Merge pull request #100 from florianhartig/0.2.0.3

0.2.0.3

Code/DHARMaCRAN/Checklist.R

@@ -0,0 +1,8 @@
+Submission Checklist
+
+
+
+
+
+
+https://cran.r-project.org/web/checks/check_results_DHARMa.html

Code/DHARMaPackageSupport/glmmTMB/glmmTMB.Rmd

@@ -6,36 +6,18 @@ output:
   html_document: 
     keep_md: yes
     toc: yes
-editor_options: 
-  chunk_output_type: console
 ---
 
 ```{r setup, include=FALSE}
 knitr::opts_chunk$set(echo = TRUE, cache = T)
 ```
 
 ```{r}
-devtools::install_github("glmmTMB/glmmTMB/glmmTMB")
-
 library(DHARMa)
 library(glmmTMB)
 ```
 
-# Testing general package behavior general problems
-
-```{r}
-testData = createData(sampleSize = 200, overdispersion = 0.5, family = poisson())
-
-m1 <- glmmTMB(observedResponse ~ Environment1 + (1|group), 
-                     family = "poisson", data = testData)
-
-predict(m1)
-
-
-simulationOutput <- simulateResiduals(fittedModel = m1)
-
-```
-
+# Testing general package behavior / general problems
 
 
 ## Pearson residuals don't work with zi terms (not implemented)
@@ -60,30 +42,6 @@ boxplot(Salamanders$mined, predict(m1, type = c("response")))
 ```
 
 
-```{r}
-testData = createData(sampleSize = 200, randomEffectVariance = 2, family = poisson())
-
-m1 <- glmmTMB(observedResponse ~ Environment1 + (1|group), 
-                     family = "poisson", data = testData)
-
-summary(m1)
-
-res <- simulateResiduals(fittedModel = m1)
-
-# this plot shows the residual pattern, despite the model being correct
-plot(res)
-
-# recalculate predictions by hand - see help ?predict.glmmTMB
-# To compute population-level predictions for a given grouping variable (i.e., setting all random effects for that grouping variable to zero), set the group value to NA.
-newdata=testData
-newdata$group = NA
-pred = predict(m1, newdata = newdata)
-
-# Now perform the plot
-plotResiduals(pred, res$scaledResiduals)
-```
-
-
 # DHARMa with glmmTMB example cases 
 
 ## Test 1 - Simple poisson model
@@ -354,5 +312,23 @@ sessionInfo()
 ```
 
 
+Hi guys, I'm testing glmmTMB for DHARMA, and I'm encountering a problem with the development versin on GitHub, the following code runs with the CRAN version, but not with the GitHub version
+
+https://github.com/glmmTMB/glmmTMB/commit/b0084337f92736dfe84fd0f6386c1180a6103181
+
+
+```{r}
+library(DHARMa)
+library(glmmTMB)
+testData = createData(sampleSize = 500, intercept = 2, fixedEffects = c(1), overdispersion = 0, family = poisson(), quadraticFixedEffects = c(-3), randomEffectVariance = 0, pZeroInflation = 0.6)
+fittedModel <- glmmTMB(observedResponse ~ Environment1 + I(Environment1^2) + (1|group), ziformula = ~1 , family = "poisson", data = testData)
+summary(fittedModel)
+predict(fittedModel, type = "response") 
+
+simulationOutput <- simulateResiduals(fittedModel = fittedModel)
+plot(simulationOutput)
+```
+
+
 
 

Code/DHARMaPackageSupport/spaMM/spaMM-test.R

@@ -1,67 +1,50 @@
 library(spaMM)
 
 data("Loaloa") ## from spaMM package
-loafit <- corrHLfit(cbind(npos,ntot-npos)~maxNDVI1+seNDVI
+fittedModel <- corrHLfit(cbind(npos,ntot-npos)~maxNDVI1+seNDVI
                     +Matern(1|longitude+latitude),
                     data=Loaloa[1:30,],family=binomial(),
                     init.corrHLfit=list(Nugget=0.1),ranFix=list(nu=0.5))
 
+testModel(fittedModel)
 
-########################
 
-fittedModel = loafit
 
-family(fittedModel)
-class(fittedModel)[1]
-nobs(fittedModel)
-getResponse(fittedModel)
-
-getResponse.HLfit <- function(object, ...){
-  response(object, ...)
-}
+DHARMa:::getFixedEffects(fittedModel)
 
-simulate.HLfit(fittedModel, nsim = 2)
-predict.HLfit(fittedModel)
-coef(fittedModel)
-ranef(fittedModel)
-fixef(fittedModel)
+res = simulateResiduals(fittedModel)
+plot(res)
 
-DHARMa:::getFixedEffects(fittedModel)
+testData = createData(sampleSize = 10000, overdispersion = 0.0, randomEffectVariance = 1, family = poisson())
 
+fittedModel <- HLfit(observedResponse ~ Environment1 + (1|group) , family = "poisson",  data = testData)
 res = simulateResiduals(fittedModel)
 plot(res)
 
