Need to check for arm64 processor

Sys.info()["machine"]
"arm64"

It might be helpful for people if there was a way they could create template analysis files, where they just "fill in the blanks". See usethis::use_test as an example.

Apparently resplot is quite slow compared to some other plots. Can we improve this?

Investigate automatic sizing

Especially Sommer but also lm/aov and lme4

Does not install because lmeInfo (a dependency of predictmeans) requires R >= 4.1.0.
Can we investigate why, or perhaps switch away from predictmeans?

We can just pass arguments to predict.asreml() via ...

If treatments are not factors, the multiple_comparisons() function will fail, while the rest of an aov() analysis will work fine.

This package installs 181 other packages on installation, can we reduce that a little?

To get a p-value for every comparison between pairs.

At present there is a default order for interactions, we could introduce an argument/option to swap the order. Or perhaps specify the order of factors?
Perhaps this is better just as DIY ggplot code?

More general power transformations could be provided, if determined to be necessary (e.g. via Box-Cox).

It does not install properly because it requires an additional folder to be created.

To rename the package

• Clone to new repo

BiometryTraining

• Add notice about rename of package
• Merge BiometryTraining dev branch to Master
• Build final binary versions

biometryassist

• Change all instances of BiometryTraining to biometryassist
• Update function names
  • mct to multiple_comparison
  • logl.test to logl_test()
  • des.info() to des_info()
• Update function arguments from . to _
• Run check on Rhub
• Submit

We need to automatically test against W2 code, to check that updates don't introduce bugs or errors affecting the code.

Unless specifically specified, they just add confusion.

For example, enable entering factorial treatments with labels and levels straight into treatments =
For example:

design("crossed:crd", treatments = list(PD = c(200, 260, 330), Irr = c(20, 100, 180)), reps = 4, nrows = 6, ncols = 6)

Also, allow subplot treatments to be specified in the same way in split plot if desired.

autoplot(pred.out, rotation = 90, label_rotation = 45)

Should rotate the axes by 90 degrees and the labels by 45 degrees, but doesn't.

use theme_bw if possible

viridisLite has new palettes: rocket, turbo and mako

This can replace ellipsis and we can also:

• check for interactivity to enable testing of the package startup message
• check required packages are installed
• check arguments missing or required
• Tidy evaluation of arguments?

Useful for plotting response against spatial variables

Hi, I have a weird issue that I'm not sure how to solve. I have a data .csv file with columns 'name', 'block', 'year', 'rep', and maturity date('mat'). On the same file is a column for larvae ('la') and another column for stem breakage ('sb'), which are the response data. Everything works great with the larvae reponse data, but get a error for running the stem breakage data...

I read that there are ways to correct by how the data is read in, but not sure where to start with this since the file is the same for data that works and for the data that doesn't.

Code:
e<-read.csv2("720_raw_blocking.csv",header=TRUE,sep=",")

library(data.table)
library(ggplot2)
library(jsonlite)
library(biometryassist)
library(asreml)

e$year <- as.factor(e$year)
e$name <- as.factor(e$name)
e$Block <- as.factor(e$Block)
e$rep <- as.factor(e$rep)
e$mat <- as.numeric(e$mat)
e$la_pro <- as.numeric(e$la_pro)
e$sb_pro <- as.numeric(e$sb_pro)

current.asr <- asreml (fixed = la_pro ~ name + year + mat, random = ~Block + rep + Block:rep,
data=e,
family=asr_binomial(link = "logit", dispersion = 1, total = 5),
na.action = na.method(x="include")
)

##or wih the response variale sb_pro if running for the sb data.

wald(current.asr)

pred.out <-multiple_comparisons(
current.asr,
classify = "name",
sig = 0.05,
int.type = "ci",
trans = NA,
offset = NA,
decimals = 2,
descending = TRUE,
plot = FALSE,
label_height = 0.1,
rotation = 0,
save = TRUE,
savename = "predicted_values_la2", #savename changes pending on the analysis
)
pred.out

