Overfitting Bayesian mixture models with an unknown number of components

This paper proposes solutions to three issues pertaining to the estimation of finite mixture models with an unknown number of components: the non-identifiability induced by overfitting the number of components, the mixing limitations of standard Markov Chain Monte Carlo (MCMC) sampling techniques, and the related label switching problem. An overfitting approach is used to estimate the number of components in a finite mixture model via a Zmix algorithm. Zmix provides a bridge between multidimensional samplers and test based estimation methods, whereby priors are chosen to encourage extra groups to have weights approaching zero. MCMC sampling is made possible by the implementation of prior parallel tempering, an extension of parallel tempering. Zmix can accurately estimate the number of components, posterior parameter estimates and allocation probabilities given a sufficiently large sample size. The results will reflect uncertainty in the final model and will report the range of possible candidate models and their respective estimated probabilities from a single run. Label switching is resolved with a computationally light-weight method, Zswitch, developed for overfitted mixtures by exploiting the intuitiveness of allocation-based relabelling algorithms and the precision of label-invariant loss functions. Four simulation studies are included to illustrate Zmix and Zswitch, as well as three case studies from the literature. All methods are available as part of the R package Zmix, which can currently be applied to univariate Gaussian mixture models.

Combining propensity and influence models for product adoption prediction

This paper studies the problem of selecting users in an online social network for targeted advertising so as to maximize the adoption of a given product. In previous work, two families of models have been considered to address this problem: direct targeting and network-based targeting. The former approach targets users with the highest propensity to adopt the product, while the latter approach targets users with the highest influence potential – that is users whose adoption is most likely to be followed by subsequent adoptions by peers. This paper proposes a hybrid approach that combines a notion of propensity and a notion of influence into a single utility function. We show that targeting a fixed number of high-utility users results in more adoptions than targeting either highly influential users or users with high propensity.

Exact solutions of coupled multispecies linear reaction–diffusion equations on a uniformly growing domain

Embryonic development involves diffusion and proliferation of cells, as well as diffusion and reaction of molecules, within growing tissues. Mathematical models of these processes often involve reaction–diffusion equations on growing domains that have been primarily studied using approximate numerical solutions. Recently, we have shown how to obtain an exact solution to a single, uncoupled, linear reaction–diffusion equation on a growing domain, 0 < x < L(t), where L(t) is the domain length. The present work is an extension of our previous study, and we illustrate how to solve a system of coupled reaction–diffusion equations on a growing domain. This system of equations can be used to study the spatial and temporal distributions of different generations of cells within a population that diffuses and proliferates within a growing tissue. The exact solution is obtained by applying an uncoupling transformation, and the uncoupled equations are solved separately before applying the inverse uncoupling transformation to give the coupled solution. We present several example calculations to illustrate different types of behaviour. The first example calculation corresponds to a situation where the initially–confined population diffuses sufficiently slowly that it is unable to reach the moving boundary at x = L(t). In contrast, the second example calculation corresponds to a situation where the initially–confined population is able to overcome the domain growth and reach the moving boundary at x = L(t). In its basic format, the uncoupling transformation at first appears to be restricted to deal only with the case where each generation of cells has a distinct proliferation rate. However, we also demonstrate how the uncoupling transformation can be used when each generation has the same proliferation rate by evaluating the exact solutions as an appropriate limit.

The Feeding Practices and Structure Questionnaire (FPSQ-28): A parsimonious version validated for longitudinal use from 2-5 years

Prospective studies and intervention evaluations that examine change over time assume that measurement tools measure the same construct at each occasion. In the area of parent-child feeding practices, longitudinal measurement properties of the questionnaires used are rarely verified. To ascertain that measured change in feeding practices reflects true change rather than change in the assessment, structure, or conceptualisation of the constructs over time, this study examined longitudinal measurement invariance of the Feeding Practices and Structure Questionnaire (FPSQ) subscales (9 constructs; 40 items) across 3 time points. Mothers participating in the NOURISH trial reported their feeding practices when children were aged 2, 3.7, and 5 years (N = 404). Confirmatory Factor Analysis (CFA) within a structural equation modelling framework was used. Comparisons of initial cross-sectional models followed by longitudinal modelling of subscales, resulted in the removal of 12 items, including two redundant or poorly performing subscales. The resulting 28-item FPSQ-28 comprised 7 multi-item subscales: Reward for Behaviour, Reward for Eating, Persuasive Feeding, Overt Restriction, Covert Restriction, Structured Meal Setting and Structured Meal Timing. All subscales showed good fit over 3 time points and each displayed at least partial scalar (thresholds equal) longitudinal measurement invariance. We recommend the use of a separate single item indicator to assess the family meal setting. This is the first study to examine longitudinal measurement invariance in a feeding practices questionnaire. Invariance was established, indicating that the subscales of the shortened FPSQ-28 can be used with mothers to validly assess change in 7 feeding constructs in samples of children aged 2-5 years of age.

