The software management and engineering in the AI-oriented projects tutorial

SETN 2010 - TUTORIAL

The software management and engineering in the AI-oriented projects

www. .co.uk


Dr W Pietruszkiewicz

SDART Ltd



www. .co.uk

Plan of presentation

Introduction

Software project management

Software design

AI implementation

Examples of AI software



www. .co.uk

Introduction



www. .co.uk

Artificial Intelligence



www. .co.uk

Project success



www. .co.uk

Research oriented software

Time and budget aren’t usually fixed – it is

possible to extend the research

Resources are flexible

People mainly have the personal interests (fame,

dissertations, publications …)

Quality (functionality) is the main concern



www. .co.uk

Product oriented software

Time and budget are usually fixed

Quality is adjusted to the available budget and

schedule

People mainly have the business interests

Resources aren’t flexible



www. .co.uk

AI & software in projects

Proper management

Struggle



www. .co.uk

This tutorialReasons to propose this tutorial:

•There exist the software projects without AI but I’m not

sure if there exist AI projects without the software

(embedded count too!) …

•The knowledge about the software management is

unpopular among AI researchers

•At SDART we deal with the AI-oriented software projects



www. .co.uk

This tutorial

•Will:• present a selection of the popular techniques and methods for software

development

• evaluate them from AI perspective

• propose what could be used

Will not:• say what have to be used

• be a guide ready to be applied step-by-step



www. .co.uk

Software project management



www. .co.uk

AI & soft – PM nightmare

In the project management two types of projects

are the symbols of unpredictability:

Research projects

…. software projects

What could we do dealing with an AI-oriented

software projects? Professionally manage them!



www. .co.uk

Software development schema



www. .co.uk

DM management

Data mining management seems to be more

organised than AI

There are different models of DM organisation:

CRISP-DM

SEMMA

DMAIC/DMADV



www. .co.uk

CRISP-DM



www. .co.uk

SEMMA



www. .co.uk

DMAIC&DMADV



www. .co.uk

Management model

Can’t we use any DM model to properly manage

the AI&software projects? Yes we could …. only if

the software does not constitute any significant

part of the project. Meanwhile …

AI projects require software to research

These projects require software to use AI



www. .co.uk

Common risk sources



www. .co.uk

Risk strategies

Avoidance

Retention

Sharing

Reduction



www. .co.uk

Software qualityThere are many software quality factors, among

them:



www. .co.uk

Software quality



www. .co.uk

Software qualityIf an application is a product users perceive only

its „front” parts:

•GUI

•Errors

•Documentation



www. .co.uk

Software metrics

Tom DeMarco said „You can’t control what you

can’t measure”

Software metrics measure some properties of a

software or its specification e.g. complexity,

quality, efficiency or size



www. .co.uk

Software estimation

Estimation problem

I think I’ve

forgotten

something, but

what was it???



www. .co.uk

Methodology & Life cycle

Software methodology defines the structure,

plan and management of developed software

Methodology contains development philosophy

and tools, methods and models supporting the

process

Life cycle model defines the phases and activities



www. .co.uk

M&LC– staged delivery

Requirements

Design

Implementation

Verification

Maintenance



www. .co.uk

M&LC - spiral

Spiral model – Boehm 1988



www. .co.uk

M&LC – staged deliveryConception

Analysis

Desing

Stage 1

Stage 2

Stage n



www. .co.uk

M&LC – design to scheduleConception

Analysis

Desing

Stage 1

Stage 2

Stage n

Software release

Time? Costs?



www. .co.uk

M&LC - AgileThe Agile development is based on the iterative model

Initial planning

Planning

Requirements Analysis and design

Implementation

Deployment

TestingEvaluation



www. .co.uk

M&LC - XP

Extreme programming is an Agile implementation:

Programming in pairs

Extensive code review

Avoids the programming required in the future

Automated tasks

Nothing is fixed i.e. everything could be redone



www. .co.uk

M&LC - SCRUM

Project is being developed by Pigs & Chickens

Pig is committed to the project i.e. ScrumMaster,

Team, ProductOwner

Chicken is involved i.e.: Stakeholder, Manager



www. .co.uk

M&LC - SCRUM



www. .co.uk

M&LC – Unified Process

Is an iterative software development framework



www. .co.uk

M&LC – UP versions

Rational Unifed Process

Agile Unified Process

Open Unified Process

Enterprise Unified Process



www. .co.uk

M&LC– selectionThe factors that influences life cycle selection are:

