Sunteți pe pagina 1din 64

IBM Watson Application Developer

C7020-230 Certification Exam V3


December 2016

Kip Harris
Business Partner Technical Enablement
IBM Cloud Partner Services Practice

2015 IBM Corporation


Gartners Top 10 Strategic
Trend No. 1: Technology Trends for 2017
AI & Advanced
AI and machine learning (ML), which
Machine Learning include technologies such as deep
learning, neural networks and natural-
language processing, can also
encompass more advanced systems
that understand, learn, predict, adapt
and potentially operate autonomously.
Systems can learn and change future
behavior, leading to the creation of more
intelligent devices and programs.

Source: gartner.com article , published 18 Oct 2016, retrieved Dec 2016

2015 IBM Corporation Page 2


IBM Watson Developer Cloud

2015 IBM Corporation Page 3


C7020-230 IBM Watson V3 Application Development

2015 IBM Corporation Page 4 Source: IBM Professional Certification Program


Agenda
Time Topic

09:00 Introductions

09:15 Neural networks

09:45 Exam objectives, Sections 1 & 2

10:45 Exam objectives, Section 3

11:30 Exam objectives, Section 4 & 5

Noon Lunch

Afternoon Hands-on tutorials, students explore demos and SDK

4:30 Last departure

2015 IBM Corporation Page 5


Fundamentals of
Cognitive Computing

Exam objectives, section (1.)

2015 IBM Corporation Page 6


Cognitive systems:
1.a. Define the main
Learn at scale
characteristics of a Reason with purpose
cognitive system Interact with people naturally

Cognitive systems also:


Work with unstructured data, as well
as with structured data
Employ probabilistic algorithms
Source: Computing, Cognition and the Future of
Knowing, John Kelley, p. 2 Augment, not replace, our capabilities
2015 IBM Corporation Page 7
Study guide suggests reviewing:
1.b. Explain neural
Role of synapse & neuron
networks
Understand weights and biases
List various approaches to neural
networks
Forward, backward propagation
Gradiant descent
Activation fns in gradiant descent

Sources:
Sigmoid
Presentation Deep Neural Networks, Tara Sainath
Hyperbolic tangent
2015 IBM Corporation Page 8
Supervised
1.c. Explain machine Characterized by training activity
Sample {Input, target or labeled
learning technologies Output} tuples are provided
Applications: Classification,
Regression / Prediction

Unsupervised
No target outputs, example labeling,
etc.
Applications: Data mining, Cluster
analysis, Outlier Detection
Sources URLs in Study Guide
List of Machine Learning Concepts, Wikipedia
Ref to Study Guide for sub-
A Tour of Machine Learning Algorithms, website, categorizations
retrieved Dec 2016
2015 IBM Corporation Page 9
Small sampling from Study Guide:
1.d. Define a common
Customer call center
set of use cases for
Agent assist for QA support
cognitive systems Ticket routing based on all kinds of
criteria

Expert advisor
Physician assistant

Insights from social media

Many more in documentation on


Sources:
Services, in SDKs, Starter Kits, and
2015 IBM Corporation Page 10
Watson Academy
Measures for evaluating Classification algorithms
1.e. Define
Precision
Precision, % of retrieved results that are relevant
Positive predictive value
Recall, and
Recall
Accuracy % of relevant results that are retrieved
Sensitivity or Completeness

Accuracy
% of correctly classified items
Intuitively, inverse of the error rate
Source:
Precision and Recall, Wikipedia,
retrieved 12/2016
See also Watson Knowledge Studio Glossary
2015 IBM Corporation Page 11
1.e. Precision, Recall,
Accuracy (cond)
Precision and Recall often related
Inverse relationship
Cited as a pair: precision x at recall y

Precision and Recall often related


Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
Accuracy = (TP + TN) / (TP+TN+FN+FP)

