Machine Learning at Scale TensorFlow in the Cloud Yufeng Guo - - PowerPoint PPT Presentation

▶
machine learning at scale
SMART_READER_LITE
LIVE PREVIEW

Machine Learning at Scale TensorFlow in the Cloud Yufeng Guo - - PowerPoint PPT Presentation

Oct 04, 2023 117 likes •657 views

Machine Learning at Scale TensorFlow in the Cloud Yufeng Guo Developer Advocate @YufengG yufengg.com @YufengG Machine Learning is using many examples to answer questions @YufengG @YufengG Training Prediction many answer examples

slide-1
SLIDE 1

@YufengG

Machine Learning at Scale

TensorFlow in the Cloud

Yufeng Guo Developer Advocate @YufengG yufengg.com

slide-2
SLIDE 2

@YufengG @YufengG

Machine Learning is using many examples to answer questions

slide-3
SLIDE 3

@YufengG @YufengG

Training many examples Prediction answer questions

slide-4
SLIDE 4

@YufengG @YufengG

Training many examples Prediction answer questions

slide-5
SLIDE 5

@YufengG @YufengG

  • Fast, flexible, and scalable
  • pen-source machine learning

library

  • For research and production
  • Distributed training and serving

predictions

  • Apache 2.0 license

https://research.googleblog.com/2016/11/celebrating-tensorflows-first-year.html

slide-6
SLIDE 6

@YufengG @YufengG

A multidimensional array. A graph of operations.

3 + 2 5

slide-7
SLIDE 7

@YufengG @YufengG

TensorFlow Supports Many Platforms...

Raspberry Pi Android iOS TPU GPU CPU

slide-8
SLIDE 8

@YufengG @YufengG

TensorFlow Distributed Execution Engine CPU GPU Android iOS ... C++ Frontend Python Frontend … more coming

slide-9
SLIDE 9

@YufengG @YufengG

TensorFlow Distributed Execution Engine CPU GPU Android iOS ... C++ Frontend Python Frontend ... Layers

Build models

slide-10
SLIDE 10

@YufengG @YufengG

TensorFlow Distributed Execution Engine CPU GPU Android iOS ... C++ Frontend Python Frontend ... Layers

Build models

Estimator

Train and evaluate models

Keras Model

slide-11
SLIDE 11

@YufengG @YufengG

TensorFlow Distributed Execution Engine CPU GPU Android iOS ... C++ Frontend Python Frontend ... Layers Estimator

Train and evaluate models Build models

Keras Model

Models in a box. Takes care of

  • ptimizer, training loop,

learning rate, etc

Canned Estimators

slide-12
SLIDE 12

@YufengG @YufengG

area = real_valued_column("square_foot"), rooms = real_valued_column("num_rooms"), zip_code = sparse_column_with_integerized_feature("zip_code", 10000) classifier = DNNClassifier( feature_columns=[area, rooms, embedding_column(zip_code, 8)], hidden_units=[1024, 512, 256, 128]) classifier.fit(train_input_fn) results = classifier.evaluate(eval_input_fn) print(results)

slide-13
SLIDE 13

@YufengG @YufengG

classifier = DNNLinearCombinedRegressor( linear_feature_columns=[area, rooms, embedding_column(zip_code, 8)], linear_optimizer=tf.train.FtrlOptimizer(learning_rate=0.01, l2_regularization_strength=0.1), dnn_feature_columns=[real_valued_column(area), real_valued_column(rooms)] dnn_optimizer=tf.train.AdagradOptimizer(learning_rate=0.01, initial_accumulator_value=0.1), dnn_activation_fn=tf.nn.relu, dnn_dropout = 0.5, gradient_clip_norm=0.1, hidden_units=[1024, 512, 256, 128]) classifier.fit(train_input_fn) classifier.evaluate(eval_input_fn)

slide-14
SLIDE 14

@YufengG

<Storytime>

slide-15
SLIDE 15

@YufengG @YufengG

Motivation - a "magical" food app

slide-16
SLIDE 16

@YufengG @YufengG

Just Launch and Iterate

  • Naive character matching
  • Say "Fried chicken"
  • Get "Chicken Fried Rice"
  • Oops. Now what?
  • Machine learning to the rescue!
slide-17
SLIDE 17

@YufengG @YufengG

v2.0: memorize all the things!

  • Train a linear TF model
  • Your app is gaining traction!
slide-18
SLIDE 18

@YufengG @YufengG

Problem: Your users are bored!