Size of project

Understood requirements

Desired reliability

Allowed overheads

Progress visibility

Sophisticated management

Allowed corrections



www. .co.uk

M&LC - summary

Waterfall is still the most frequently used model by

the biggest software companies

But they do it preparing „yet another version of X”

AI projects are usually new for you as well as for

everyone else, so do not expect it to be as easy as

software mentioned above



www. .co.uk

Collaboration

Homogeneous research team Heterogeneous software team



www. .co.uk

Project controlling



www. .co.uk

Software design



www. .co.uk

Architecture

Architecture is a high level planning for the

software that does not involve any functional

decisions



www. .co.uk

ViewsThe views are perspectives that are used to represent different aspect of

the system:

Code/module view

Concurrency/process/thread view

Data view

Development/structural view

Functional/logic view

Physical/deployment view

User action/feedback view



www. .co.uk

FrameworksSoftware architecture is defined by various frameworks

defining its necessary elements:

4+1

RM-ODP

DODAF

SOMF

…



www. .co.uk

Frameworks - 4+1



www. .co.uk

Design concepts

Abstraction & refinement – generalisation

Modularity

Hierarchy

Information & functionality enclosure

Structural partitioning – vertical & horizontal



www. .co.uk

UML

Unified Modeling Language:

Object oriented

Unified Booch, OOSE, OMT modelling methods

Currently in 2.2 version – the most of software

supports 2.0



www. .co.uk

UML Diagrams



www. .co.uk

UML Advantages

UML became an industry standard due to:

Visualisation

Documenting

Specifying

Constructing



www. .co.uk

UML ToolsPapyrus – http://www.papyrusuml.org

Poseidon for UML - http://www.gentleware.com

Visual Paradigm for UML - http://www.visual-

paradigm.com

Umbrello UML Modeller – http://uml.sourceforge.net

StarUML – http://staruml.sourceforge.net



www. .co.uk

Supporting tools

Let’s take a look at the supporting tools

If something could be done simpler and at least

with the same quality

Use them and forget the time-consuming

repetive tasks and focus on the difficult ones



www. .co.uk

Version controlling

How to avoid the mess dealing with the different

versions of files in projects?

The source code is constantly changing …

The AI files too …



www. .co.uk

Version controllingVersion controlling allows the development team to manage

different versions of software. They keep track of work and

allow the team to look at the project’s past.

There are 3 kinds of version controlling systems:

Local

Distributed

Client-server



www. .co.uk

Version controlling – local systems

Revision Control System

Source Conde Control System

MKS Implementer

History Explorer



www. .co.uk

Version controlling - distributed

DCVS (Distributed CVS)

Monotone

GNU arch

Plastic SCM

BitKeeper

TeamWare

PC Rep

PC Rep

PC Rep



www. .co.uk

Version controlling – client/server

CVS

SVN

Codendi

ClearCase

Visual SourceSafe

Autodesk Valut

Client

Client

Client

Server Rep



www. .co.uk

Supporting tools & technologies



www. .co.uk

Automated builds - advantagesEfficiency increase

Development process simplification

Increased software quality

Reduced costs

False version removal

Reduction of the development team or task relocation

Project archiving



www. .co.uk

Automated buildsThe automated build tools include:

make

Maven

Ant

MSBuild

Cmake

and many others



www. .co.uk

Automated testing - make

An example of makefile taken from QT library.

