No / Model interpretation is one of the key aspects of the model evaluation process. The explanation of the relationship between model variables and outputs is relatively easy for statistical models, such as linear regressions, thanks to the availability of model parameters and their statistical significance . For “black box” models, such as random forest, this information is hidden inside the model structure. This work presents an approach for computing feature contributions for random forest classification models. It allows for the determination of the influence of each variable on the model prediction for an individual instance. By analysing feature contributions for a training dataset, the most significant variables can be determined and their typical contribution towards predictions made for individual classes, i.e., class-specific feature contribution “patterns”, are discovered. These patterns represent a standard behaviour of the model and allow for an additional assessment of the model reliability for new data. Interpretation of feature contributions for two UCI benchmark datasets shows the potential of the proposed methodology. The robustness of results is demonstrated through an extensive analysis of feature contributions calculated for a large number of generated random forest models.

In this paper, we begin in Chapter 1 by giving a brief overview of the history of fractal geometry, focusing on six of the most important mathematicians in this field. Chapter 2 explains the main definitions needed for the remainder of the paper. In Chapter 3, we clarify the process of creating fractals from iterated function systems. Chapter 4 consists of an examination of the properties of traditional fractals. Next, in Chapter 5, we examine the newly discovered V-variable fractals and their properties. Finally we consider applications and future research in the field of fractal geometry.

This paper examines the response of Swedish local governments to a targeted intergovernmental stimulus grant aimed at increasing staffing levels in elderly care. The focus is on two key outcomes: municipal elderly care personnel costs relative to total municipal costs and the number of full-time employees in elderly care per elderly user. An OLS regression based on panel data between 2011 and 2018 initially estimates the grant’s spending effects. An instrumental variable (IV) model is then employed to address potential endogeneity, utilizing an update in the grant allocation formula. Both the OLS and IV estimates suggest that the stimulus grant has no discernible effect on the ratio of elderly care personnel costs to total municipal spending. Furthermore, the IV results show insignificant short-run effects on full-time employment in elderly care. However, significant increases are observed three years after the allocation formula update. The overall effects confirm standard economic grant theory predicting how non-matching targeted grants only contribute to an income effect.

Geomantic culture, deeply rooted in China's rich traditions, stands out as one of the most captivating and mysterious aspects of the country's heritage. In thousands of years, this enigmatic blend of metaphysics and geomancy has significantly shaped the lives of the Chinese people. Its historical success in predicting and judging risks has made it a valuable tool for investors in the field of real estate. In this case, research on the influence of risks of metaphysics and geomancy becomes necessarily important to facilitate investors to own relatively feasible tools when participating in economic activities. In this dissertation, I collect property transaction data in a district of city of Beijing in China in August 2023, and examine how various geomantic characteristics affect property prices. I identify sixteen indicators of geomantic factors, among which four are variables of community characteristics (location, shape, main entrance and road), four are building characteristics (position, orientation, floor and house number), and eight are internal characteristics of residences (house type, layout, daylighting, ventilation, color scheme, five elements, door and window). Quantification of these 16 characteristic variables can be divided as quantitative quantification and qualitative quantification, and the method to quantify indicators in accordance with the actual situation is an innovative content of indicator quantification of this paper. Employing multiple linear regressions, I find that 10 out of the 16 characteristic variables, including location, shape, main entrance, road, position, orientation, floor, house number, house type, and layout, are significantly related to property prices. I also compare the strength of the effect of each geomantic characteristic variable on property prices through standardized regression. Community location of the property has the greatest influence on prices. a building's position is at the second most important factor. House number and layout seem having the least effect on property transaction prices. My research suggests that homeowners in China care about the geomantic characteristics of their property, hence are willing to pay a higher price for more favorable geomantic characteristics. This study also offers useful guidance for real estate developers how to maximize profits through improving geomantic characteristics of their property development. / Global Finance

