Browsing by Subject "clustering"

Active galactic nuclei (AGN) are compact, luminous objects found in the central region of many galaxies. In the standard paradigm, the AGN is fueled by accretion of matter into a supermassive black hole (SMBH). In fact, the properties of many galaxies and their respective SMBHs are linked, which hints at the importance of AGN as factors in galaxy formation and evolution. The bulk of the matter in the Universe is some form of dark matter, which is still poorly understood. AGN are biased tracers of the underlying dark matter distribution. By comparing the clustering of AGN with that of the dark matter, the bias may be quantified and further, the bias can be linked to a characteristic mass of the dark matter halo hosting the AGN. The advent of high-resolution X-ray telescopes, namely Chandra and XMM-Newton, has made unprecedently large samples available for study. With detailed spectroscopic follow-up programs, the study of X-ray selected AGN clustering has received a major boost. The clustering measurements tell of the typical environments that are likely to host AGN and thus shed light on what actually triggers the AGN. In this thesis, the clustering of ∼ 600 X-ray selected AGN with z < 2.5 (z = 1.19) in the COS- MOS (Cosmic evolution survey) field surveyed with XMM-Newton (XMM-COSMOS) is studied. The full sample is split into subsamples based on the host galaxy stellar mass M∗ and the ratio between the X-ray luminosity and the stellar mass LX/M∗ which is a proxy for the Eddington ratio. For the full sample the bias is 3.61+0.37−0.40, which corresponds to a characteristic halo mass of log M halo /h−1 M⊙ = 13.52+0.12−0.16 , consistent with the overall picture of X-ray selected AGN residing in massive haloes with 12.5 < logMhalo/h−1M⊙ < 13.5. The low M∗ and high M∗ samples have biases 3.53+0.58−0.70 and 4.13+0.85−1.07, respectively and the data do not support a difference in the typical masses of the hosting haloes. For the LX/M∗ subsamples, there is marginal evidence that low L X /M∗ AGN (logM halo /h−1M⊙ = 13.52+0.22−0.37) reside in more massive haloes than high L X /M∗ AGN (logM halo /h−1M⊙ = 13.29+0.28−0.58). One possible explanation would be that the environment of the low LX /M∗ AGN reduces the amount of gas available for accretion and thus results in lower accretion rates.

Programming courses often receive large quantities of program code submissions to exercises which, due to their large number, are graded and students provided feedback automatically. Teachers might never review these submissions therefore losing a valuable source of insight into student programming patterns. This thesis researches how these submissions could be reviewed efficiently using a software system, and a prototype, CodeClusters, was developed as an additional contribution of this thesis. CodeClusters' design goals are to allow the exploration of the submissions and specifically finding higher-level patterns that could be used to provide feedback to students. Its main features are full-text search and n-grams similarity detection model that can be used to cluster the submissions. Design science research is applied to evaluate CodeClusters' design and to guide the next iteration of the artifact and qualitative analysis, namely thematic synthesis, to evaluate the problem context as well as the ideas of using software for reviewing and providing clustered feedback. The used study method was interviews conducted with teachers who had experience teaching programming courses. Teachers were intrigued by the ability to review submitted student code and to provide more tailored feedback to students. The system, while still a prototype, is considered worthwhile to experiment on programming courses. A tool for analyzing and exploring submissions seems important to enable teachers to better understand how students have solved the exercises. Providing additional feedback can be beneficial to students, yet the feedback should be valuable and the students incentivized to read it.

The subject of the thesis is the comparison of lexical semantics and phonetics. The thesis investigates with computational methods if there is significantly more phonetic variance in words that belong to the same semantic domains than with phonetically similar words from other semantic domains. In other words, phonetically very similar words and especially phonological minimal pairs would be in separate semantic domains. The method clusters word embedding vectors and distinctive phonological feature vectors from multiple languages, and the phonetic and semantic standard deviations are calculated for each cluster, and the mean standard deviations of cluster sets are compared. In addition to semantic and phonetic clusters, two test clusters are constructed which have the same number and the same size of clusters as the semantic clusters. The first test clusters use the words from phonetic clusters in order and the second test clusters are randomly permuted. These different cluster sets are compared by their mean standard deviations and cluster set similarity index. The results imply that words on the same semantic domains contain rarely phonetically very similar words, and those words are usually in separate semantic domains.

The self-organizing map (SOM) is a form of unsupervised neural network and a method for data analysis that allows reducing the dimensionality of data, exploring the variation and dependencies between variables and presenting their similarity relations. Being a powerful visualization instrument and having a strong disposition for clustering, the self-organizing map could be implemented to the analysis of survey data, particularly, collected with the questionnaires. This thesis provides a relevant example of dealing with the limited size mixed survey data set. The self-organizing map algorithm is implemented to analyze the data obtained from the faculty well-being project organized at the Faculty of Social Sciences in the University of Helsinki. The set of experiments utilize the self-organizing map algorithm to explore a possible clustering structure of the data and identify the profiles of the survey participants. Each of three experiments illustrates different variable encoding approaches for the sets of closed background and Likert scale questions. The largest number of the profiles was obtained from the final experiment. Four out of seven profiles represent clusters of the individuals with mainly neutral, negative or very negative experiences related to the well-being at the faculty. The data analysis experiments also illustrate the possible challenges of the SOM method implementation to survey data. The existence of categorical variables, the necessity of choosing a set of parameters for the SOM training and dealing with the missing values are discussed as main challenges of the SOM implementation to survey data analysis using the R package “kohonen”.

This work presents a novel concept of categorising failures within test logs using string similarity algorithms. The concept was implemented in the form of a tool that went through three major iterations to its final version. These iterations are the following: 1) utilising two state-of-the-art log parsing algorithms, 2) manual log parsing of the Pytest testing framework, and 3) parsing of .xml files produced by the Pytest testing framework. The unstructured test logs were automatically converted into a structured format using the three approaches. Then, structured data was compared using five different string similarity algorithms, Sequence Matcher, Jaccard index, Jaro-Winkler distance, cosine similarity and Levenshtein ratio, to form the clusters. The results from each approach were implemented and validated across three different data sets. The concept was validated by implementing an open-sourced Test Failure Analysis (TFA) tool. The validation phase revealed the best implementation approach (approach 3) and the best string similarity algorithm for this task (cosine similarity). Lastly, the tool was deployed into an open-source project’s CI pipeline. Results of this integration, application and usage are reported. The achieved tool significantly reduces software engineers’ manual work and error-prone work by utilising cosine similarity as a similarity score to form clusters of failures.