Browsing by Department "Professorship Vehtari Aki"

Now showing 1 - 6 of 6
  • Thumbnail Image
    Item
    Bayesian leave-one-out cross-validation for large data
    (PMLR, 2019) Magnusson, Måns; Andersen, Michael; Jonasson, Johan; Vehtari, Aki; Professorship Vehtari Aki; Department of Computer Science; Chalmers University of Technology; Probabilistic Machine Learning
    Model inference, such as model comparison, model checking, and model selection, is an important part of model development. Leave-one-out cross-validation (LOO) is a general approach for assessing the generalizability of a model, but unfortunately, LOO does not scale well to large datasets. We propose a combination of using approximate inference techniques and probability-proportional-to-size-sampling (PPS) for fast LOO model evaluation for large datasets. We provide both theoretical and empirical results showing good properties for large data.
  • Thumbnail Image
    Item
    Hidden impacts of conservation management on fertility of the critically endangered kākāpō
    (PeerJ, 2023-02) Digby, Andrew; Eason, Daryl; Catalina, Alejandro; Lierz, Michael; Galla, Stephanie; Urban, Lara; Le Lec, Marissa F.; Guhlin, Joseph; Steeves, Tammy E.; Dearden, Peter K.; Joustra, Tineke; Lees, Caroline; Davis, Tane; Vercoe, Deidre; , Kākāpō Recovery Team; Professorship Vehtari Aki; Justus Liebig University Giessen; University of Canterbury; University of Otago; Department of Computer Science
    Background. Animal conservation often requires intensive management actions to improve reproductive output, yet any adverse effects of these may not be immediately apparent, particularly in threatened species with small populations and long lifespans. Hand-rearing is an example of a conservation management strategy which, while boosting populations, can cause long-term demographic and behavioural problems. It is used in the recovery of the critically endangered kākāpō (Strigops habroptilus), a flightless parrot endemic to New Zealand, to improve the slow population growth that is due to infrequent breeding, low fertility and low hatching success. Methods. We applied Bayesian mixed models to examine whether hand-rearing and other factors were associated with clutch fertility in kākāpō. We used projection predictive variable selection to compare the relative contributions to fertility from the parents’ rearing environment, their age and previous copulation experience, the parental kinship, and the number of mates and copulations for each clutch. We also explored how the incidence of repeated copulations and multiple mates varied with kākāpō density. Results. The rearing status of the clutch father and the number of mates and copulations of the clutch mother were the dominant factors in predicting fertility. Clutches were less likely to be fertile if the father was hand-reared compared to wild-reared, but there was no similar effect for mothers. Clutches produced by females copulating with different males were more likely to be fertile than those from repeated copulations with one male, which in turn had a higher probability of fertility than those from a single copulation. The likelihood of multiple copulations and mates increased with female:male adult sex ratio, perhaps as a result of mate guarding by females. Parental kinship, copulation experience and age all had negligible associations with clutch fertility. Conclusions. These results provide a rare assessment of factors affecting fertility in a wild threatened bird species, with implications for conservation management. The increased fertility due to multiple mates and copulations, combined with the evidence for mate guarding and previous results of kākāpō sperm morphology, suggests that an evolutionary mechanism exists to optimise fertility through sperm competition in kākāpō. The high frequency of clutches produced from single copulations in the contemporary population may therefore represent an unnatural state, perhaps due to too few females. This suggests that opportunity for sperm competition should be maximised by increasing population densities, optimising sex ratios, and using artificial insemination. The lower fertility of hand-reared males may result from behavioural defects due to lack of exposure to conspecifics at critical development stages, as seen in other taxa. This potential negative impact of hand-rearing must be balanced against the short-term benefits it provides.
  • Thumbnail Image
    Item
    State space Gaussian processes with non-Gaussian likelihood
    (PMLR, 2018) Nickisch, Hannes; Solin, Arno; Grigorevskiy, Alexander; Philips Research Laboratories Germany; Professorship Solin A.; Professorship Vehtari Aki; Department of Computer Science; Dy, Jennifer; Krause, Andreas
    We provide a comprehensive overview and tooling for GP modelling with non-Gaussian likelihoods using state space methods. The state space formulation allows for solving one-dimensonal GP models in O(n) time and memory complexity. While existing literature has focused on the connection between GP regression and state space methods, the computational primitives allowing for inference using general likelihoods in combination with the Laplace approximation (LA), variational Bayes (VB), and assumed density filtering (ADF) / expectation propagation (EP) schemes has been largely overlooked. We present means of combining the efficient O(n) state space methodology with existing inference methods. We also furher extend existing methods, and provide unifying code implementing all approaches.
  • Thumbnail Image
    Item
    thurstonianIRT: Thurstonian IRT Models in R
    (2019) Burkner, Paul-Christian; Professorship Vehtari Aki; Department of Computer Science
  • Thumbnail Image
    Item
    Using Stacking to Average Bayesian Predictive Distributions (with Discussion)
    (2018-09) Yao, Yuling; Vehtari, Aki; Simpson, Daniel; Gelman, Andrew; Columbia University; Professorship Vehtari Aki; University of Toronto; Department of Computer Science
    Bayesian model averaging is flawed in the M-open setting in which the true data-generating process is not one of the candidate models being fit. We take the idea of stacking from the point estimation literature and generalize to the combination of predictive distributions. We extend the utility function to any proper scoring rule and use Pareto smoothed importance sampling to efficiently compute the required leave-one-out posterior distributions. We compare stacking of predictive distributions to several alternatives: stacking of means, Bayesian model averaging (BMA), Pseudo-BMA, and a variant of Pseudo-BMA that is stabilized using the Bayesian bootstrap. Based on simulations and real-data applications, we recommend stacking of predictive distributions, with bootstrapped-Pseudo-BMA as an approximate alternative when computation cost is an issue.
  • Thumbnail Image
    Item
    Variable selection for Gaussian processes via sensitivity analysis of the posterior predictive distribution
    (PMLR, 2019-04-16) Paananen, Topi; Piironen, Juho; Andersen, Michael; Vehtari, Aki; Professorship Vehtari Aki; Probabilistic Machine Learning; Department of Computer Science
    Variable selection for Gaussian process models is often done using automatic relevance determination, which uses the inverse lengthscale parameter of each input variable as a proxy for variable relevance. This implicitly determined relevance has several drawbacks that prevent the selection of optimal input variables in terms of predictive performance. To improve on this, we propose two novel variable selection methods for Gaussian process models that utilize the predictions of a full model in the vicinity of the training points and thereby rank the variables based on their predictive relevance. Our empirical results on synthetic and real world data sets demonstrate improved variable selection compared to automatic relevance determination in terms of variability and predictive performance.