Natalya Pya Arnqvist

Short track:

2019 - present: Associate Professor, Umeå University
2017 - 2019: Project assistant, Umeå University
2016 - 2017: Senior research engineer, Umeå University
2015 - 2020: Assistant Professor (on leave 2016-2020), Nazarbayev University, Kz
2011 - 2015: Research Associate, University of Bath, UK
2010 - 2011: Assistant Professor,  KIMEP University, Kz
2006 - 2017: Senior research fellow (part-time), Institute for Mathematics,  Kz
2007 - 2010: PhD in Statistics, University of Bath, UK
2005 - 2007: Senior Lecturer, KIMEP University, Kz
2000 - 2005: CSc, Candidate of Physical and Math Sciences, Institute for Mathematics, Kz

Research:

I have two main research interests, statistical regression modelling and functional data analysis. I am specifically interested in developing methods for shape preserving smoothing within generalized additive models and applications of shape constrained additive models (SCAMs). I've written an R package `scam' which implements SCAM. The short description of this package is given below. I am also interested in methods for clustering functional data with optional scalar covariates and applications of functional clustering.

scam: Shape constrained additive models

• `scam' is an R package that implements generalized additive modelling under shape constraints on the component functions of the linear predictor.
• Models can contain multiple shape constrained and unconstrained terms as well as bivariate smooths with double or single monotonicity.
• Univariate smooths under various possible shape constraints including monotonically increasing/decreasing, convex/concave, increasing/decreasing and convex, increasing/decreasing and concave, are available as model terms.
• `scam' implements tensor product smooths for creating bivariate functions with shape constraints in one of the covariates or both covariates.
• The model set up is the same as in `gam' in the package `mgcv'  with the added shape constrained smooths. So the unconstrained smooths can be of more than one variable. Other user defined smooths can be also included as model terms.
• `scam' is based on penalized regression splines with automatic smoothness estimation.
• Smoothness selection in `scam' is by GCV or UBRE/AIC.
• A Bayesian approach is used to obtain a covariance matrix of the model coefficients and credible intervals for each smooth.
• as in `gam' in the package `mgcv' the linear preditor of a model in `scam' can depend on a bounded linear functional of a smooth (via a summation convention used in model specification). This allows scalar-on-function regression to be performed.

-------------------------------------------------------------------------------------------

fdaMocca: Model-based clustering for functional data with covariates. N. Pya Arnqvist, P. Arnqvist, S. Sjöstedt de Luna

• `fdaMocca' provides functions for model-based functional cluster analysis for functional data with optional covariates.
• The aim is to cluster a set of independent functional subjects into homogenous groups by using basis function representation of the functional data and allowing scalar covariates.
• A functional subject is defined as a curve and covariates. The spline coefficients and the (potential) covariates are modelled as a multivariate Gaussian mixture model, where the number of mixtures corresponds to the number of (predefined) clusters.
• `mocca' allows for different cluster covariance structures for the basis coefficients and for the covariates.

-------------------------------------------------------------------------------------------

nilde: Nonnegative integer solutions of linear diophantine equations with applications. N. Pya Arnqvist, V. Voinov, Y. Voinov

• `nilde' is an R package that provides functions for enumerating all existing nonnegative integer solutions of a linear Diophantine equation.
• `nilde' also includes functions for solving 0-1, bounded and unbounded knapsack problems; 0-1, bounded and unbounded subset sum problems; a problem of additive partitioning of natural numbers; and one-dimensional bin-packing problem
• The algorithm is based on a generating function of Hardy and Littlewood used by Voinov and Nikulin (1997)

-------------------------------------------------------------------------------------------

Some other R packages:

• fiberLD: Fiber Length Determination. N. Pya Arnqvist, S. Sjöstedt de Luna, K. Abramowicz 
• mvnTest: Goodness of Fit Tests for Multivariate Normality. N. Pya, V. Voinov, R. Makarov, Y. Voinov