##Output from when code from biometryassist is run ('pred.out <- multiple comparisions..") with the sb_pro response data (no error with the la_pro response data is used):

Binomial; Logit Mu=P=1/(1+exp(-XB)); V=Mu(1-Mu)/N
Note: The LogLik value is unsuitable for comparing GLM models

Deviance from GLM fit: 1812.25
Variance heterogenity factor (Deviance/df): 0.79
(assuming 2297 degrees of freedom)
Binomial; Logit Mu=P=1/(1+exp(-XB)); V=Mu(1-Mu)/N
Note: The LogLik value is unsuitable for comparing GLM models

Deviance from GLM fit: 1812.25
Variance heterogenity factor (Deviance/df): 0.79
(assuming 2297 degrees of freedom)
Calculating denominator DF
Binomial; Logit Mu=P=1/(1+exp(-XB)); V=Mu(1-Mu)/N
Note: The LogLik value is unsuitable for comparing GLM models

Deviance from GLM fit: 1812.25
Variance heterogenity factor (Deviance/df): 0.79
(assuming 2297 degrees of freedom)
Binomial; Logit Mu=P=1/(1+exp(-XB)); V=Mu(1-Mu)/N
Note: The LogLik value is unsuitable for comparing GLM models

Deviance from GLM fit: 1812.25
Variance heterogenity factor (Deviance/df): 0.79
(assuming 2297 degrees of freedom)
Calculating denominator DF

Error in levels<-(*tmp*, value = as.character(levels)) :
factor level [20] is duplicated

##More info on the raw data.

'data.frame': 6144 obs. of 6 variables:
$ year : Factor w/ 2 levels "2020","2021": 1 1 1 1 1 1 1 1 1 1 ...
$ rep : Factor w/ 2 levels "1","2": 1 1 1 1 1 1 1 1 1 1 ...
$ Block : Factor w/ 5 levels "1","2","3","4",..: 2 2 2 3 3 3 3 3 3 3 ...
$ name : Factor w/ 724 levels "FC004002B","FC029333",..: 1 2 3 4 5 6 7 8 9 10 ...
$ mat : num 104 102 104 98 100 98 108 108 107 105 ...
$ la_pro : num 0.6 0.4 0.8 1 0.2 1 0.4 0.4 0.4 0.8 ...
$ sb_pro : num 0.2 0.0667 0.36 0.12 0.04 ...

Thank you in advance for looking this over!

Set default value of offset when missing to 0 and warn.

Green being low values and red being high? Although that gives the impression that high is "bad" which is not necessarily the case.

autoplot function doesn't show up in the R help pages (pkgdown website is fine), and conflicts with ggplot2 when searching

Aliased levels are added as a single string rather than a vector, which makes printing a little difficult.

Currently gives an unhelpful error such as:

Error in names(diffs) <- m : 
  'names' attribute [45] must be the same length as the vector [0]

It's a common question, so let's write a vignette to explain the steps.
I wonder if we can have the asreml key behind a login website like the guest wifi password?

This has been a change for a while, time to start removing resplt().

THe install_asreml() function could be refactored into a download component and an install component, possibly even refactoring some of the other functions within it as well.
This may make testing easier and enable more robust checking across different operating systems.

Check lmPerm for inspiration

A couple of people have suggested that coefficient of variation is sometimes requested, and they would like to be able to get it.

For example:

dat.asr <- asreml(response ~ Genotype, random = ~ Row + Block, 
                  residual = ~ id(Column):id(Row), data = dat)

logl.test(dat.asr, rand.terms = NULL, resid.terms = "Row")

should fail or warn, because Row is already id(Row) in resid.