The current investigation is framed within variationist sociolinguistics, an area of study that examines how language varies and is influenced by linguistic and extralinguistic factors. Variationist research has reported that the use of verbal moods (the subjunctive and indicative) in Spanish is variable. One of the reasons for this variation appears to be a centuries-long process of language change, during which the subjunctive mood seemed to show a decline in favor of the indicative mood. Considering this gradual process of change, various investigations have examined how sociolinguistic factors condition variability in Spanish, focusing their analyses on different Spanish-speaking regions. To contribute to the understanding of mood variation in Spanish, this study explored a range of sociolinguistic factors across three Spanish-speaking regions (Rosario, Argentina; Barcelona, Spain; and Seville, Spain) by means of a conceptual replication of Gudmestad (2021), which studied variable mood use in Spanish across three regions (Quito, Ecuador; Mérida, Mexico; and Seville, Spain). The current study's participant pool (N = 107) consisted of Spanish speakers—men and women over 18 years old—residing in the three aforementioned metropolitan areas, who had lived in the same location for at least 15 years at the moment of the data collection. The data were collected via a written clause-elicitation task, with the purpose of having enough verbal-mood contexts to analyze, and a background questionnaire with basic demographic information. The results suggested that there was geographical variation of mood use, which reinforced the original study's findings on regional variation of mood use. Like Gudmestad (2021), the patterns of verbal moods with individual governors (e.g., preferir que 'to prefer that'), semantic category, and time reference diverged across the regions. The examination of gender and individual participants also pointed to a possible connection between these factors and variable mood use across regions. These findings indicate that the envelopes of variation of mood use may be diverse across the regions under investigation, which suggests that different geographical regions may have slightly distinct grammars. / Master of Arts / The current investigation is framed within variationist sociolinguistics, an area of study that examines how language varies, and which factors may condition this variation. Research within this field has found that Spanish speakers use subjunctive and indicative moods in a variable way. This variability in mood use may be connected to a long process of language change, over which speakers seem to be using the indicative mood in instances in which, in the past, they would have used the subjunctive mood. Taking this into account, various investigations have studied how the use of verbal moods in Spanish varies by analyzing different locations where Spanish is spoken. To contribute to this body of knowledge, the present investigation studied variable mood use in three regions (Rosario, Argentina; Barcelona, Spain; and Seville, Spain), by partially replicating an original investigation conducted by Gudmestad (2021) on variable mood use in Spanish across the regions of Quito, Ecuador; Mérida, Mexico; and Seville, Spain. Participants of the replication study (a total of 107) were three geographical groups of Spanish speakers. They were men and women over 18 years old who were residing in the three aforementioned urban areas and who had lived in the same location for at least 15 years. The participants completed a story with blanks that elicited the use of verbal moods and a questionnaire with basic demographic information. The findings indicated that Spanish speakers may use the subjunctive and indicative moods variably, according to different linguistic contexts and extralinguistic characteristics. Regarding geographical variation, the results showed that mood use was variable across the different locations analyzed, which reinforced the original study's findings on the presence of regional variation of mood use in Spanish.