PNGMAK=scripts/makefile.linux

all:cd zlib; ./configure; makecd libpng; make -f $(PNGMAK)

clean:-cd zlib; make clean-cd libpng; make -f $(PNGMAK) clean



www. .co.uk

Automated testing - Ant<project name="MyProject" default="dist" basedir="."> <description>simple example build file</description>  <property name="src" location="src"/> <property name="build" location="build"/> <property name="dist" location="dist"/> <target name="init">  <tstamp/>  <mkdir dir="${build}"/> </target> <target name="compile" depends="init” description="compile the source " >  <javac srcdir="${src}" destdir="${build}"/> </target> <target name="dist" depends="compile” description="generate the distribution" >  <mkdir dir="${dist}/lib"/>  <jar jarfile="${dist}/lib/MyProject-${DSTAMP}.jar" basedir="${build}"/> </target> <target name="clean” description="clean up" >  <delete dir="${build}"/> <delete dir="${dist}"/> </target></project>



www. .co.uk

Automated testing - Maven<project xmlns="http://maven.apache.org/POM/4.0.0" mlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd"> <modelVersion>4.0.0</modelVersion> <groupId>com.mycompany.app</groupId> <artifactId>my-app</artifactId> <packaging>jar</packaging> <version>1.0-SNAPSHOT</version> <name>Maven Quick Start Archetype</name> <url>http://maven.apache.org</url> <dependencies> <dependency> <groupId>junit</groupId> <artifactId>junit</artifactId> <version>3.8.1</version> <scope>test</scope> </dependency> </dependencies></project>



www. .co.uk

Automated testing - unit

Test ATest ABCD

vs

Test B Test C Test D

Data

Data Data Data Data

The most popular unit testing tool is xUnit available in many versions for all popular languages



www. .co.uk

Automated testing - system

Automated GUI tests

perceive the application as

a black-box and test it from

the user’s perspective.



www. .co.uk

Code documenting

Automatic API documentation tools:

JavaDoc

DoxyGen

RoboDoc

…

Manual documentation

Automatic documentation

Code Doc CodeDoc



www. .co.uk

Bug trackingIssue/Bug Tracking Systems usage leads to:

Better quality control

Gaining users’ feedback

Better resource allocation

Increased users’ comfort

Better information flow

Feature request available



www. .co.uk

Bug tracking – bug categories



www. .co.uk

Bug tracking

Bugzilla

Flyspray

Mantis

Trac

Redmine



www. .co.uk

Bug tracking



www. .co.uk

AI implementation



www. .co.uk

Implementation

Transforms the design (plan) into the code

Problems:

Technology

Programming language



www. .co.uk

Levels of programming languages



www. .co.uk

1st generation

The code lines are equal to the processor

instructions – one to one

Machine code



www. .co.uk

2nd generation

Symbolic representation

of processor’s

instructions

Assembler language



www. .co.uk

3rd generation

Instructions represent the basic operations

Oriented on programmer

Most popular generation



www. .co.uk

4th generation

Specialised languages oriented on advanced

operations

RAD tools are 4GL build over 3GL

Matlabdisp('Hello, world!')

SASdata _null_; put 'Hello, world!'; run;

Postscript(Hello, world!\n) print



www. .co.uk

5th generation

Represent not algorithms but desired outcome

Logical programming

Programming for non-programmers

Common Lisp(write-line "Hello, world!")

Prolog:- write('Hello, world!'),nl.



www. .co.uk

Hybrid programming – embedding

Hybrid programming (embedding languages)

could be synergic but:

Increases the technological risk

Increases the demand on skilled developers

Blurs the project



www. .co.uk

Programming paradigms

Paradigm is a fundamental style of computer

programming and defines the concept of

programming language

The most of programming languages support

various paradigms (different approaches to

problem)



www. .co.uk

Programming paradigms

The most popular paradigms are:

Agent-oriented

Declarative

Event-driven

Imperative

Metaprogramming



www. .co.uk

Desktop applications

C/C++

Java

C#

Python



www. .co.uk

Web services – server sideJava EE

PHP

Python

Ruby

Perl

.NET

CGI (every language that could be run by server)



www. .co.uk

Web services - RIAsFlash

AJAX

Java SE

JavaFX

FLEX

OpenLaszlo

Silverlight



www. .co.uk

Mobile applications

Java ME

C++

Objective-C



www. .co.uk

Languages popularity

http://www.tiobe.com/index.php/content/paperinfo/tpci/index.html



www. .co.uk

Characteristics - domain

Specialised – good at something but below

average on other things

General purpose – no specific domain and a

compromise on all tasks



www. .co.uk

Characteristics - speed

Compiled faster

Interpreted slower

But:

Compiled start-up slower

Interpreted start-up faster (write-run-test cycle)

JIT compilers for interpreted languages



www. .co.uk

Characteristics - portability

Software could be portable on two levels:

Source code

Binary code

Portability does not relate only to the programming

language but also to used libraries



www. .co.uk

Characteristics - prototyping

Ability to verify the assumption:

Dynamic language

Simple syntax

Efficiency isn’t a main concern

Checks if the idea works



www. .co.uk

Characteristics – target platform

Target platform:

PC

Mobile

Web



www. .co.uk

Computer languages

Scientific perspective Programming perspective



www. .co.uk

Why use a specialised languages

Popular in the research community

Specialized languages

Many libraries/toolkits and examples are

available



www. .co.uk

Why not use them

Specialised not universal

High price

Simple or no GUI

No networking

Research not production languges



www. .co.uk

Important programming features

Dynamic languages:

Do not care about types of data

Focus on the algorithms

But reject a full data control



www. .co.uk


Prototyping:

Checks algorithms

Easily visualise data or make GUI

Speed is measured as the time of coding not

running



www. .co.uk


Scientific libraries:

How often do we construct new algorithms

(Charlie Eppes Syndrome – NUMB3RS)?

The application is more like a large panel building

than brick building – applies to the AI problems as

well as to the „ordinary” soft



www. .co.uk


Visual design (graphical programming):

GUI could be constructed using components

Link the components visually

Evaluate application faster



www. .co.uk

More important part of AI software

KnowledgeModelsRules

Graphs



www. .co.uk

Hybrid software development

Research Development

Ide

a

Pro

duc

t

Replanning& redesign

Technology X

Technology X Technology Y??

???

??



www. .co.uk

Programming libraries – C/C++MLC++ http://www.sgi.com/tech/mlc/

FANN http://leenissen.dk/fann/

Waffles http://waffles.sourceforge.net/

OpenCV ML

http://opencv.willowgarage.com/documentation/cpp/ml._mach

ine_learning.html

dlib C++ http://dlib.net/

VFML http://www.cs.washington.edu/dm/vfml/



www. .co.uk

Programming libraries – Java

Java-ML http://java-ml.sourceforge.net/

Weka Engine

http://www.cs.waikato.ac.nz/ml/weka/

MLJ http://sourceforge.net/projects/mldev/

MALLET http://mallet.cs.umass.edu/

LIBSVM http://www.csie.ntu.edu.tw/~cjlin/libsvm/



www. .co.uk

Programming libraries – C#Infer.NET http://research.microsoft.com/en-us/um/cambridge/

projects/infernet/

AForge.NET http://code.google.com/p/aforge/

C# Neural Network Library

http://franck.fleurey.free.fr/NeuralNetwork/

NeuronDotNet http://neurondotnet.freehostia.com/

NeuroBox Neural Network Library

http://www.cdrnet.net/projects/neuro/



www. .co.uk

Programming libraries – Python

Orange engine http://www.ailab.si/orange/

PyML http://pyml.sourceforge.net/

MDP http://mdp-toolkit.sourceforge.net/

PyBrain http://www.pybrain.org/

LIBSVM http://www.csie.ntu.edu.tw/~cjlin/libsvm/



www. .co.uk

Programming libraries – PHP

FANN PHP Extension

http://fann.sourceforge.net/fann.html

Tremani Neural Network http://www.tremani.nl/

open-source/neural-network/

ANN http://ann.thwien.de/index.php/Main_Page



www. .co.uk

Example I – Java

A simple data mining task

Uses Java-ML library



www. .co.uk

Example II – Python

Adaptive filtering algorithm

Use as a web service (server part)

Use mod_python on Apache



www. .co.uk

Example III – Flex

Particle Swarm Optimisation example

RIA application

Uses a standard Flex without any additional

libraries



www. .co.uk

Thank you. Any questions?Get in touch: www.sdart.co.uk

The software management and engineering in the AI-oriented projects tutorial

Technology

aioriented projects

ukthe software management

ai projects

ukai software

software projectswhat

ai perspective

software embedded count

ai researchersat sdart