Scalable Visualization Methods for Modern Generalized Additive Models

Fasiolo, Matteo; Nedellec, Raphaël; Goude, Yannig; Wood, Simon N.

doi:10.1080/10618600.2019.1629942

Cited by 145 publications

(105 citation statements)

References 27 publications

(32 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We used R v.3.6.2 (R Core Team, 2019) and the packages lme4 v.1.1.21 (Bates et al, 2015), lmerTest v.3.0.1 (Kuznetsova et al, 2017), effects v.4.0.3 (Fox et al, 2019), car v.3.0.2 (Fox et al, 2011), tidyverse v.1.3.0 (Wickham et al, 2019), lattice v.0.20-38 (Sarkar, 2008), itsadug v.2.3 (van Rij et al, 2016, mgcv v.1.8-31 (Wood, 2006), mgcViz v.0.1.4 (Fasiolo et al, 2019) and rgl v.0.1.3 (Nenadic & Greenacre, 2007). Raincloud plots were produced to visualise behavioural data using the code provided by (Allen et al, 2019).…”

Section: Resultsmentioning

confidence: 99%

Oscillatory and aperiodic neural activity jointly predict language learning

Cross

Corcoran

Schlesewsky

et al. 2020

Preprint

View full text Add to dashboard Cite

word count: 250 Introduction word count: 1,309 Discussion word count: 2,372ABSTRACT Memory formation involves the synchronous firing of neurons in task-relevant networks, with recent models postulating that a decrease in low frequency oscillatory activity underlies successful memory encoding and retrieval. To date, this relationship has predominantly been investigated using objects (e.g., faces, natural scenes); however, considerably less is known about the oscillatory correlates of complex rule learning (e.g., language acquisition). Further, recent work has shown that aperiodic (non-oscillatory) 1/ƒ activity is functionally and behaviourally relevant, yet its interaction with oscillatory activity during complex rule learning remains virtually unknown. Using spectral decomposition and power-law exponent estimation of human EEG data, we show for the first time that 1/ƒ and oscillatory activity jointly influence the learning of different word order rules of a miniature language system. Fixed word order rules were associated with an increased power-law exponent (i.e. steeper 1/ƒ slope) compared to flexible word order rules. We also show that stronger anterior beta synchronisation predicts fixed word order rule learning and subsequent behavioural performance, while stronger theta/alpha synchronisation is associated with the learning of flexible word order rules. These results also revealed nonlinear differences between word order rules as a function of time and sensor space. Moreover, we demonstrated that inter-individual variations in spectral power across the learning task interacted with aperiodic activity to influence subsequent behavioural performance. Together, these results suggest that 1/ƒ activity plays an important role in higherorder cognition, including language processing, and that grammar learning is modulated by different word order permutations, which manifest in distinct oscillatory profiles.

show abstract

Section: Resultsmentioning

confidence: 99%

Oscillatory and aperiodic neural activity jointly predict language learning

Cross

Corcoran

Schlesewsky

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Data were processed and visualised in R using the contributed packages ‘tidyverse’ v1.3.0 (Wickham et al 2019), ‘lubridate’ v1.7.8 (Grolemund and Wickham 2011) and’ ‘clifro’ (Seers and Shears 2015) for wind roses. The best fitting GAM was visualised using a custom function created for this work that uses the contributed package ‘mgcViz’ v0.1.6 (Fasiolo et al 2019). Mean wind speed was calculated using a circular averaging function from ‘SDMTools’ v1.1-221.2 (VanDerWal et al 2014).…”