Hey Sam, as suggested:

pred.out2 <- multiple_comparisons(
  model2,
  classify = 'liming_treatment:rate:year',
  present = c('year', 'control', 'liming_treatment', 'rate')
)

werks

But:

pred.out2 <- multiple_comparisons(
  model2,
  classify = 'liming_treatment:rate:year',
  present = c('year', 'control', 'liming_treatment', 'rate'),
  sed = TRUE
)

doesn't, and errors with this:

Error in asreml::predict.asreml(model.obj, classify = classify, sed = TRUE,  : 
  formal argument "sed" matched by multiple actual arguments
Error:
! Arguments in `...` must be used.
✖ Problematic arguments:
• present = c("year", "control", "liming_treatment", "rate")
• sed = TRUE
Run `rlang::last_error()` to see where the error occurred.

Looks like it's reading in the sed argument twice, so perhaps a check to make sure that arguments passed with ... are needed or not?

Hi

We preparing the next release of ggplot2 and our reverse dependency checks show that your package is failing with the new version. Looking into it we see that this is due to changes in the warnings and errors thrown by ggplot2 that you test for in your package

You can install the release candidate of ggplot2 using devtools::install_github('tidyverse/[email protected]') to test this out.

We plan to submit ggplot2 by the end of October and hope you can have a fix ready before then

Kind regards
Thomas

Otherwise it will get annoying. Need to think about how to do this without adding loads more dependencies though

Agricolae has very poor documentation which could confuse users - can we improve this somehow?

We could wrap the design functions in shiny to enable people to use them graphically.

Currently when not provided, it gives an unhelpful error about

Error in 0:(ncols - 1) : result would be too long a vector

A little more in depth example than what's shown in the examples

Some ideas:

• Generating a complex factorial design (e.g. crossed but with control treatment(s) that aren't crossed)
• Analysis of a complex factorial design as above
• Interpreting residual plots
• Interpreting variograms
• updates in the new version

Hi I have NA's in my data and getting an error that 'NA is not allowed'. Is there a way to say 'include' or 'omit'?

Here is my code if at all helpful:
asreml_Iblock_di <- asreml (fixed = cbind (la_pro, sb_pro) ~ name + year + mat, random = ~rep + Block + rep:Block,
residual = ~id(units):us(trait),
data=e,
family=asr_binomial(link = "logit", dispersion = 1, total = 5),
na.action = na.method(x="include")
)
wald(asreml_Iblock_di, denDF = "default", ssType = "conditional")
pred.asr <- predict(asreml_Iblock_di, classify = "name", sed = TRUE)
pred.out <- mct.out(model.obj = asreml_Iblock_di, pred.obj = pred.asr,
classify = "name", order = "descending", decimals = 5, save = TRUE, savename = "predicted_values_05_23.csv")

E.g.

library(tidyverse)

dat <- read.csv("example1.csv")
dat <- dat %>% mutate(across(c(1:3), factor))
dat.aov <- aov(RL ~ trt, data = dat)            # fitting the model
pred.out <- mct.out(model.obj = dat.aov, classify = "trt")

example1.csv

If a column name is provided that is too similar to the output columns, multiple_comparisons() will change the names of all the columns between the first and the maximum matching column to factors.

See example code below:

# Any of these names will break the function
c("predicted.value", "std.error", "Df",
  "groups", "PredictedValue", "ApproxSE", "ci", "low", "up")
library(biometryassist)
dat <- design("crd", LETTERS[1:4], 4, nrow = 4, ncols = 4)$design
names(dat)[5] <- "groups"
dat$response <- rnorm(16, 10)
dat.aov <- aov(response~groups, data = dat)
multiple_comparisons(dat.aov, classify = "groups", trans = "log", offset = 0)

Hi! This is not so much a bug but a question. Can anyone recommend a r package that can take an asreml model and do a kruskal wallis with a multiple comparison output? My data is far from normal...
Thank you for your time,
Sarah

Need to update multiple_comparisons() to use Mean and SE instead of previous values, but with defence against old package versions.

We should defend against this a bit, so that it can be used for other designs with just the addition of the design class to the data frame.

Probably not very useful to have the warnings if the model is not converging after 10 iterations. Could try changing step size after the final iteration if not converging.

Can we do this without the predictmean package? This may make biometryhub/BiometryTraining#78 easier?