Για να δεις στα
ελληνικά
Statistical Analysis of Networks
- Module on Time Series Networks 2018-2019
The module "Time Series Networks"
constitutes part of the course "Statistical Analysis
of Networks" of the postgraduate program
"Web Science"
organized by the Department of Mathematics
of AUTh.
Instructor:
Dimitris Kugiumtzis
Content:
Syllabus, suggested literature
Syllabus:
The module is about the construction of networks from
multivariate time-series, where each time-series defines the node, and the
association of two time-series determines the link between the respective nodes.
The main objective of the module is the study of different
types of association (connection / link) between two time series and the
networks derived from such associations.
The module is divided in three parts and each part is taught
at a 4-hour
course (see Course transperancies).
-
Association networks, correlation networks,
cross-correlation, partial cross-correlation, parametric and non-parametric
tests of significance of correlation, correlation network from time-series.
Examples.
-
Uni-variate time-series (stationarity,
time series decomposition, detrending),
autocorrelation and mutual information, test for independence,
linear autoregressive models.
Bi-variate time-series, cross-correlation
and cross-mutual information. Examples.
-
Linear stochastic and nonlinear
deterministic systems. Correlation measures. Construction of
correlation networks from time series. Examples.
-
Nonlinear and nonlinear causality measures. Connectivity
networks using causality measures on multivariate time series. Applications
to neuroscience and finance.
-
Networkd from time series in Matlab. Practical application
of methods presented in the first four parts using the computational
software Matlab.
Suggested literature:
Suggested books:
-
J. Pearl
(1998), "Probabilistic Reasoning in Intelligent Systems: Networks of
Plausible Inference", Morgan Kaufmann.
-
E.D. Kolaczyk (2009),
"Statistical Analysis of Network Data",
Springer
-
Horvath, S. (2011) Weighted Network Analysis,
Applications in Genomics and Systems Biology. New York:
Springer
-
Fornito, A., Bullmore, E. T. & Zalesky, A. (2016)
Fundamentals of Brain Network Analysis, 1st edn.
Cambridge,
Massachusetts, United States: Academic Press, Elsevier
Suggested edited collection of papers:
-
A.
Reggiani and P. Nijkamp (2006), "Spatial Dynamics, Networks And Modelling
(New Horizons in Regional Science Series)",
Edward Elgar Publishing.
-
T. Gross and H. Sayama (2009),
"Adaptive Networks: Theory, Models and Applications (Understanding Complex
Systems)", Springer.
-
R. Menezes, S. Fortunato, G. Mangioni and V. Nicosia (2009),
"Complex Networks: Results of the 1st International Workshop on Complex
Networks (CompleNet 2009) (Studies in Computational Intelligence)",
Springer.
-
N. Ganguly, A. Deutsch and A. Mukherjee (2009),
"Dynamics On and Of Complex Networks: Applications to Biology, Computer
Science, and the Social Sciences (Modeling and Simulation in Science,
Engineering and Technology)",
Birkhauser Boston.
-
S. K. Dana, P. K. Roy, and J. Kurths, Eds., Complex
Dynamics in Physiological Systems: From Heart to Brain, ser. Series:
Understanding Complex Systems. Berlin / Heidelberg: Springer, 2009
To view the networks, the software
pajek will be used (see
http://pajek.imfm.si)
Course
transparencies
- Course 1:
Introduction to networks, correlation and time series (in PDF)
[last update 8/11/2018]
- Introduction to association networks. Example from
scientometrics.
- Correlation networks. Cross-correlation,
parametric and non-parametric tests of significance of correlation.
Partial cross-correlation and tests of significance. Example for
correlation network from gene regulation of micro-array data.
- Correlation network and time-series. Example of correlation
network from world market indices.
Data files for course 1:
1.
Ecoliv4Build6ex1.xls (excel format),
Ecoliv4Build6ex1.txt
(plain text format): Micro-array data.
Source:
http://m3d.bu.edu/cgi-bin/web/array/index.pl (data base
Palsson03).
2. WorldMarkets.dat
(plain text format). The data are in a matrix form of size 101 x 8,
containing the indices over 101 days from 8 world stock markets
(USA, Australia, United Kingdom, Germany, Greece, Malaysia, South
Africa, Croatia).
Source:
Morgan Stanley Capital International’s market capitalization
weighted index data
3.
WorldIndicesCorrCoeff.net (format suitable for pajek software).
It contains the information of nodes (vertices) and links (arcs) of
the correlation network for the 8 world markets.
4.
WorldIndicesPartCorr.net The same as for
"WorldIndicesCorrCoeff.net" but for the partial correlation network.
- Course 2:
Measure of correlation, complexity and coupling
for time-series (in PDF)
[last update 29/3/2011]
-
Time series, stationarity, time series
decomposition, detrending. Examples.
- Autocorrelation, test for independence, linear
autoregressive models.
Examples.
- Mutual information and test for independence. Examples.
- Bi-variate time-series,
cross-correlation and cross-mutual
information. Examples.
Data files for course 2:
1.
GPIC2001_2005residuals.dat
(plain text format): The monthly General Consumer Price Index for
the period Jan 2001 - Aug 2005.
2. PacketArrival.dat
(plain text format). A trace containing packet arrivals seen on an
Ethernet at the Bellcore Morristown Research and Engineering
facility, regarding LAN traffic, period 11:25 on August 29, 1989.
2. USAreturns.dat
(plain text format). Daily returns of the USA marker index for an
unspecified period of 300 days.
Source:
Morgan Stanley Capital International’s market capitalization
weighted index data
3. Matlab program and functions:
Exercise2.m,
ARm.m,
mutual.m,
portmanteauLB.m.
- Course 3:
Multivariate
time series and networks (in PDF)
[last update 17/3/2010]
- Spurious cross-correlation, vector
autoregressive models and dynamic regression models. Examples.
- Nonlinear dynamical systems, coupled dynamical systems.
Examples.
- Correlation and causality measures and
construction of networks from time series. Examples.
- Course 4 :
Connectivity Networks
and Applications (σε
PDF)
- Survey of measures for causality and
correlation
- Granger causality index and conditional Granger causality
index
- Measures of Granger causality based on information
theory
- Evaluation of Granger causality
measures on known systems (stochastic linear and nonlinear, coupled
Henon maps)
- Application of Granger causality
measures on electroencephalograms
(EEG).
- Course 5: Exercises on
Matlab.
Exercises and
computer programs
To view multiple time series in a graphical Matlab environment
you may use the toolkit Measures of Time Series Analysis (MATS) that can be
downloaded at
http://eeganalysis.web.auth.gr (Software).
All Matlab programs for
the course in a compressed file:
matlab.zip.
Data files
in a compressed file: Data.zip
Course examination
and marking
The mark for the module "Time series networks" will be
determined based on the project the students have to deliver, formed in groups
of up to two students, as well as their presence and discussion at the
presentation of the projects.
The project will be given at the end of the course.
The teams of one or two students should send the presentation
(e.g. in PowerPoint, LaTeX) by e-mail to the teacher (dkugiu@auth.gr)
not later than the last day/night before the date of "examination". The examination will take place on
13/2/2019, at
17.00, at the Math computer lab,
that
is there will be presentations of 8 minutew maximum followed by
questions and discussions.
Updated on 8/11/2018
from Dimitris Kugiumtzis
dkugiu@auth.gr