Janis Zuters

Near Real-Time Data Warehousing with Multi-stage Trickle and Flip

A data warehouse typically is a collection of historical data designed for decision support, so it is updated from the sources periodically, mostly on a daily basis. Today’s business however asks for fresher data. Real-time warehousing is one of the trends to accomplish this, but there are a number of challenges to move towards true real-time. This paper proposes ‘Multi-stage Trickle & flip’ methodology for data warehouse refreshment. It is based on the ‘Trickle & flip’ principle and extended in order to further insulate loading and querying activities, thus enabling both of them to be more efficient.

Sequence Q-learning: A memory-based method towards solving POMDP

Partially observable Markov decision process (POMDP) models a control problem, where states are only partially observable by an agent. The two main approaches to solve such tasks are these of value function and direct search in policy space. This paper introduces the Sequence Q-learning method which extends the well known Q-learning algorithm towards the ability to solve POMDPs through adding a special sequence management framework by advancing from action values to “sequence” values and including the “sequence continuity principle”.

Can SQ and EQ Values and Their Difference Indicate Programming Aptitude to Reduce Dropout Rate

A crucial problem that we are currently facing at the Faculty of Computing of the University of Latvia is that during the first study semester on average 30% of the first-year students drop out, whereas after the first year of studies the number of dropouts increases up to nearly 50%. Thus, our overall goal is to determine in advance applicants that most likely will not finish the first study year successfully. A hypothesis formulated in another research study was that programming aptitude could be predicted based on the results of two personality self-report questionnaires − Systemizing Quotient (SQ) and Empathy Quotient (EQ) − taken by students. The difference between the SQ and EQ scores had a strong correlation with grades received for programming test. We reproduced the circumstances of mentioned empirical study with our first-year students using similar tests to calculate SQ and EQ, and semester grades in introductory programming course as a quantitative measure to evaluate programming ability. In this paper, we elaborate on the empirical setting, measures, and estimation methods of our study, which produced the results that made us call the stated hypothesis into question and disprove it.

Modelling of Adequate Costs of Utilities Services

The paper propose methodology for benchmark modelling of adequate costs of utilities services, which is based on the data analysis of the factual cases (key performance indicators of utilities as the predictors). The proposed methodology was tested by modelling of Latvian water utilities with three tools: (1) a classical version of the multi-layer perceptron with error back-propagation training algorithm was sharpened up with task-specific monotony tests, (2) the fitting of the generalized additive model using the programming language R ensured the opportunity to evaluate the statistical significance and confidence bands of predictors, (3) the sequential iterative nonlinear regression process with minimizing mean squared error provided the notion of the impact of each predictor on the searched regularity. The quality of models is high: the adjusted determination coefficient is greater than 0.75, explained deviance exceeds 0.80, while the correlation between the respective modelled values exceeds even 0.95.

CN2-R: Faster CN2 with randomly generated complexes

Among the rule induction algorithms, the classic CN2 is still one of the most popular ones; a great amount of enhancements and improvements to it is to witness this. Despite the growing computing capacities since the algorithm was proposed, one of the main issues is resource demand. The proposed modification, CN2-R, substitutes the star concept of the original algorithm with a technique of randomly generated complexes in order to substantially improve on running times without significant loss in accuracy.

Realizing Undelayed N-step TD prediction with neural networks

There exist various techniques to extend reinforcement learning algorithms, e.g., eligibility traces and planning. In this paper, an approach is proposed, which combines several extension techniques, such as using eligibility-like traces, using approximators as value functions and exploiting the model of the environment. The obtained method, ‘Undelayed n-step TD prediction’ (TD-P), has produced competitive results when put in conditions of not fully observable environment.