVE DIRECTORY ........................................................................................... 17

4.2 BACKUP ................................................................................................................... 17

4.3 CDN ........................................................................................................................... 17

4.4 HDINSIGHT ............................................................................................................. 17

4.5 MACHINE LEARNING .......................................................................................... 17

4.6 SITE RECOVERY ................................................................................................... 18

4.7 STORAGE ................................................................................................................ 18

4.8 VIRTUAL MACHINES........................................................................................... 18

4.9 WEBSITES ............................................................................................................... 18

5 AZURE MACHINE LEARNING ........................................................................... 19

5.1 HISTORIE STROJOVHO UEN VE SPOLENOSTI MICROSOFT ........ 19

5.2 PKLAD STROJOVHO UEN V MICROSOFT MALWARE

PROTECTION CENTER ....................................................................................... 20

5.3 PPADOV STUDIE MACHINE LEARNING ................................................. 21

5.3.1 JJ FOOD SERVICE ................................................................................................. 21

5.3.2 CARNEGIE MELLON UNIVERSITY ......................................................................... 22

5.4 MACHINE LEARNING STUDIO ......................................................................... 22

5.5 SLUBA AZURE ML API ...................................................................................... 24

5.6 CENY AZURE MACHINE LEARNING .............................................................. 24

II. PRAKTICK ST ................................................................................................ 27

6 MICROSOFT AZURE MACHINE LEARNING V PRAXI ............................... 28

6.1 PRVN KONTAKT S AZURE MACHINE LEARNING .................................... 28

6.2 VYTVOEN VLASTNHO EXPERIMENTU.................................................... 35

6.2.1 ZSKN DAT ........................................................................................................ 35

6.2.2 PEDZPRACOVN DAT ......................................................................................... 38

6.2.3 DEFINOVN PARAMETR .................................................................................... 43

6.2.4 VBR A APLIKACE UCHO ALGORITMU ............................................................ 44

6.2.5 PEDPOVDI NAD NOVMI DATY .......................................................................... 47

7 ANALZA DATABZE FILM A PREDIKCE HODNOCEN....................... 50

7.1 DATA DOSTUPN Z IMDB .................................................................................. 50

7.2 TVORBA MODELU V MICROSOFT AZURE ................................................... 51

7.3 PEDZPRACOVN DAT ..................................................................................... 53

7.4 POROVNN MODEL PEDPOVDI ............................................................. 56

7.4.1 STUDIE 1: BOOSTED DECISION TREE, PARAMETR PRMRNCH HERC ............... 56

7.4.2 STUDIE 2: BOOSTED DECISION TREE, VECHNY PARAMETRY ............................... 56

7.4.3 STUDIE 3: NEURAL NETWORK REGRESSION, PARAMETR PRMRNCH HERC .................................................................................................................. 56

7.4.4 STUDIE 4: NEURAL NETWORK REGRESSION, VECHNY PARAMETRY .................... 57

7.4.5 STUDIE 5: DECISION FOREST REGRESSION, PARAMETR PRMRNCH HERC ...... 57

7.4.6 STUDIE 6: DECISION FOREST REGRESSION, VECHNY PARAMETRY ...................... 57

7.4.7 STUDIE 7: LINEAR REGRESSION, PARAMETR PRMRNCH HERC ...................... 58

7.4.8 STUDIE 8: LINEAR REGRESSION, VECHNY PARAMETRY ...................................... 58

7.4.9 STUDIE 9: BOOSTED DECISION TREE, PODSTATN KORELACE .............................. 58

7.4.10 STUDIE 10: BOOSTED DECISION TREE, PODSTATN KORELACE, EXPERIMENTLN NASTAVEN .............................................................................. 59

7.4.11 STUDIE 11: BOOSTED DECISION TREE, PODSTATN KORELACE, DOSTATEK HODNOT................................................................................................................ 59

7.4.12 STUDIE 12: BOOSTED DECISION TREE, EXPERIMENTLN ZJITN PARAMETRY I NASTAVEN ..................................................................................... 60

7.4.13 STUDIE 13: BOOSTED DECISION TREE, EXPERIMENTLN ZJITN PARAMETRY, NASTAVEN DLE SWEEP PARAMETERS ............................................. 60

7.4.14 STUDIE 14: BOOSTED DECISION TREE, EXPERIMENTLN ZJITN PARAMETRY, NASTAVEN DLE SWEEP PARAMETERS, VCE NE 30

HODNOCEN .......................................................................................................... 61

7.5 ZHODNOCEN VSLEDK ................................................................................. 63

ZVR ............................................................................................................................... 64

SEZNAM POUIT LITERATURY .............................................................................. 65

SEZNAM POUITCH SYMBOL A ZKRATEK ..................................................... 70

SEZNAM OBRZK ....................................................................................................... 71

SEZNAM TABULEK ........................................................................................................ 73

UTB ve Zln, Fakulta aplikovan informatiky 8

VOD

Strojov uen je v souasn dob velmi populrn tma, jeliko poadavky na pedpovdi a

doporuovn se tkaj velk sti komern i nekomern sfry.

Microsoft Azure je cloudov platforma, kter poskytuje velk mnostv slueb pro firmy i

jednotlivce. Machine Learning pat k nejnovjm slubm dostupnm v Microsoft Azure

a nabz techniky strojovho uen, kter svou jednoduchost a kvalitn zpracovanou doku-

mentac uspokoj jak bn uivatele se zkladn znalost statistiky a strojovho uen, tak i

nron velk firmy, kter mohou vyut podporu jazyka R nebo Python, garantovanou do-

stupnost a technickou podporou.

Clem prce je vytvoen praktickho nvodu pro prci s Microsoft Azure Machine Learning

a praktick ukzka monost na vlastnm pkladu.

Vyuit strojovho uen se nachz i v oblasti automatickho zen, napklad pro inteli-

gentn zen budov za elem sniovn nklad na provoz a zvyovn komfortu.

Tato prce se zabv monostmi vyuit Azure Machine Lear