High dimensional data have been frequently collected in many fields of scientific research and technological development. The traditional idea of best subset selection methods, which use penalized L_0 regularization, is computationally too expensive for many modern statistical applications. A large number of variable selection approaches via various forms of penalized least squares or likelihood have been developed to select significant variables and estimate their effects simultaneously in high dimensional statistical inference. However, in modern applications in areas such as genomics and proteomics, ultra-high dimensional data are often collected, where the dimension of data may grow exponentially with the sample size. In such problems, the regularization methods can become computationally unstable or even infeasible. To deal with the ultra-high dimensionality, Fan and Lv (2008) proposed a variable screening procedure via correlation learning to reduce dimensionality in sparse ultra-high dimensional models. Since then many authors further developed the procedure and applied to various statistical models. However, they all focused on single type of predictors, that is, the predictors are either all continuous or all discrete. In practice, we often collect mixed type of data, which contains both continuous and discrete predictors. For example, in genetic studies, we can collect information on both gene expression profiles and single nucleotide polymorphism (SNP) genotypes. Furthermore, outliers are often present in the observations due to experimental errors and other reasons. And the true trend underlying the data might not follow the parametric models assumed in many existing screening procedures. Hence a robust screening procedure against outliers and model misspecification is desired. In my dissertation, I shall propose a robust feature screening procedure for mixed type of data. To gain insights on screening for individual types of data, I first studied feature screening procedures for single type of data in Chapter 2 based on marginal quantities. For each type of data, new feature screening procedures are proposed and simulation studies are performed to compare their performances with existing procedures. The aim is to identify a best robust screening procedure for each type of data. In Chapter 3, I combine these best screening procedures to form the robust feature screening procedure for mixed type of data. Its performance will be assessed by simulation studies. I shall further illustrate the proposed procedure by the analysis of a real example. / Ph. D. / In modern applications in areas such as genomics and proteomics, ultra-high dimensional data are often collected, where the dimension of data may grow exponentially with the sample size. To deal with the ultra-high dimensionality, Fan and Lv (2008) proposed a variable screening procedure via correlation learning to reduce dimensionality in sparse ultra-high dimensional models. Since then many authors further developed the procedure and applied to various statistical models. However, they all focused on single type of predictors, that is, the predictors are either all continuous or all discrete. In practice, we often collect mixed type of data, which contains both continuous and discrete predictors. Furthermore, outliers are often present in the observations due to experimental errors and other reasons. Hence a robust screening procedure against outliers and model misspecification is desired. In my dissertation, I shall propose a robust feature screening procedure for mixed type of data. I first studied feature screening procedures for single type of data based on marginal quantities. For each type of data, new feature screening procedures are proposed and simulation studies are performed to compare their performances with existing procedures. The aim is to identify a best robust screening procedure for each type of data. Then i combined these best screening procedures to form the robust feature screening procedure for mixed type of data. Its performance will be assessed by simulation studies and the analysis of real examples.

