Search

BEF studies aim to understand how ecosystems respond to a gradient of species diversity. Diversity-Interactions (DI) models are suitable for analysing the BEF relationship. These models relate an ecosystem function response of a community to the identity of the species in the community, their evenness (proportions) and interactions. The number of species in the community (richness) is also implicitly modelled through this approach. It is common in BEF studies to model an ecosystem function as a function of richness; while this can uncover trends in the BEF relationship, by definition, species diversity is much broader than richness alone, and important patterns in the BEF relationship may remain hidden. In this paper, we introduce the DImodels R package for implementing DI models. We also compare DI models to traditional modelling approaches to highlight the advantages of using a multi-dimensional definition of species diversity. We show that using DI models can lead to considerably improved model fit over other methods; it does this by incorporating variation due to the multiple facets of species diversity. Predicting from a DI model is not limited to the study design points, the model can extrapolate to predict for any species composition and proportions (assuming there is sufficient coverage of this space in the study design). Expressing the BEF relationship as a function of richness alone can be useful to capture overall trends. However, collapsing the multiple dimensions of species diversity to a single dimension (such as richness) can result in valuable ecological information being lost. DI modelling provides a framework to test the multiple components of species diversity in the BEF relationship. It facilitates uncovering a deeper ecological understanding of the BEF relationship and can lead to enhanced inference.

We present vivid, an R package for visualizing variable importance and variable interactions in machine learning models. The package provides a range of displays including heatmap and graph-based displays for viewing variable importance and interaction jointly and partial dependence plots in both a matrix layout and an alternative layout emphasizing important variable subsets. With the intention of increasing a machine learning models' interpretability and making the work applicable to a wider readership, we discuss the design choices behind our implementation by focusing on the package structure and providing an in-depth look at the package functions and key features. We also provide a practical illustration of the software in use on a data set., Comment: 15 pages, 7 figures

Tree-based regression and classification has become a standard tool in modern data science. Bayesian Additive Regression Trees (BART) has in particular gained wide popularity due its flexibility in dealing with interactions and non-linear effects. BART is a Bayesian tree-based machine learning method that can be applied to both regression and classification problems and yields competitive or superior results when compared to other predictive models. As a Bayesian model, BART allows the practitioner to explore the uncertainty around predictions through the posterior distribution. In this paper, we present new visualization techniques for exploring BART models. We construct conventional plots to analyze a model's performance and stability as well as create new tree-based plots to analyze variable importance, interaction, and tree structure. We employ Value Suppressing Uncertainty Palettes (VSUP) to construct heatmaps that display variable importance and interactions jointly using color scale to represent posterior uncertainty. Our new visualizations are designed to work with the most popular BART R packages available, namely BART, dbarts, and bartMachine. Our approach is implemented in the R package bartMan (BART Model ANalysis)., Comment: 25 pages, 15 figures

Variable importance, interaction measures, and partial dependence plots are important summaries in the interpretation of statistical and machine learning models. In this paper we describe new visualization techniques for exploring these model summaries. We construct heatmap and graph-based displays showing variable importance and interaction jointly, which are carefully designed to highlight important aspects of the fit. We describe a new matrix-type layout showing all single and bivariate partial dependence plots, and an alternative layout based on graph Eulerians focusing on key subsets. Our new visualizations are model-agnostic and are applicable to regression and classification supervised learning settings. They enhance interpretation even in situations where the number of variables is large. Our R package vivid (variable importance and variable interaction displays) provides an implementation., Comment: 31 pages, 10 figures

Machine learning models fit complex algorithms to arbitrarily large datasets. These algorithms are well-known to be high on performance and low on interpretability. We use interactive visualization of slices of predictor space to address the interpretability deficit; in effect opening up the black-box of machine learning algorithms, for the purpose of interrogating, explaining, validating and comparing model fits. Slices are specified directly through interaction, or using various touring algorithms designed to visit high-occupancy sections or regions where the model fits have interesting properties. The methods presented here are implemented in the R package \pkg{condvis2}., Comment: 35 pages, 11 figures

The condvis package is for interactive visualization of sections in data space, showing fitted models on the section, and observed data near the section. The primary goal is the interpretation of complex models, and showing how the observed data support the fitted model. There is a video accompaniment to this paper available at https://www.youtube.com/watch?v=rKFq7xwgdX0. This is a preprint version of an article to appear in the Journal of Statistical Software.