  • Too many & waffles
  • Show me similar, but different food
  • Your users are picky
slide-19
SLIDE 19

@YufengG @YufengG

v3.0: More generalized recommendations for all

slide-20
SLIDE 20

@YufengG @YufengG

No good deed goes unpunished

  • Some recommendations are "too general"

○ Irrelevant dishes are being sent

  • Your users are still picky
slide-21
SLIDE 21

@YufengG @YufengG

No good deed goes unpunished

  • 2 types of requests: specific and general
  • "iced decaf latte with nonfat milk" != "hot latte with

whole milk"

  • “seafood” or “italian food” or “fast food”
  • How to balance this?
slide-22
SLIDE 22

@YufengG @YufengG

v4.0: Why not both?

slide-23
SLIDE 23

@YufengG @YufengG

Wide & Deep

generalization diversity memorization relevance

slide-24
SLIDE 24

@YufengG @YufengG

Wide and Deep

slide-25
SLIDE 25

@YufengG

</Storytime>

slide-26
SLIDE 26

@YufengG @YufengG

Meet our dataset: US Census Data

  • Task: predict whether the household has an

annual income of over $50K

  • Over 32k training examples
  • Extracted from the 1994 US Census by Barry

Becker.

slide-27
SLIDE 27

@YufengG @YufengG

Meet our dataset: US Census Data

Column Name Type Description age Continuous The age of the individual workclass Categorical The type of employer the individual has (government, military, private, etc.). fnlwgt Continuous The number of people the census takers believe that

  • bservation represents (sample weight). Not used.

education Categorical The highest level of education achieved for that individual. education_num Continuous The highest level of education in numerical form. marital_status Categorical Marital status of the individual.

slide-28
SLIDE 28

@YufengG @YufengG

Meet our dataset: US Census Data

Column Name Type Description

  • ccupation

Categorical The occupation of the individual. relationship Categorical Wife, Own-child, Husband, Not-in-family, Other-relative, Unmarried. race Categorical White, Asian-Pac-Islander, Amer-Indian-Eskimo, Other, Black. gender Categorical Female, Male. capital_gain Continuous Capital gains recorded. capital_loss Continuous Capital Losses recorded.

slide-29
SLIDE 29

@YufengG @YufengG

Meet our dataset: US Census Data

Column Name Type Description hours_per_week Continuous Hours worked per week. native_country Categorical Country of origin of the individual. income_bracket Categorical ">50K" or "<=50K", meaning whether the person makes more than $50,000 annually.

slide-30
SLIDE 30

@YufengG @YufengG

Wide & Deep

generalization diversity memorization relevance Dense/Real Continuous Sparse Categorical

slide-31
SLIDE 31

@YufengG @YufengG

To the code! bit.ly/widendeep-census

slide-32
SLIDE 32

@YufengG @YufengG

tensorboard --logdir=models/

slide-33
SLIDE 33

@YufengG @YufengG

Training many examples Prediction answer questions

slide-34
SLIDE 34

RPC Server

What is Serving?

TensorFlow Data Scientist Model

App Data

slide-35
SLIDE 35
  • C++ Libraries

○ TensorFlow model save / export formats ○ Generic core platform

  • Binaries

○ Best practices out of the box ○ Docker containers, K8s tutorial

  • Hosted Service across

○ Google Cloud ML Engine ○ Internal service

What is TensorFlow Serving?

slide-36
SLIDE 36

@YufengG

Model Creation

slide-37
SLIDE 37

@YufengG @YufengG

slide-38
SLIDE 38

@YufengG @YufengG

slide-39
SLIDE 39

@YufengG @YufengG

slide-40
SLIDE 40

@YufengG @YufengG

slide-41
SLIDE 41

@YufengG @YufengG

slide-42
SLIDE 42

@YufengG @YufengG

slide-43
SLIDE 43

@YufengG @YufengG

slide-44
SLIDE 44

@YufengG @YufengG

slide-45
SLIDE 45

@YufengG

Instance Prediction

slide-46
SLIDE 46

@YufengG @YufengG

{ "age": 25, "workclass": " Private", "education": " 11th", "education_num": 7, "marital_status": " Never-married", "occupation": " Machine-op-inspct", "relationship": " Own-child", "race": " Black", "gender": " Male", "capital_gain": 0, "capital_loss": 0, "hours_per_week": 40, "native_country": " United-States" } { "age": 42, "workclass": " Self-emp-inc", "education": " HS-grad", "education_num": 9, "marital_status": " Married-civ-spouse", "occupation": " Exec-managerial", "relationship": " Husband", "race": " White", "gender": " Male", "capital_gain": 5178, "capital_loss": 0, "hours_per_week": 50, "native_country": " United-States" }

slide-47
SLIDE 47

@YufengG @YufengG

slide-48
SLIDE 48

@YufengG @YufengG { "probabilities": [ 0.11601490527391434, 0.8839850425720215 ], "logits": [ 2.030721426010132 ], "classes": 1, "logistic": [ 0.8839850425720215 ] } { "probabilities": [ 0.9948562383651733, 0.005143760237842798 ], "logits": [

  • 5.2648138999938965

], "classes": 0, "logistic": [ 0.005143760237842798 ] }

slide-49
SLIDE 49

@YufengG @YufengG

Training many examples Prediction answer questions

slide-50
SLIDE 50

@YufengG @YufengG

Training many examples Prediction answer questions

slide-51
SLIDE 51

@YufengG @YufengG

Training many examples Prediction answer questions

slide-52
SLIDE 52

@YufengG @YufengG

Thank you!

@YufengG yufengg.com

Cloud Machine Learning Engine cloud.google.com/ml-engine TensorFlow tensorflow.org To the code! bit.ly/widendeep-census bit.ly/widendeep-code

Resources:

slide-53
SLIDE 53

@YufengG

The End