File server outage

Since a few days (since the 07/24/2017 approximately), the server of the Eric laboratory that hosts the Tanagra project files (software, books, course materials, tutorials...) is idle. After a power outage, there is nobody to restart the server during the summer period. And the server is located in a room in which I do not have access.

So we wait. And it will take a little time, the summer break lasts a month, our University (and Lab) is officially reopened on August 21st! I am sorry for users that work from the documents that I put online. This difficulty is totally beyond my control and I cannot do anything about it.

Some internet users are reported to me the problem. I take the initiative to inform you. As soon as the situation is back in order, I will let you know.

Kind regards,

Ricco (July 27th, 2017).

Interpreting cluster analysis results

Interpretation of the clustering structure and the clusters is an essential step in unsupervised learning. Identifying the characteristics that underlie differentiation between groups allows to ensuring their credibility.

In this course material, we explore the univariate and multivariate techniques. The first ones have the merit of the ease of calculation and reading, but do not take into account the joint effect of the variables. The seconds are a priori more efficient, but require additional expertise to fully understand the results.

Keywords: cluster analysis, clustering, unsupervised learning, percentage of variance explained, V-Test, test value, distance between centroids, correlation ratio
Slides: Characterizing the clusters
References:
Tanagra Tutorial, "Understanding the 'test value' criterion", May 2009.
Tanagra Tutorial, "Hierarchical agglomerative clustering", June 2017.
Tanagra Tutorial, "K-Means clustering", June 2017.

Kohonen map with R

This tutorial complements the course material concerning the Kohonen map or Self-organizing map (June 2017). In a first time, we try to highlight two important aspects of the approach: its ability to summarize the available information in a two-dimensional space; Its combination with a cluster analysis method for associating the topological representation (and the reading that one can do) to the interpretation of the groups obtained from the clustering algorithm. We use the R software and the “Kohonen” package (Wehrens et Buydens, 2007). In a second time, we carry out a comparative study of the quality of the partitioning with the one obtained with the K-means algorithm. We use an external evaluation i.e. we compare the clustering results with pre-established classes. This procedure is often used in research to evaluate the performance of clustering methods. It takes on its meaning when it is applied to artificial data where the true class membership is known. We use the K-Means and Kohonen-Som components of Tanagra.

This tutorial is based on the Shane Lynn's article on the R-bloggers website (Lynn, 2014). I completed it by introducing the intermediate calculations to better understand the meaning of the charts, and by conducting the comparative study.

Keywords: som, self organizing map, kohonen network, data visualization, dimensionality reduction, cluster analysis, clustering, hierarchical agglomerative clustering, hac, two-step clustering, R software, kohonen package, k-means, external evaluation, heatmaps
Components: KOHONEN-SOM
Tutorial: Kohonen map with R
Program and dataset: waveform - som
References:
Tanagra tutorial, "Self-organizing map (slides)", June 2017.
Tanagra Tutorial, "Self-organizing map (with Tanagra)", July 2009.

Cluster analysis with Python - HAC and K-Means

This tutorial describes a cluster analysis process. We deal with a set of cheeses (29 instances) characterized by their nutritional properties (9 variables). The aim is to determine groups of homogeneous cheeses in view of their properties. We inspect and test two approaches using two Python procedures: the Hierarchical Agglomerative Clustering algorithm (SciPy package) ; and the K-Means algorithm (scikit-learn package).

One of the contributions of this tutorial is that we had conducted the same analysis with R previously, with the same steps. We can compare the commands used and the results provided by the available procedures. We observe that these tools have comparable behaviors and are substitutable in this context.

Keywords: python, scipy, scikit-learn, cluster analysis, clustering, hac, hierarchical agglomerative clustering, , k-means, principal component analysis, PCA
Turorial: hac and k-means with Python 
Dataset and cource code: hac_kmeans_with_python.zip
References :
Marie Chavent, Teaching Page, University of Bordeaux.

Tanagra Tutorials, "Cluster analysis with R - HAC and K-Means", July 2017.

Cluster analysis with R - HAC and K-Means

This tutorial describes a cluster analysis process. We deal with a set of cheeses (29 instances) characterized by their nutritional properties (9 variables). The aim is to determine groups of homogeneous cheeses in view of their properties.

We inspect and test two approaches using two procedures of the R software: the Hierarchical Agglomerative Clustering algorithm (hclust) ; and the K-Means algorithm (kmeans).

The data file "fromage.txt" comes from the teaching page of Marie Chavent from the University of Bordeaux. The excellent course materials and corrected exercises (commented R code) available on its website will complete this tutorial, which is intended firstly as a simple guide for the introduction of the R software in the context of the cluster analysis.

Keywords: R software, cluster analysis, clustering, hac, hierarchical agglomerative clustering, , k-means, fpc package, principal component analysis, PCA
Components: hclust, kmeans, kmeansruns
Turorial: hac and k-means with R 
Dataset and cource code: hac_kmeans_with_r.zip
References :
Marie Chavent, Teaching Page, University of Bordeaux.

k-medoids clustering (slides)

K-medoids is a partitioning-based clustering algorithm. It is related to the k-means but, instead of using the centroid as reference data point for the cluster, we use the medoid which is the individual nearest to all the other points within its cluster. One of the main consequence of this approach is that the resulting partition is less sensible to outliers.

This course material describes the algorithm. Then, we focus on the silhouette tool which can be used to determine the right number of clusters, a recurring open problem in cluster analysis.

Keywords: cluster analysis, clustering, unsupervised learning, paritionning method, relocation approach, medoid, PAM, partitioning aroung medoids, CLARA, clustering large applications, silhouette, silhouette plot
Slides: Cluster analysis - k-medoids algorithm
References:
Wikipedia, "k-medoids".