La coupe à rétention de bouquets est un type de coupe à rétention variable visant à atténuer l'impact de la coupe totale sur la diversité biologique et sur la qualité visuelle des paysages. Puisque l'utilisation des types de coupe associés à l'aménagement écosystémique est encore récente au Québec, très peu d'études ont été réalisées afin d'en vérifier l'impact réel sur les écosystèmes forestiers. La présente étude vise à déterminer si les bouquets réussissent à préserver des attributs qui sont perdus lors de coupes traditionnelles en comparant la structure et les communautés de plantes de sous-bois retrouvées dans ces deux milieux. Nos résultats indiquent que la diversité végétale des deux milieux est semblable. Cependant, l'approche par traits fonctionnels montre que la tolérance à l'ombre semble jouer un rôle significatif dans la répartition des espèces entre les deux environnements. À court terme, la coupe à rétention de bouquet devrait également préserver une plus grande quantité de bois mort qu'une coupe avec protection de la régénération et des sols (CPRS), ce qui pourrait avoir des effets bénéfiques à long terme pour le maintien de la biodiversité.

Le travail présenté dans ce mémoire s’est déroulé dans le cadre d’une collaboration entre le groupe Convertisseur Statique du laboratoire LAPLACE et le département " Installations Fixes de Traction Electrique " de la Direction de l’Ingénierie de la SNCF. Aujourd'hui, la majorité du trafic sur le réseau ferré monophasé 25 kV / 50 Hz est assurée par des locomotives équipées de redresseurs contrôlés à thyristors. Du fait de l'augmentation du trafic, l'utilisation de ces locomotives nécessite la mise en place de moyen de compensation de puissance réactive afin de maintenir la tension caténaire à un niveau acceptable et de réduire la facture en énergie réactive. La correction du facteur de déplacement est réalisée en partie par des batteries de compensation fixes dont lapuissance est limitée par la tension maximale admissible à vide sur la caténaire. Afin d’adapter le niveau de compensation à la consommation, la partie fixe est complétée par un dispositif réglable basé sur une réactance contrôlée par des thyristors. Bien qu’il soit simple dans son principe, ce dispositif nécessite un filtrage des harmoniques en basse fréquence avec des circuits LC volumineux. L’objectif de cette thèse est de proposer une nouvelle topologie de compensateur de puissance réactive à haut rendement et utilisant un contrôle à modulation de largeur d’impulsion dans le but de minimiser le volume des éléments de filtrage. La première partie de ce mémoire est consacrée à une étude comparative de différentes topologies du point de vue des pertes dans les semi-conducteurs et du dimensionnement des éléments de filtrage associés. Les résultats de cette étude montrent que les topologies à base de gradateurs MLI, constituent les solutions les plus intéressantes pour réaliser le compensateur. La deuxième partie du travail concerne l’étude de l’insertion d’un compensateur à base de gradateurs MLI sur le réseau ferré avec la prise en compte des interactions harmoniques. Le cas d’étude concerne une sous-station où doit être implantée une compensation variable de 3 MVAR. Un relevé des courants délivrés par la sous-station a été effectué dans le but d’analyser leur contenu harmonique. Une modélisation de la sous-station et du compensateur est ensuite proposée et des simulations temporelles de l’ensemble sont réalisées en prenant en compte les formes d’ondes réelles des courants absorbés par les trains. Finalement, cette étude par simulation permet d’affiner le dimensionnement du compensateur à gradateurs MLI et des éléments de filtrage associés. Elle met en évidence l’avantage d’une solution, avec des gradateurs MLI en montage élévateur de tension, qui fonctionne sans transformateur et réutilise les batteries de compensation fixes déjà installées en sous-station à la fois comme diviseur de tension et éléments de filtrage. Afin de valider le principe de cette nouvelle topologie de compensateur statique de puissance réactive, un démonstrateur de 1,2 MVAR est mis en oeuvre et testé sur une plateforme d'essai de la SNCF. / The work presented in this thesis is the result of collaboration between the Static Converters research group of LAPLACE Laboratory and the department "Installations Fixes de Traction Electrique" of the Engineering Division of the French Railways company, SNCF. Nowadays, most of the traffic in 25 kV – 50 Hz lines is achieved by old locomotives equipped with thyristor rectifiers. As traffic and load increase, reactive power compensation devices are required to keep the overhead line voltage at acceptable level and to reduce the spending for reactive power. The basic power factor correction is completed by fixed compensation banks. But the difficulty of such configuration is the no-load operation of overhead lines. The voltage can increase out of the 29 kV standard limit and to avoid this problem, variable reactive power compensator is often added to the fixed compensation banks. Nowadays, SNCF is equipped with thyristor based static VAR compensators (SVC). The main drawback of this topology is the requirement of a large LC shunt filter tuned for the third harmonic. The goal of this thesis is to find a new high-efficiency topology of STATic COMpensator (STATCOM) using PWM control to minimize the filtering components. The first part of this thesis focuses on a comparative study of different topologies in terms of semiconductor losses and filter elements size. The results of this study show that the topologies based on PWM AC Chopper are the most interesting solutions to achieve the compensator. The second part of the work concerns the influence of the connection of a STATCOM to a substation and focuses on the harmonic interactions with locomotives. The study case concerns a high traffic substation where a 3 MVAR STATCOM should be installed. Substation output current measurement was carried out in order to analyze its harmonic content. Models of the substation and the STATCOM are then proposed and simulations using real current waveforms are performed. Finally, this simulation study helps to refine the features of the AC Chopper topology and its filter elements. It highlights the advantage of a solution, with boost AC Choppers which operate without transformer and reuse the fixed compensation banks already installed in the substation as voltage divider and filter. To validate the principle of this new topology of STATCOM, a 1.2 MVAR prototype is built in LAPLACE laboratory and tested on a SNCF test platform.