Section: Methodsmentioning

confidence: 99%

The role of conidia in the dispersal of Ascochyta rabiei

Khaliq

Fanning

Melloy

et al. 2020

Preprint

View full text Add to dashboard Cite

29Ascochyta rabiei asexual spores (conidia) were assumed to spread over short distances 30 (~10 m) in a combination of rain and strong wind. We investigated the potential distance of 31 conidial spread in three rainfall and three sprinkler irrigation events. Chickpea trap plants 32 were distributed at the distances of 0, 10, 25, 50 and 75 m from infected chickpea plots 33 before scheduled irrigation and forecast rainfall events. Trap plants were transferred to a 34 controlled temperature room (20 °C) for 48 h (100% humidity) after being exposed in the field 35 for 2-6 days for rainfall events, and for one day for irrigation events. After a 48 h incubation 36 period, trap plants were transferred to a glasshouse (20 °C) to allow lesion development. 37Lesions on all plant parts were counted after two weeks, which gave an estimate of the 38 number of conidia released and the distance travelled. Trap plants at all distances were 39 infected in all sprinkler irrigation and rainfall events. The highest number of lesions on trap 40 plants were recorded closest to the infected plots -the numbers decreased as the distance 41 from the infected plots increased. There was a positive relationship between the amount of 42 rainfall and the number of lesions recorded. A generalised additive model was developed that 43 efficiently described spatial patterns of conidial spread. With further development, the model 44 can be used to predict the spread of A. rabiei. This is the first systematic study to show that 45 conidia distribute A. rabiei over longer distances than previously reported. 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 Keywords: Ascochyta blight, epidemiology, conidial spread, wind-driven rain, chickpea, 62 sprinkler irrigation.63 Introduc)on 64 Chickpea (Cicer arietinum L.) is the second most important legume crop globally and is the 65 most widely grown legume grain crop in Australia with >1,060,000 ha harvested in 2018 66 (FAOSTAT 2020). Ascochyta blight caused by Ascochyta rabiei (syn. Phoma rabiei) is one of 67 the most devastating chickpea diseases worldwide (Pande et al. 2005). With the exception of 68 the Ord region in northern Western Australia, A. rabiei is the major biotic constraint to 69 chickpea production in Australia with almost all chickpea growing areas affected (Bretag et 70 al. 2008). Ascochyta rabiei survives on infected seed, volunteer chickpea plants and infested 71 stubble, forming conidia that initiate primary infection. Infected seed gives rise to infected 72 seedlings through transmission from germinating seeds. Conidia are spread by rain splash or 73 wind driven rain. Pycnidial formation, conidial production, host infection and disease 74 development are favoured by temperatures between 5 and 30 °C (optimum 20 °C), relative 75 humidity > 95 % (Nene 1982), and wetness period of 10 h or more (Khan 1999). A longer 76 wetness period is required for spore germination at sub-optimal temperatures, and infection 77 is rare in hot and dry conditions (Jhorar et al. 1998). Symptoms develop within 5...

show abstract

“…To assess model significance, tests with 1000 permutations were employed. The relationships between environmental variables and bacterial family richness as well as diversity (represented by the Shannon Index) were explored with generalized additive models (GAM), using the mgcv and mgcViz packages (Wood and Wood 2015; Fasiolo et al 2019), and with Kruskal-Wallis tests by rank and Mann-Whitney tests for comparisons between the variables. Finally, random forest models were applied to identify the critical environmental factors affecting the richness and diversity of the community (number of trees=10001).…”

Section: Methodsmentioning

confidence: 99%

Variation of bacterial communities along the vertical gradient in Lake Issyk Kul, Kyrgyzstan

Rojas-Jiménez

Araya-Lobo

Quesada-Perez

et al. 2019

Preprint

View full text Add to dashboard Cite

30In this study, we explored the diversity and community composition of bacteria along a 31 vertical gradient in Lake Issyk Kul, Kyrgyzstan, one of the largest and deepest brackish lakes 32 in in the world. We identified 4904 bacterial ASVs based on analysis of 16S rRNA gene 33 sequences and determined significant changes in the composition responding mainly to the 34 variables depth and salinity. A higher abundance of Proteobacteria and Bacteroidetes was 35 observed in the surface waters and the lake tributaries. Cyanobacteria were more abundant in 36 the deep chlorophyll maximum (28.5 to 128 m), while Planctomycetes and Chloroflexi were 37 dominant at depths between 128 to 600 m. According to our machine learning analysis used 38 for identifying the most critical environmental factors, depth and temperature revealed the 39 strongest effect on members of Proteobacteria, Planctomycetes, and Chloroflexi, while 40 oxygen is associated with the variations in Cyanobacteria. Also, a notable increase in alpha 41 diversity estimations was observed with increasing water depth. This work evidences 42 significant differences in the structure of bacterial communities along the depth gradient in 43 deep, transparent lake ecosystems. Notably, there is a dominance of Planctomycetes and 44

show abstract

Scalable Visualization Methods for Modern Generalized Additive Models

Cited by 145 publications

References 27 publications

Oscillatory and aperiodic neural activity jointly predict language learning

Oscillatory and aperiodic neural activity jointly predict language learning

The role of conidia in the dispersal of Ascochyta rabiei

Variation of bacterial communities along the vertical gradient in Lake Issyk Kul, Kyrgyzstan

Contact Info

Product

Resources

About