leodido / m.sc-thesis

My M.Sc thesis on Continuous time Bayesian Network Classifiers (CTBNCs)

CTBN

This repository contains the tex source of my M.Sc. thesis on Continuous time Bayesian Networks (CTBNs).

Topics

• Theory
  • Continuous time Bayesian Networks
  • Continuous time Bayesian Network Classifiers
  • Continuous time Bayesian Network structure learning
• Development
  • Analysis of the relative R packages developed
  • Analysis and description of the .dll extension developed to generate traffic dataset from TSIS
• Case Studies
  • Tutorials about the main R package
  • Experimentations and performance analysis

Tex

Based on a customization of the ClassicThesis style, obtained through the ArsClassica package and other customizations made by me.

How to compile

Prerequisites

• Texlive (or similar) complete distribution
• ArsClassica package installed
• Sublime Text 2
• Sublime Text 2 LaTeXTools package installed
• Graphviz installed (dot utility have to be in the $PATH variable)
• Compiling: pdflatex with option -shell-escape

Steps

1. Use the LaTeX sublime text 2 build tool provided to compile tesi.tex
2. Open tesi-frn.tex and compile it to generate the frontispiece related tesi-frn.pdf file
3. Compile again tesi.tex to insert the frontispiece
4. Run makeindex -s classic tesi in the shell (into the project directory) to build the definitive (and better) arclassica analytical index^(*)
5. Recompile tesi.tex

Note: tesi.sublime-project file set tesi.tex file as the tex root file for the project.

_{(*) Thi step is required each time a new indexed term (or keyword) is added.}

m.sc-thesis's Issues

Analytical index classic style

Compile and use the classic style for analytical index.

It is included by arsclassica package.

Remove Sublime Text 2 workspace file from repository

Subsection about CTBNC learning

Analytical index

What is it?

An analytical index is an alphabetic index of argument keywords.-

To do

• insert common keywords
• categorize common keywords where required
• build to arsclassica analytical index with its classic.ist included file

Section "Definizioni Preliminari"

CTBNC learning listing

Fix README prerequisites

Introduce the structural learning

What is structural learning?

State-of-art of structural learning of Bayesian Networks (see Friedman1997 and Langley1992)

Mathematical notation of the parent set instances

Correct the mathematical notation to represent the instances of a parent set.

Notation:

Section about the score function

Introduction

Inference listing

Talk about the not-fixed time granularity

E.g. "Le CTBN rappresentano esplicitamente il tempo: non è necessario scegliere la granularità temporale a priori (contro DBNs)".

Fix errors signaled by draft modality

Concepts

This issue is a memo/todo list.

Technicalities

• hill climbing: score caching/memoization

Concepts

• explain computed evaluation metrics (see google doc document already prepared)

Add other when founded.

Fix errors signaled by draft modality

Example about CIM

As in Nodelman2007

Is used to make it more understandable to the reader.

Solve question about J-evidence-stream term and J-time-segment

Not an italian term.

So:

• it is possible to translate it?
• how to use in the corpus?
  • quoted (e.g. "J-evidence-stream")
  • guillemeted (e.g. «J-evidence-stream»)
  • emphatized (e.g. J-evidence-stream)

Section "Apprendimento"

CTNBC learning listing

Add a subsection related to CTBN semantic?

See Nodelman2007, 3.2*.

To say absolutely somewhere: "Le variabili casuali devono rappresentare aspetti distinti del mondo perciò si assume che esse non possano effettuare transizioni nello stesso istante."

Is this addition really needed or this information can be said elsewhere?

Citation:
"It is important to note that wemake a fundamental assumption in the construction of the CTBN model: two variables cannot transition at the same time. This can be viewed as a formalization of the view that variables must represent distinct aspects of the world. We should not, therefore, model a domain in which we have two variables that functionally and deterministically change simultaneously".

Subsection "Processi di Markov"

Section "Rappresentazione"

Cite the point evidence

CTBN allow to have point evidence (reference: paper of Prof. Stella), say it.

Subsection "Bayesian Network"

Describe hill climbing greedy search

Insert references in section 2.2

Missing many references. Particularly in subsection "Apprendimento e Inferenza".

E.g. Pearl, Heckermann1995, Yedidia2000 (for generalized belief propagation).

Comment learning listings

Fix errors signaled by draft modality

Split likelihood formulas on more lines

Insert keywords for analytical index

Allow multi-line chapter titles

Talk about computation of CIMs from hyper-parameters

Make sure it is clear how to calculate the CIMs starting from the hyper-parameters.

Add section about CTBN inference?

See Nodelman2007, sections 8.1 and 8.2.

Rewrite of subsection "Apprendimento e Inferenza"

Particularty the part about Structural learning (see Scutari website about its bnlearn R package).

Fix frontespiece

Fix the logo and recompile.

Add subsection talking of bayesian approach?

See Nodelman2007, section 5.1.1.

Verify if this is needed in the future (e.g. for the chapter 4).

Specify what means index k

Subsection "Definizioni Preliminari", definition of "Matrice di intensità condizionale": what means index k into the matrix? Specify it.

Review all chapter listings

Comment inference listing

Describe the CTNB classifier.

Subsection "Tree Augumented Naive Bayes"

Highlight the lack of acyclicity constrait entails

• optimization done separately for each node attribute.

Mark keywords for analytical index

• follow best practices: link
• generate italian sorting with texindy: link

Support graphviz files inclusion

Section about CTBNC inference

Modify git ignore list

Include *.sublime-workspace file

Configure the listing objects

Arrange an analytical index

An analytical index is an alphabetic index of argument keywords.

Section about CTBNC

Continuos time Bayesian Network classifiers.

Insert CTBN image/s

Create a .png image (with the apposite R script) containing an example CTBN and insert in related subsection of chapter 2.