La caséine [bêta] (CN-[bêta]) et la caséine [alpha]s1 (CN-[alpha]s1) du lait de jument possèdent un taux variable de phosphorylation et sont de bons modèles d’étude de l’influence du degré de phosphorylation et de la séquence peptidique sur la chélation de caséinophosphopeptides (CPP) avec des minéraux d’intérêt nutritionnel. Avant d’envisager une telle étude, la structure des caséines doit être déterminée précisément. Notre travail a été consacré à la caractérisation de variants post-transcriptionnels et post-traductionnels de CN-[alpha]s1 et de CN-[bêta]. Après fractionnement chromatographique et analyse par spectrométrie de masse, la CN-[alpha]s1 entière, trois variants d’épissage alternatif des exons 7 et 14 de la CN-[alpha]s1 et quatre variants délétés du résidu Gln91, résultat de l’utilisation d’un site d’épissage cryptique, ont été identifiés dans le lait équin. Nous avons montré que le degré de phosphorylation de ces isoformes varie de 2 à 8 groupes phosphates. Au total, 36 isoformes différentes de CN-[alpha]s1 ont été caractérisées. La cartographie bidimensionnelle de la CN-[bêta] a été établie avec précision après avoir isolé par chromatographie chacun des variants de phosphorylation (possédant de 3 à 7 groupes phosphates) et après avoir caractérisé les formes de CN-[bêta] modifiées par désamidation non enzymatique du résidu Asn135. Des CPP trypsiques de chaque variant de phosphorylation ont été préparés avec une technique récente de chromatographie d’affinité au dioxyde de titane, ce qui a permis de localiser par spectrométrie de masse en tandem les sites phosphorylés de la CN-[bêta] (Ser9, Ser10, Thr12, Ser18, Ser23, Ser24, Ser25) et de montrer que la phosphorylation de la CN-[bêta] n’est pas aléatoire mais séquentielle / Equine [bêta]-casein ([bêta]-CN) and [alpha]s1-casein ([alpha]s1-CN) have a variable phosphorylation degree and are good models for the study of the influence of phosphorylation degree and peptide sequence on chelation of caseinophosphopeptides (CPP) with minerals of nutritional interest. Before considering such study, structure of caseins must be precisely determined. Our work has been devoted to the characterization of post-transcriptional and post-translational variants of [alpha]s1-CN and [bêta]-CN. Concerning [alpha]s1-CN, the full-length protein, three alternative splicing variants involving exons 7 and 14 and four variants involving cryptic splice site resulting in deletion of residue Gln91 have been identified in mare’s milk after chromatographic fractionation and mass spectrometric analysis. The phosphorylation degree of these variants varies between 2 and 8 phosphate groups. Finally, 36 isoforms of [alpha]s1-CN have been identified. Isolation of each phosphorylation variant (having 3 to 7 phosphate groups) of [bêta]-CN by chromatography, and characterization of modified forms of [bêta]-CN by non enzymatic deamidation of residue Asn135 permits the establishment of bidimensional cartography of [bêta]-CN with precision. After hydrolysis by trypsin, CPP of each phosphorylation variant have been prepared by affinity chromatography to titanium dioxide, a recent technology, which allowed to locate by mass tandem spectrometry the phosphorylated sites of [bêta]-CN (Ser9, Ser10, Thr12, Ser18, Ser23, Ser24, Ser25). It was shown that the phosphorylation of [bêta]-CN is not a random process but follows a sequential way

La dynamique instable des jets tournants est étudiée, en tenant compte des variations de masse volumique au sein de l'écoulement. Un code de simulation numérique directe permettant de résoudre les équations de Navier-Stokes à masse volumique variable a été développé, en utilisant une méthode originale et efficace pour résoudre le champs de pression. Analytiquement, deux modes instables bidimensionnels ont été mis en évidence, et sont identifiés comme des modes de Couette-Taylor et de Rayleigh-Taylor, ainsi qu'un troisième mode tridimensionnel, du à un couplage de vitesse. La dynamique instable de cet écoulement résulte d'une compétition entre ces trois modes, et les simulations numériques montrent que ces modes perdurent non linéairement. Ensuite, le comportement spatio-temporel de cette instabilité est étudiée par simulation numérique directe, et il a été montré qu'il existe une transition vers des modes absolument instables, sous l'effet du rapport de densité s ainsi que du taux de rotation q. Cette dynamique est également étudiée expérimentalement au travers de plusieurs méthodes de mesures, et la présence de mode globaux auto-entretenus est mise en évidence qui sont en bon accord avec les résultats numériques. Finalement, le phénomène de l'éclatement tourbillonnaire est étudié, et montre le rôle prépondérant de la viscosité réelle. En effet, l'éclatement tourbillonnaire est un mécanisme permettant de soulager le système de l'intensification de la vorticité, au travers de la viscosité, alors qu'il n'apparaît pas en traitant les équations d'Euler tronquées. / The unstable dynamics of a swirling jet flow is studied, including density variations within the flow. A direct numerical simulation method was developed to solve variable density Navier-Stokes equations, using an accurate and efficient pressure solver. Analitically, two unstable bi-dimensionnal modes are highlighted, and are identified as Couette-Taylor and Rayleigh-Taylor modes. A three-dimensionnal mode is also highlighted, wich is created by the shear. Numerical simulations show that those modes are nonlinearly persistant. Then, the spatio-temporal instability behaviour is studied numerically, and show that the instability undergoes to a convective/absolute transition with density ratio s and rotation rate q. This dynamic is also studied experiementally through different methods, and Global selfsustained modes are highlighted wich are in ggod agreement with numerical results. Finally, the vortex breakdown phenomenon is studied, and show the crucial role of real viscosity. Indeed, the vorticity intensification is relaxed through the viscosity effect, while it is not treating the truncated Euler Equations.