Section: New Results
FocusLab platform: software part
Participants : Brigitte Trousse, Yves Lechevallier, Semi Gaieb, Xavier Augros, Guillaume Pilot, Florian Bonacina.
FocusLab v1.3 (software component) done inside the ELLIOT project (cf. section 184.108.40.206 ) and for the purposes of the CPER Telius (cf. section 8.1.5 ) corresponds to the design and the implementation of a set of web-services providing basic and advanced functionalities for data analysis and some other tools supporting the living lab process.
In this version, five data analysis web services are proposed including three generic web services: a classical linear regression and two AxIs methods:
SMDS/SCDS  : SCDS (Sequence Clustering in Data Stream) is a clustering algorithm for mining sequential patterns (Java) in data streams developed by A. Marascu during her thesis. This software takes batches of data in the format "Client-Date-Item" and provides clusters of sequences and their centroids in the form of an approximate sequential pattern calculated with an alignment technique. We propose in this version to return the apparition frequency (min, max, average, slope) of a sequential pattern from data streams (SCDS algorithm) (see references
GEAR for data streams compression  ,  ,  ,  : GEAR (REGLO in french) is an implementation of the history management strategy proposed in Marascu’s thesis . It takes a set of time series and provides a memory representation of these series based on a new principle, where salient events are important (in contrast to the recent events of decaying models) .
Other data analysis services and tools have been added for Living Labs needs. We propose also two clustering methods which must be downloaded as standalone software and used for mining data from living labs:
ATWUEDA (Axis Tool for Web Usage Evolving Data Analysis) for Analysing Evolving Web Usage Data (Da Silva ‘thesis 2009  ,  ,  ,  ) was developed in Java and uses the JRI library (http://www.r-project.org/). The ATWUEDA tool is able to read data from a cross table in a MySQL database, split the data according to the user specifications (in logical or temporal windows) and then apply the proposed approach in order to detect changes in dynamic environment. Such an approach characterizes the changes undergone by the usage groups (e.g. appearance, disappearance, fusion and split) at each timestamp. Graphics are generated for each analysed window, exhibiting statistics that characterizes changing points over time. This application for the next experiment of Green services use case is under study.
MND method (Dynamic Clustering Method for Multi-Nominal Data)  : The proposed MND method (developed in C++ language) determines iteratively a series of partitions which improves at each step the underlying clustering criterion. The algorithm is based on: a) Prototypes for representing the classes; b) Representation space; c) Proximities (distances or similarities) between two individuals; d) Context-dependent proximity functions for assigning the individuals to the classes at each step. The clustering criterion to be optimized is based on the sum of proximities between individuals and the prototype of the assigning clusters.
This method has been also successfully applied on Web logs in 2003. This year we improved our code and tested it on IoT data (temperature) issued from the ECOFFICES project (cf. sections 6.5.3 and 8.1.3 ).
The application of the services provided by FocusLab 1.3 and other AxIS data mining methods for the purposes of ELLIOT use cases and other experimental projects are under study.