Computation of confined swirling flow using RNG and k-epsilon turbulence models

In this work we numerically model isothermal turbulent swirling flow in a cylindrical burner. Three versions of the RNG k-epsilon model are assessed against performance of the standard k-epsilon model. Sensitivity of numerical predictions to grid refinement, differing convective differencing schemes and choice of (unknown) inlet dissipation rate, were closely scrutinised to ensure accuracy. Particular attention is paid to modelling the inlet conditions to within the range of uncertainty of the experimental data, as model predictions proved to be significantly sensitive to relatively small changes in upstream flow conditions. We also examine the characteristics of the swirl--induced recirculation zone predicted by the models over an extended range of inlet conditions. Our main findings are: - (i) the standard k-epsilon model performed best compared with experiment; - (ii) no one inlet specification can simultaneously optimize the performance of the models considered; - (iii) the RNG models predict both single-cell and double-cell IRZ characteristics, the latter both with and without additional internal stagnation points. The first finding indicates that the examined RNG modifications to the standard k-e model do not result in an improved eddy viscosity based model for the prediction of swirl flows. The second finding suggests that tuning established models for optimal performance in swirl flows a priori is not straightforward. The third finding indicates that the RNG based models exhibit a greater variety of structural behaviour, despite being of the same level of complexity as the standard k-e model. The plausibility of the predicted IRZ features are discussed in terms of known vortex breakdown phenomena.

Genetic analysis of structural brain connectivity using DICCCOL models of diffusion MRI in 522 twins

Genetic and environmental factors affect white matter connectivity in the normal brain, and they also influence diseases in which brain connectivity is altered. Little is known about genetic influences on brain connectivity, despite wide variations in the brain's neural pathways. Here we applied the 'DICCCOL' framework to analyze structural connectivity, in 261 twin pairs (522 participants, mean age: 21.8 y ± 2.7SD). We encoded connectivity patterns by projecting the white matter (WM) bundles of all 'DICCCOLs' as a tracemap (TM). Next we fitted an A/C/E structural equation model to estimate additive genetic (A), common environmental (C), and unique environmental/error (E) components of the observed variations in brain connectivity. We found 44 'heritable DICCCOLs' whose connectivity was genetically influenced (α2>1%); half of them showed significant heritability (α2>20%). Our analysis of genetic influences on WM structural connectivity suggests high heritability for some WM projection patterns, yielding new targets for genome-wide association studies.

The Pandolf load carriage equation is a poor predictor of metabolic rate while wearing explosive ordnance disposal protective clothing

This investigation aimed to quantify metabolic rate when wearing an explosive ordnance disposal (EOD) ensemble (~33kg) during standing and locomotion; and determine whether the Pandolf load carriage equation accurately predicts metabolic rate when wearing an EOD ensemble during standing and locomotion. Ten males completed 8 trials with metabolic rate measured through indirect calorimetry. Walking in EOD at 2.5, 4.0 and 5.5km·h−1 was significantly (p < 0.05) greater than matched trials without the EOD ensemble by 49% (127W), 65% (213W) and 78% (345W), respectively. Mean bias (95% limits of agreement) between predicted and measured metabolism during standing, 2.5, 4 and 5.5km·h−1 were 47W (19 to 75W); −111W (−172 to −49W); −122W (−189 to −54W) and −158W (−245 to −72W), respectively. The Pandolf equation significantly underestimated measured metabolic rate during locomotion. These findings have practical implications for EOD technicians during training and operation and should be considered when developing maximum workload duration models and work-rest schedules.

