Skills Showcases

SKILLS SHOWCASE 1
Date: 30 June 2022
Time: 11h00 – 12h30
Venue: Venue 6 and On-line

Deep learning and bioacoustics
Dr Emmanuel Dufourq1,2,3
1 African Institute for Mathematical Sciences, Cape Town, South Africa
2 School for Data Science and Computational Thinking, Stellenbosch University, Stellenbosch, South Africa
3 Department of Industrial Engineering, Stellenbosch University, Stellenbosch, South Africa

Tremendous progress has been made in bioacoustics research via deep learning algorithms. Recent scientific articles focusing on the use and development of bioacoustic classification models all use convolutional neural networks. This skills showcase will provide participants with fundamental knowledge in convolutional neural networks to better understand how they are implemented to create an acoustic classifier model.

Participants will be introduced to theoretical basics of 2D convolutional neural networks. This will be followed by an introduction to audio processing in Python and finally, participants will build a convolutional neural network to classify vocalisation events of the critically endangered Hainan gibbons. The same approach could be applied to species.

Programming will be done in Python 3 and Tensorflow using Google Colab. No software installation is required other than an Internet browser with a working Internet connection and a Google account.

SKILLS SHOWCASE 2
Date: 30 June 2022
Time: 11h00 – 12h30
Venue: Venue 4 & On-line

Model-based data integration: a primer and practical guide
Dr Saras Windecker University of Melbourne1, Associate Professor Nick Golding2
1 University of Melbourne, Parkville, VIC, Australia
2 Curtin University, Perth, WA, Australia

To make the most accurate prediction of a species’ distribution it is important to make use of all relevant data. Although opportunistically collected species records are more readily abundant and contain lots of data, they are also more likely to be biased. Standardised surveys may suffer less observation bias, but they are expensive to conduct and are likely to be more limited in spatial extent. Combining the two data types not only gives more data to a model, but also provides a chance for structured survey data to help correct biases in opportunistic data. Model-based data integration explicitly accounts for imperfect observation processes in both datasets, and propagates information contained in each while accounting for appropriate biases. In this skills showcase we will explain the concepts behind model-based data integration, and demonstrate its implementation in a Bayesian framework for broad spatial-scale presence-only data and spatial abundance data from structured surveys. We will illustrate how Point process models can be used to facilitate this integration. We expect that participants will leave more confident in the theory and execution of integrated models.

SKILLS SHOWCASE 3
Date: 30 June 2022
Time: 11h00 – 12h30
Venue: Venue 5 & On-line

Advances in piecewise estimation of path models
Dr. Jacob (Bob) Douma +31317482140 +31317482140 1, Dr. Jonathan Lefcheck (443) 482-2443 (443) 482-2443 2
1 Wageningen University, Wageningen, Gelderland, Netherlands
2 Smithsonian Environmental Research Center, Edgewater, Maryland, USA

Path modelling has become an indispensable tool for ecologists to understand and test hypothesis about the causal dependency between measured variables. Ecological data is oftentimes not normally distributed, non-linear, and clustered in time or space, limiting the application and reliability of inferences made from classical path analysis.

The method of d-separation, and its implementation in the popular R package ‘piecewiseSEM’, allows one to test causal hypotheses derived from causal models that can be described as directed acyclic graphs (DAGs, a causal structure only allowing directed causal effects and no feedback loops), and does not impose constraints on the normality, linearity or clustering of the data.

In the last few years, a number of advances have been for piecewiseSEM,: 1) an alternative statistical method for testing DAGs that is more in parity with classical path analysis; 2) an extension of this method and the d-sep method to test for differences among groups, and 3) extending the d-sep test to causal graphs that includes correlated errors, through the so-called m-sep test.

In this skills workshop you will be updated on the theory behind these advances and we will show you how these can be implemented in R, and in the R package piecewiseSEM in particular.

SKILLS SHOWCASE 4
Date: 30 June 2022
Time: 11h00 – 12h30
Venue: Venue 6 & On-line

Defensive Programming: How to Help Shield Your Code From Error
Dr Michael Bertolacci1
1 University Of Wollongong, Wollongong, New South Wales, Australia

Reflecting the use of increasingly sophisticated statistical methods, the computer programs we write to analyse data are becoming more and more complex. With complexity comes the risk of coding errors, which can waste time or even lead us to wrong conclusions. The more we can do to avoid these errors, the better.

In this skills showcase, I will share some best-practices from my former career as a software engineer that can help avoid coding errors before they arise, with a focus on the kinds of errors that can happen in coding for statistics. Wrong assumptions are the root cause of all programming errors, so I will focus on three techniques, coding standards, defensive programming, and automated testing, that help address that cause. Coding standards are the programming equivalent of good spelling and grammar, and help a programmer to understand their own creation so as to understand what assumptions have actually been made. Defensive programming makes those assumptions explicit in the code so that it’s clear when they’re violated. Automated testing reduces the burden on the programmer by using custom computer programs that check the truth of the assumptions. The techniques will be illustrated with R and Python examples.

Unfortunately, best-practice in software engineering is a vast area that is impossible to cover in a short session. So, at the conclusion of the session, I’ll provide some pointers to further material to help you improve as a programmer.