Sources:
Precision and Recall, Wikipedia, retrieved 12/2016
Graphic image from Wikipedia, retrieved 12/2016.
2015 IBM Corporation Page 12
Training data
1.f. Importance of
Known, good inputs and outputs
separating training, Used to determine parameters on
classifier
validation and test data
Validation data
Used to access performance of
initial training
Select among candidate
algorithms, final tuning
Sources:
[1] stats.stackexchange.com Test
[2] stackoverflow.com
Assess final performance

2015 IBM Corporation Page 13


Recap Accuracy definition
1.g. Measure
Importance of sample size
accuracy of a
Rationale separation of training,
service validation, test data
Risks of failing to do so

Recognizing when to stop training

Sources: exam Study Guide

2015 IBM Corporation Page 14


This objective explores familiarity with
1.h. Define types of Watson Knowledge Studio (WKS)
WKS is an authoring tool for learning models
entities, relationships Define linguistics of specialized industry
and co-references Model consumed by Alchemy Language and
other Watson services

Terms entity and relationship


Analogous to database E-R modeling

Sources: co-reference is two equivalent entities in


Watson Knowledge Studio glossary same text
Watson Knowledge Studio overview
Alchemy Language service documentation E.g. she taught herself

2015 IBM Corporation Page 15


Class
1.i. Define
One instance of a Classification
Intents and Is a statistical clustering of data
E.g. grouping of words in a natural language,
Classes or audio sounds from speech

Classifiers
Module within a cognitive services that perform
classification
Classifiers generally require training

Intent
A term used to describe Classes defined for
text interactions with end users
2015 IBM Corporation Page 16
Use cases
of Cognitive Services

Exam objectives, section (2.)

2015 IBM Corporation Page 17


Discussion
2.a.
Language, Image, Speech, Data
Select cognitive Generic services per Study Guide

technologies fit for Explore docs and demos using


these resources:
use with given use-
WDC starter kits (next topic 2.b.)
case & data format discuss appropriate application

Intended Use on top doc page of


every service (one sample here)
Sources: Study Guide, WDC, Watson Academy
Watson Academy > Intro to WDC
2015 IBM Corporation Page 18
> Learning Lab Use Cases
Starter Kits are organized by use
2.b. Use cases of cases. Each kit comes with:
services found in the Scenario description
Hosted, running demonstration
Application Starter Source code
Kits See also
Usage notes on each service in
the Node SDK on Github
Study Guide notes for Section 2a

Sources:

2015 IBM Corporation Page 19


Multiple kits using:
2.b. (cond) AlchemyData News
Services used in AlchemyLanguage
Natural Language Classifier
Starter Kits* Personality Insights
Retrieve and Rank
Tone Analyzer
Tradeoff Analytics

Single instances of kits using:


Conversation
Speech to Text, Text to Speech
*Note: retrieved 01 Dec 2016
Language Translator
Application Starter Kits on WDC Visual Recognition
2015 IBM Corporation Page 20
Conversation service (GA 7/2016)
2.c. Combines Dialog and NLC services
Dialog (now retired) controls flow
Describe the Watson NLC determines Intent from text entered by user

Conversational Agent Discussion


Superior user experience from natural language
interface to your application
Enables construction of virtual agents
Can automate responses to FAQ
Can monitor social media for instant engagement
with community

Resources
WDC > Conversation
Text Message Chatbot starter kit

2015 IBM Corporation Page 21


Integration with these enormous
2.d. Use cases for information sources allows citizens
integrating external to draw sophisticated insight using
natural language interactions
systems such as
Twitter, Weather API

Sources:

2015 IBM Corporation Page 22


Fundamentals of
IBM Watson Developer Cloud

Exam objectives, section (3.)

