Uwe Ligges

klaR Analyzing German Business Cycles

Decision making often asks for classification. We will present a new R package klaR including functions to build, check, tune, visualize, and compare classification rules. The software is illustrated by means of a case study of prediction of the German economy’s business cycle phases.

Classification of high and low achievers in a music sight-reading task

The unrehearsed performance of music, called ‘sight-reading’ (SR), is a basic skill for all musicians. It is of particular interest for musical occupations such as the piano accompanist, the conductor, or the correpetiteur. However, up until now, there is no theory of SR which considers all relevant factors such as practice-related variables (e.g. expertise), speed of information processing (e.g. mental speed), or psychomotor speed (e.g. speed of trills). Despite the merits of expertise theory, there is no comprehensive model that can classify subjects into high- and low-performance groups. In contrast to previous studies, this study uses a data mining approach instead of regression analysis and tries to classify subjects into predetermined achievement classes. It is based on an extensive experiment in which the total SR performance of 52 piano students at a German music department was measured by use of an accompanying task. Additionally, subjects completed a set of psychological tests, such as tests of mental speed, reaction time, working memory, inner hearing, etc., which were found in earlier studies to be useful predictors of SR achievement. For the first time, classification methods (cluster analysis, regression trees, classification trees, linear discriminant analysis) were applied to determine combinations of variables for classification. Results of a linear discriminant analysis revealed a two-class solution with four predictors (cross-validated error: 15%) and a three-class solution with five predictors (cross-validated error: 33%).

Classification in music research

Since a few years, classification in music research is a very broad and quickly growing field. Most important for adequate classification is the knowledge of adequate observable or deduced features on the basis of which meaningful groups or classes can be distinguished. Unsupervised classification additionally needs an adequate similarity or distance measure grouping is to be based upon. Evaluation of supervised learning is typically based on the error rates of the classification rules. In this paper we first discuss typical problems and possible influential features derived from signal analysis, mental mechanisms or concepts, and compositional structure. Then, we present typical solutions of such tasks related to music research, namely for organization of music collections, transcription of music signals, cognitive psychology of music, and compositional structure analysis.

A Convergence Criterion for Multiobjective Evolutionary Algorithms Based on Systematic Statistical Testing

A systematic approach for determining the generation number at which a specific Multi-Objective Evolutionary Algorithm (MOEA) has converged for a given optimization problem is introduced. Convergence is measured by the performance indicators Generational Distance, Spread and Hypervolume. The stochastic nature of the MOEA is taken into account by repeated runs per generation number which results in a highly robust procedure. For each generation number the MOEA is repeated a fixed number of times, and the Kolmogorow-Smirnov-Test is used in order to decide if a significant change in performance is gained in comparison to preceding generations. A comparison of different MOEAs on a problem with respect to necessary generation numbers becomes possible, and the understanding of the algorithms behaviour is supported by analysing the development of the indicator values. The procedure is illustrated by means of standard test problems.

Detection of Locally Stationary Segments in Time Series

In many applications it is required to segment a time series into its locally stationary parts. Two applications are presented:

Parameter Optimization in Automatic Transcription of Music

Based on former work on automatic transcription of musical time series into sheet music (Ligges et al. (2002), Weihs and Ligges (2003, 2005)) in this paper parameters of the transcription algorithm are optimized for various real singers. Moreover, the parameters of various artificial singer models derived from the models of Rossignol et al. (1999) and Davy and Godsill (2002) are estimated. In both cases, optimization is carried out by the Nelder-Mead (1965) search algorithm. In the modelling case a hierarchical Bayes extension is estimated by WinBUGS (Spiegelhalter et al. (2004)) as well. In all cases, optimal parameters are compared to heuristic estimates from our former standard method.

Antepartal insulin-like growth factor 1 and insulin-like growth factor binding protein 2 concentrations are indicative of ketosis in dairy cows.

A study involving a small number of cows found that the concentrations of insulin-like growth hormone 1 (IGF1) may be a useful predictor of metabolic disease. Further, IGF1 may provide also a pathophysiological link to metabolic diseases such as ketosis. The objective of the current study was to test whether the low antepartal total IGF1 or IGF1 binding protein (IGFBP) concentrations might predict ketosis under field conditions. Clinical examinations and blood sampling were performed antepartum (262-270 d after artificial insemination) on 377 pluriparous pregnant Holstein Friesian cows. The presence of postpartum diseases were recorded (ketosis, fatty liver, displacement of the abomasum, hypocalcemia, mastitis, retention of fetal membranes, and clinical metritis or endometritis), and the concentrations of IGF1, IGFBP2, IGFBP3, and nonesterified fatty acids were measured. Cows with postpartum clinical ketosis had lower IGF1 concentrations antepartum than healthy cows. The sensitivity of antepartal IGF1 as a marker for postpartum ketosis was 0.87, and the specificity was 0.43; a positive predictive value of 0.91 and a negative predictive value of 0.35 were calculated. The cows with ketosis and retained fetal membranes had lower IGFBP2 concentrations compared with the healthy cows. It can be speculated that lower IGF1 production in the liver during late pregnancy may increase growth hormone secretions and lipolysis, thereby increasing the risk of ketosis. Lower IGFBP2 concentrations may reflect the suppression of IGFBP2 levels through higher growth hormone secretion. In conclusion, compared with nonesterified fatty acids as a predictive parameter, IGF1 and IGFBP2 may represent earlier biomarkers of inadequate metabolic adaptation to the high energy demand required postpartum.

Statistik mit R

Bei R handelt es sich um eine ”Sprache und Umgebung fur Statistisches Rechnen“ (R Development Core Team, 2008a). Ein Kapitel uber die Anwendung statistischer Verfahren mit R darf und soll in diesem Buch nicht fehlen. Der Fokus des Buches ist jedoch auf die Programmierung mit R gerichtet, und es soll in erster Linie das Verstandnis der Sprache vermittelt werden. Daher werden nur haufig verwendete und wichtige Verfahren kurz erklart sowie syntaktische Feinheiten der Modellspezifikation beschrieben.

SVM Based Instrument and Timbre Classification

In this paper we propose a method that allows for instrument and timbre classification from a single tone. Features are derived from a pre-filtered time series divided into small windows. Afterwards, features from the (transformed) spectrum, Perceptive Linear Prediction (PLP), and Mel Frequency Cepstral Coefficients (MFCCs) as known from speech processing are selected. Clustering methods (e.g. k-means) are applied yielding a reduced number of aggregated features for the final classification task. It turns out that a polynomial kernel with reasonable complexity can be used for the SVM. Accuracy of the results is very convincing given a misclassification error of roughly 19% for 59 different classes of instruments. Misclassification error is much smaller for a reasonable small number of classes, of course. During methodological work, we ported the ‘rastamat’ library (Ellis 2005) functionality from Matlab to R. This means feature extraction as known from speech processing is now easily available from the statistical programming language R.

From local to global analysis of music time series

Local and more and more global musical structure is analyzed from audio time series by time-series-event analysis with the aim of automatic sheet music production and comparison of singers. Note events are determined and classified based on local spectra, and rules of bar events are identified based on accentuation events related to local energy. In order to compare the performances of different singers global summary measures are defined characterizing the overall performance.