Forecasting day-ahead electricity load using a multiple equation time series approach

The quality of short-term electricity load forecasting is crucial to the operation and trading activities of market participants in an electricity market. In this paper, it is shown that a multiple equation time-series model, which is estimated by repeated application of ordinary least squares, has the potential to match or even outperform more complex nonlinear and nonparametric forecasting models. The key ingredient of the success of this simple model is the effective use of lagged information by allowing for interaction between seasonal patterns and intra-day dependencies. Although the model is built using data for the Queensland region of Australia, the method is completely generic and applicable to any load forecasting problem. The model’s forecasting ability is assessed by means of the mean absolute percentage error (MAPE). For day-ahead forecast, the MAPE returned by the model over a period of 11 years is an impressive 1.36%. The forecast accuracy of the model is compared with a number of benchmarks including three popular alternatives and one industrial standard reported by the Australia Energy Market Operator (AEMO). The performance of the model developed in this paper is superior to all benchmarks and outperforms the AEMO forecasts by about a third in terms of the MAPE criterion.

Is the mental wellbeing of young Australians best represented by a single, multidimensional or bifactor model?

Internationally there is a growing interest in the mental wellbeing of young people. However, it is unclear whether mental wellbeing is best conceptualized as a general wellbeing factor or a multidimensional construct. This paper investigated whether mental wellbeing, measured by the Mental Health Continuum-Short Form (MHC-SF), is best represented by: (1) a single-factor general model; (2) a three-factor multidimensional model or (3) a combination of both (bifactor model). 2,220 young Australians aged between 16 and 25 years completed an online survey including the MHC-SF and a range of other wellbeing and mental ill-health measures. Exploratory factor analysis supported a bifactor solution, comprised of a general wellbeing factor, and specific group factors of psychological, social and emotional wellbeing. Confirmatory factor analysis indicated that the bifactor model had a better fit than competing single and three-factor models. The MHC-SF total score was more strongly associated with other wellbeing and mental ill-health measures than the social, emotional or psychological subscale scores. Findings indicate that the mental wellbeing of young people is best conceptualized as an overarching latent construct (general wellbeing) to which emotional, social and psychological domains contribute. The MHC-SF total score is a valid and reliable measure of this general wellbeing factor.

Eddy diffusivity: A single dispersion analysis of high resolution drifters in a tidal shallow estuary

In an estuary, mixing and dispersion resulting from turbulence and small scale fluctuation has strong spatio-temporal variability which cannot be resolved in conventional hydrodynamic models while some models employs parameterizations large water bodies. This paper presents small scale diffusivity estimates from high resolution drifters sampled at 10 Hz for periods of about 4 hours to resolve turbulence and shear diffusivity within a tidal shallow estuary (depth < 3 m). Taylor's diffusion theorem forms the basis of a first order estimate for the diffusivity scale. Diffusivity varied between 0.001 – 0.02 m2/s during the flood tide experiment. The diffusivity showed strong dependence (R2 > 0.9) on the horizontal mean velocity within the channel. Enhanced diffusivity caused by shear dispersion resulting from the interaction of large scale flow with the boundary geometries was observed. Turbulence within the shallow channel showed some similarities with the boundary layer flow which include consistency with slope of 5/3 predicted by Kolmogorov's similarity hypothesis within the inertial subrange. The diffusivities scale locally by 4/3 power law following Okubo's scaling and the length scale scales as 3/2 power law of the time scale. The diffusivity scaling herein suggests that the modelling of small scale mixing within tidal shallow estuaries can be approached from classical turbulence scaling upon identifying pertinent parameters.

Ductile deformation of single inclusions in simple shear with a finite-strain hyperelastoviscoplastic rheology

We examine the 2D plane-­strain deformation of initially round, matrix-­bonded, deformable single inclusions in isothermal simple shear using a recently introduced hyperelastoviscoplastic rheology. The broad parameter space spanned by the wide range of effective viscosities, yield stresses, relaxation times, and strain rates encountered in the ductile lithosphere is explored systematically for weak and strong inclusions, the effective viscosity of which varies with respect to the matrix. Most inclusion studies to date focused on elastic or purely viscous rheologies. Comparing our results with linear-­viscous inclusions in a linear-­viscous matrix, we observe significantly different shape evolution of weak and strong inclusions over most of the relevant parameter space. The evolution of inclusion inclination relative to the shear plane is more strongly affected by elastic and plastic contributions to rheology in the case of strong inclusions. In addition, we found that strong inclusions deform in the transient viscoelastic stress regime at high Weissenberg numbers (≥0.01) up to bulk shear strains larger than 3. Studies using the shapes of deformed objects for finite-­strain analysis or viscosity-­ratio estimation should establish carefully which rheology and loading conditions reflect material and deformation properties. We suggest that relatively strong, deformable clasts in shear zones retain stored energy up to fairly high shear strains. Hence, purely viscous models of clast deformation may overlook an important contribution to the energy budget, which may drive dissipation processes within and around natural inclusions.