2015 IBM Corporation Page 23


Requires training
3.a. Conversation
Natural Language Classifier
Services which Rank and Retrieve (ranker)

require training Pre-trained, but provide option for custom


model or other extensions
AlchemyLanguage
Visual Recognition
Speech to text, Text to speech
Language Translation

Pre-trained, no custom model or config


AlchemyData News
Tone Analyzer
Sources:
WDC service and API documentation retrieved Dec 2016 Personality Insights

2015 IBM Corporation Page 24


3.b. Text classification
examples using
Natural Language
Classifier Discussed in Section 3.e.

2015 IBM Corporation Page 25


Watson Developer Cloud (WDC)
3.c. Watson SDKs
Documentation, Docs, Starters,
Community

Watson SDKs on GitHub


node-sdk / JavaScript
iOS-sdk / Swift
python-sdk / python
java-sdk / Java
unity-sdk / c
android-sdk / java

2015 IBM Corporation Page 26


3.d. REST APIs
provided by WDC
services
Discussed in Sections 3.e. 3.j.

2015 IBM Corporation Page 27


Classifies short, natural language
3.e texts
Natural Language Uses deep learning technology
Classifier service : Requires training, but simpler than
traditional neural network training
Explain, configure
Developers leverage built-in
classifications
Service can successfully classify
inputs not contained in training data
All developers benefit from pre-
Sources:
NLC documentation on WDC
training with public corpuses
NLC API Reference
Service is continually learning
2015 IBM Corporation Page 28
Exam Study Guide suggests these
3.e Spam email detection
Customer message routing
Natural Language Academic paper classification into technical fields
of interest
Classifier: Use cases Forum post classification into categories of interest
Patient reports for escalation and routing based on
symptoms
News article analysis
Investment opportunity ranking
Web page topic analysis

Additional use cases found in these Resources


(links in Appendix)
Sources: Service documentation on WDC
NLC documentation on WDC
WDC SDKs
Application Starter kits
2015 IBM Corporation Page 29
2015 IBM Corporation Page 30 Source: "Using Your Own Data" in NLC documentation on WDC
3.f.
Describe the process for training a
Visual Recognition classifier

service : Explain, Identifying images with a specified


classifier
configure Reference custom classifier(s) when
invoking Classify method

Capabilities of Face Detection /


Recognition

Capabilities of Natural Scene OCR


Sources:
Visual Recognition documentation on WDC
Visual Recognition API reference on WDC
2015 IBM Corporation Page 31
2015 IBM Corporation Page 32 Source: Visual Recognition documentation on WDC
Personality model, Big Five
3.g. Agreeableness, Conscientiousness,
Extraversion, Emotional Range, Openness
Personality Models
Needs model, 12 total attributes
(used in multiple Includes Excitement, Harmony, Curiosity,
Love, Challenge,
services)
Values model
Self-transcendence, Conservation,
Hedonism, Self-enhancement, Open to
change

Do not confuse these models with


Sources, retrieved Dec 2016 Emotional Scoring (anger, disgust, )
Personality Models, WDC
Personality Insights Basics, WDC
2015 IBM Corporation Page 33
Single operation Request Profile
3.g.
Input
Personality Insights Up to 20 MB content
Plain text, HTML, JSON, CSV formats
service
Output proflie
Scoring for Personality, Needs, and Values
models
Predictions for likely behavior available

Use cases show these predictive models


being used in business context
Sources, retrieved Dec 2016
Ref Use Cases link in the services
Personality Insights Overview for Developers, WDC Detailed Documentation page
Personality Insights Detailed Documentation, WDC
2015 IBM Corporation Page 34
Linguistic analysis determines tone of
3.h text in three dimensions
Tone Analyzer Emotion (anger, disgust, )
Personality traits (Big Five)
service Writing style (confident, analytical, )
Science behind the service

