Transcript
Page 1: Misha Bilenko @ Microsoft

http://azure.com/ml

Page 2: Misha Bilenko @ Microsoft

-

-

-

-

Page 3: Misha Bilenko @ Microsoft
Page 4: Misha Bilenko @ Microsoft
Page 5: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

Page 6: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

Page 7: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

Page 8: Misha Bilenko @ Microsoft
Page 9: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

1

Page 10: Misha Bilenko @ Microsoft
Page 11: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

1

2

Page 12: Misha Bilenko @ Microsoft
Page 13: Misha Bilenko @ Microsoft

Mark Ready for Production

Ready

For Production

Sent to

Deploy

Published

To Marketplace

Page 14: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

1

2 3

Page 15: Misha Bilenko @ Microsoft
Page 16: Misha Bilenko @ Microsoft
Page 17: Misha Bilenko @ Microsoft

Save

// Call the Marketplace api to get the Frequently Bought Together products for this product

var prediction = GetJsonObject ("https://api.datamarket.azure.com/ data.ashx/ama/mba-preview/v1/GetItemSetForItem?key='demo'&item='" + Model.Id + "'","AccountKey", "Q6pX0DpKFsoFSgbxstlgzo1wtmCMQhr0Kf4rky MuTVQ");

// Get the product information from the Databasevar boughtTogether = (prediction != null) ? GetProducts((object[]) prediction.ItemSet) : new List<ProductOverviewModel>();

}

<!-- Apply HTML template for the Frequently Bought Together Products -->@Html.Partial("FrequentlyBoughtTogether", boughtTogether)

Page 18: Misha Bilenko @ Microsoft

>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]>>//[Aeoriuapakfajoerauiepraup]<<//4348957t894qfo<<//[http://insideapps.//]//>>p//color{//yellow//]

Page 19: Misha Bilenko @ Microsoft

-

- Product: Introducing Azure ML

-

-

Page 20: Misha Bilenko @ Microsoft
Page 21: Misha Bilenko @ Microsoft

http://azure.com/ml

Page 22: Misha Bilenko @ Microsoft
Page 23: Misha Bilenko @ Microsoft

Develop Model Deploy Model

Page 24: Misha Bilenko @ Microsoft

Classification Regression Recommenders Clustering

• State of art ML Algorithms

Page 25: Misha Bilenko @ Microsoft

Dataset 1, called as:

maml.mapInputPort(1)

Dataset 2, called as:

maml.mapInputPort(2)

Contents of optional Zip port are in ./src/, called as:

source("src/yourfile.R");load("src/yourData.rdata");

Output data.frame, called as:

maml.mapOutputPort

R Device output – standard output and graphics

Page 26: Misha Bilenko @ Microsoft

Get/Prepare Data

Build/Edit Experiment

Create/Update Model

Evaluate Model Results

Publish Web

Service

Build ML Model Deploy as Web ServiceProvision Workspace

Get Azure

Subscription

Create

Workspace

Publish an App

Azure Data

Marketplace

Page 27: Misha Bilenko @ Microsoft

-

-

- Backend: Many-{task,algo,platform} ML toolbox

-

Page 28: Misha Bilenko @ Microsoft

-

-

-

-

-

-

Page 29: Misha Bilenko @ Microsoft

-

-

-

-

-public interface IPredictor<in TFeatures, out TResult>

public interface ITrainer<in TDataSet, out TPredictor>

where TPredictor : IPredictor

Page 30: Misha Bilenko @ Microsoft

-

-

-

-

--

-

-

Page 31: Misha Bilenko @ Microsoft

-

-

-

- Language: Net# for Deep Neural Net Topologies

Page 32: Misha Bilenko @ Microsoft
Page 33: Misha Bilenko @ Microsoft

-

-

-

-

-

-

-

-

Page 34: Misha Bilenko @ Microsoft

Convolutional

1 hidden layer, fully-connected

Input: 28x28

Hidden: 50

Output:

10

2 hidden layers, fully-connected

Input: 28x28

Hidden: 100

Output: 10

Hidden: 200

28

28

5

5

Conv1

Image

24

24

2

2

Pool112

12

8Conv28

22

Pool24

4

5

5

32

Hidden: 120

Output: 10

32

Page 35: Misha Bilenko @ Microsoft

input Image [28, 28];

hidden H1 [200] from Image all;

hidden H2 [100] from H1 all;

output Output [10] softmax from H2 all;

Image: 28x28

H1: 200

H2: 100

Output: 10

Page 36: Misha Bilenko @ Microsoft

const {kernelSize1 = 5; mapCount1 = 5}

input Image [28, 28];

hidden Conv1 [mapCount1,24,24] from Image convolve {KernelShape = [kernelSize1, kernelSize1]; MapCount = mapCount1; InputShape = [28, 28];

};

hidden Pool1 [5, 12, 12] from Conv1 max pool {KernelShape = [1, 2, 2]; Stride = [1, 2, 2]; InputShape = [mapCount1, 24, 24];

};

...

hidden Hidden [120] from Pool2 all;

output Output [10] softmax from H2 all;

28

28

5

5

Conv1

Image

24

242 2

Pool11212

8

Conv28

2

2

Pool24

4

5

5

32

Hidden: 120

Output: 10

32

Page 37: Misha Bilenko @ Microsoft

••

• Azure ML Net#

Page 38: Misha Bilenko @ Microsoft

-

-

- Backend: Many-{task,algo,platform} ML toolbox

-

Page 39: Misha Bilenko @ Microsoft

- http://azure.com/ml

-

-

http://aka.ms/azuremlawards [email protected]


Top Related