Use cases
Ref service overview & case studies
in WDC documentation

Differences from methods in


Sources: AlchemyLanguage service
Tone Analyzer documentation on WDC

2015 IBM Corporation Page 35


3.h Tone Analyzer usage flow

2015 IBM Corporation Page 36 Source: Tone Analyzer documentation on WDC


Collection of text analysis functions
3.i. that derive semantic information
AlchemyLanguage : Larger collection of disparate features
Explain, execute than other WDC API. Sampling:
Concepts
Entities
Keywords
Taxonomies
Relationships
subject, verb, object
Sources:
Emotional and Sentiment analysis
AlchemyLanguage documentation on WDC
AlchemyLanguage API Reference
2015 IBM Corporation Page 37
Keywords, entities, concepts
3.i. (con'd)
Taxonomy vs Knowledge Graph
AlchemyLanguage :
Document-level vs targeted sentiment
Discussion items
Emotional Analysis
anger, disgust, fear, joy, sadness
Use cases in research blog: Another Step
Closer to Building Empathetic Systems

Constructing custom models with


Watson Knowledge Studio
Sources:
AlchemyLanguage documentation on WDC
AlchemyLanguage API Reference
2015 IBM Corporation Page 38
Retrieve part built on Apache SOLR
3.j.
SOLR knowledge needed
Retrieve
Ranking part requires training examples
and Train with ground truth
Importance of iterating
Rank
Enhanced info retrieval combines Document
Conversion w R&R; improves consumability.

IBM Watson Retrieve and Rank Handbook


Short, well done slide deck
Sources: Good guidance, references, graphic explanations
Retrieve and Rank documentation on WDC

2015 IBM Corporation Page 39


2015 IBM Corporation Page 40
Source: IBM Watson Retrieve and Rank Handbook
Developing Cognitive Applications
using IBM Watson Developer Cloud

Exam objectives, section (4.)

2015 IBM Corporation Page 41


If trainable service or custom model
4.a Call a Watson Collect sample input
Sample size dependent on service
service to analyze Manually classify sample input
Perform training (create model)
content
Generic steps
(Generic steps:) Create service instance in Bluemix
Obtain credentials from instance
Select desired service operation
Understand required parameters and input /
output objects
Create service invocation using RESTful API

Recommend
Walk these steps for Alchemy Lanaguage, NLC,
and other services
2015 IBM Corporation Page 42
Conversation service product-izes Dialog + NLC
4.b. design pattern
Dialog retired, NLC remains available separately
Implementation of a
1. Configure workspace
Conversational Agent Define dialog flows, intents, entities

2. Connect application to workspace using


REST API

3. Integrate application with other systems as


required by use case
Back-end systems
Chat service
Social media
Source: Watson Conversation Service overview

4. Improve accuracy with usage


2015 IBM Corporation Page 43
4.b. Conversational Agent implementation (cond)

2015 IBM Corporation Page 44


Engage user and detect intent
4.c. and 4.d. Converatation svc, for user engagement + intent
AlchemyLanguage, for entity extraction which can be

Composing Watson used to discern additional context

Brand monitoring
services Alchemy Data News for content
Alchemy Language to detect sentiment

Video analysis (for tone, content)


Text to speech on audio track
AlchemyLanguage or Tone Analysis for sentiment

Q&A assistance
Conversation or NLC, for intent
NLC for FAQ, R&R for long tail

Sources: C7020-230 Study Guide See also: Application Starter Kits


Many compose services
2015 IBM Corporation Page 45
4.c,d. Composing Watson services (cond)

2015 IBM Corporation Page 46


Source: IBM Watson Retrieve and Rank Handbook
4.c,d. Composing Watson services (cond)

Source: IBM Watson Retrieve and Rank Handbook


2015 IBM Corporation Page 47
Implicit assumption of Bluemix Cloud Foundry
4.e. application

Deploy an No different than standard Bluemix development


WDC services follow standard Cloud Foundry
application which
Some services provide new option for credential
management
uses an IBM Covered in next section

Watson service Introductory level Bluemix training covers skills


being assessed here
Exam seems to focus on command line interface
Can use other user interfaces to Bluemix
E.g. Bluemix GUI works equally well

2015 IBM Corporation Page 48


Administration and DevOps

Exam objectives, section (5.)

2015 IBM Corporation Page 49


Credentials for WDC services are
5.a. obtained through instance of Watson
Service credentials service in Bluemix

Two models for communications


between your cognitive application
and your Watson service(s)
1. Relay request All service
invocations pass through Bluemix
server side component
2. Direct interaction Bluemix service
used only to obtain credentials and
Sources: Programming models for Watson services security tokens.

2015 IBM Corporation Page 50


5.a. (con'd) Two interaction models

red
Proxy, or
Relay request

blue
Direct interaction

(Protocol description in
slide notes and in Source
listed at left bottom.)

Source: Programming models for Watson services

2015 IBM Corporation Page 51


Implicit assumption of Bluemix
5.b. + 5.c: Resource runtime
utilization for a Bluemix Standard Bluemix story
hosted application Simple tools available in application
using IBM Watson dashboard

services* Bluemix Monitoring and Analytics


Bluemix service
*Note: Exam Objectives and Study Guide ask about Provides performance monitoring,
resource utilization for a "service" vs "application"
convenient access to application
Sources: runtime logs, more

2015 IBM Corporation Page 52


5.b. + 5.c: Resource Bluemix Monitoring and Analytics
service provides dashboards
utilization (cond) Performance
Availability
Application events
Addl analysis features
depending on plan

Sources: Bluemix Monitoring and Analytics service overview

2015 IBM Corporation Page 53


5.b. + 5.c: Resource utilization (cond)

2015 IBM Corporation Page 54 Sources: Bluemix Monitoring and Analytics service overview
Standard Bluemix story
5.d. Examine error logs
Demonstration of log access
provided by Bluemix
application runtimes*

*Note: Exam Objectives and Study Guide ask about logs


generated by "service" vs "application runtime.

Sources: Monitoring and Logging with Cloud Foundry

2015 IBM Corporation Page 55


Suggested hands-on exercises

2015 IBM Corporation Page 56


Ideally, want to gain experience with
Hands-on with
Training a service / Working with
Watson Developer custom models
Working directly with the API
Cloud services Working with each of the four major
service categories
Language, Speech, Vision, Data

Universal prerequisite: Bluemix login

Most Tutorials directly exercise APIs


Require local install curl tool

Following slides offer spectrum of


2015 IBM Corporation Page 57
options
Conversation
Guided labs Does not require installation of any local tools

Natural Language Classifier


Some redundancy to Conversation, but work at API
level

Rank and Retrieve


Work with document collection referenced by tutorial
Affinity with Document Conversion service, tutorial
here

AlchemyLanguage
Note: Unless otherwise noted, these Tutorials
require curl to generate HTTP requests. WebSphere Knowledge Studio tutorial also available

Tone Analyzer

2015 IBM Corporation Page 58


Application Starter Kits
Application Starter
Running application
Kits and SDKs Source code

SDKs in Node, iOS, Python, Java,


more
SDKs contain runnable examples

2015 IBM Corporation Page 59


Additional Service catalog on WDC

demonstration Navigate through Learn more link


Running demonstration applications
applications for most / all services

2015 IBM Corporation Page 60


Appendix

2015 IBM Corporation Page 61


Watson SDKs on Github
Resources
API Documentation

Application Starter Kits

Watson Academy
IBM external
IBM internal

IBM Exam C7020-230 Study Guide

2015 IBM Corporation Page 62


Legend to abbreviations used in this presentation
Abbreviation Unabbreviated meaning

Bmx IBM Bluemix

CF Cloud Foundry

NLC IBM Watson Natural Language Classifier service

svc Service

Watson IBM Watson

WDC IBM Watson Developer Cloud

WKS IBM Watson Knowledge Studio

2015 IBM Corporation Page 63


2015 IBM Corporation Page 64

S-ar putea să vă placă și