-coef(fittedModel)
-fixef(fittedModel)
+fittedModel <- glmer(observedResponse ~ Environment1 + (1|group) , family = "poisson",  data = testData)
+summary(fittedModel)
+res = simulateResiduals(fittedModel, re.form = ~0)
+plot(res)
+
 
 
 testData = createData(sampleSize = 200, overdispersion = 0.0, randomEffectVariance = 1, family = poisson())
 fittedModel <- HLfit(observedResponse ~ Environment1 + (1|group) , family = "poisson",  data = testData)
-res = simulateResiduals(fittedModel)
+res = simulateResiduals(fittedModel, refit = T)
 plot(res)
-testUniformity(res)$p.value < 0.001
+
+
 
 testData = createData(sampleSize = 200, overdispersion = 0, randomEffectVariance = 0, family = binomial(), binomialTrials = 20)
 fittedModel <- HLfit(cbind(observedResponse1,observedResponse0) ~ Environment1 + (1|group) , family = "binomial",  data = testData)
 res = simulateResiduals(fittedModel)
 plot(res)
-testUniformity(res)$p.value < 0.001
-
 
 testData = createData(sampleSize = 200, fixedEffects = c(1,0), overdispersion = 0, randomEffectVariance = 0, family = binomial())
 fittedModel <- HLfit(observedResponse ~ Environment1 + (1|group) , family = "binomial",  data = testData)
 res = simulateResiduals(fittedModel)
 plot(res)
-testUniformity(res)$p.value < 0.001
-
-
-########################
-
-
-simsCond = simulate(nsim=250, loafit, type="residual") # I think that this is what I call a conditional simulat, i.e. simulations are conditinal on the fitted REs
-simsUncond = simulate(nsim=250, loafit, type="marginal") # I think that this is what I call a unconditional simulation, i.e. random effects and glm distribution are re-siulated?
+testResiduals(res)
 
-createDHARMa(simulatedResponse = sims, observedResponse = loafit$y,
-             fittedPredictedResponse = predict(loafit, re.form=~0)[,1L], integer = T)
 
 

DHARMa/DESCRIPTION

@@ -1,13 +1,13 @@
 Package: DHARMa
 Title: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models
-Version: 0.2.0
-Date: 2018-06-06
+Version: 0.2.1
+Date: 2019-01-14
 Authors@R: c(person("Florian", "Hartig", email = "[email protected]", role = c("aut", "cre"), comment = "Theoretical Ecology, University of Regensburg, Regensburg, Germany"))
 Description: The 'DHARMa' package uses a simulation-based approach to create
     readily interpretable scaled (quantile) residuals for fitted (generalized) linear mixed
     models. Currently supported are (generalized) linear mixed models from 'lme4' 
-    (classes 'lmerMod', 'glmerMod') and 'glmmTMB', generalized additive models ('gam' from 'mgcv'), 
-    'glm' (including 'negbin' from 'MASS', but excluding quasi-distributions) and 'lm' model
+    (classes 'lmerMod', 'glmerMod'), 'glmmTMB' and 'spaMM', generalized additive models ('gam' from 
+    'mgcv'), 'glm' (including 'negbin' from 'MASS', but excluding quasi-distributions) and 'lm' model
     classes. Moreover, externally created simulations, e.g. posterior predictive simulations 
     from Bayesian software such as 'JAGS', 'STAN', or 'BUGS' can be processed as well. 
     The resulting residuals are standardized to values between 0 and 1 and can be interpreted 
@@ -32,7 +32,8 @@ Imports:
     MASS,
     lme4,
     mgcv,
-    glmmTMB (>= 0.2.1)
+    glmmTMB (>= 0.2.1),
+    spaMM (>= 2.6.0)
 Suggests:
     knitr,
     testthat

DHARMa/NAMESPACE

@@ -1,15 +1,22 @@
 # Generated by roxygen2: do not edit by hand
 
 S3method(fitted,gam)
+S3method(getResponse,HLfit)
+S3method(getResponse,default)
+S3method(getSimulations,HLfit)
+S3method(getSimulations,default)
 S3method(hist,DHARMa)
 S3method(plot,DHARMa)
 S3method(print,DHARMa)
+S3method(refit,HLfit)
 S3method(refit,glmmTMB)
 S3method(refit,lm)
 S3method(residuals,DHARMa)
 export(createDHARMa)
 export(createData)
 export(getRandomState)
+export(getResponse)
+export(getSimulations)
 export(plotConventionalResiduals)
 export(plotQQunif)
 export(plotResiduals)
@@ -19,6 +26,7 @@ export(runBenchmarks)
 export(simulateResiduals)
 export(testDispersion)
 export(testGeneric)
+export(testModel)
 export(testOutliers)
 export(testOverdispersion)
 export(testOverdispersionParametric)
@@ -33,4 +41,8 @@ import(graphics)
 import(stats)
 import(utils)
 importFrom(foreach,"%dopar%")
+importFrom(lme4,fixef)
+importFrom(lme4,ranef)
 importFrom(lme4,refit)
+importFrom(spaMM,response)
+importFrom(spaMM,update_resp)

DHARMa/NEWS

@@ -7,7 +7,7 @@ DHARMa 0.2.1
 New features
 
 - Outlier highlighting (in plots) and formal outlier test, implemented in https://github.com/florianhartig/DHARMa/pull/99
-- Support for the SPAMM package
+- Supporting now also models fit with the spaMM package
 
 Major changes
 

DHARMa/R/DHARMa.R

@@ -1,7 +1,7 @@
 #' @title DHARMa - Residual Diagnostics for HierArchical (Multi-level / Mixed) Regression Models
 #' @name DHARMa
 #' @docType package
-#' @description The 'DHARMa' package uses a simulation-based approach to create readily interpretable scaled (quantile) residuals for fitted generalized linear mixed models. Currently supported are generalized linear mixed models from 'lme4' (classes 'lmerMod', 'glmerMod') and 'glmmTMB', generalized additive models ('gam' from 'mgcv'), 'glm' (including 'negbin' from 'MASS', but excluding quasi-distributions) and 'lm' model classes. Alternatively, externally created simulations, e.g. posterior predictive simulations from Bayesian software such as 'JAGS', 'STAN', or 'BUGS' can be processed as well. The resulting residuals are standardized to values between 0 and 1 and can be interpreted as intuitively as residuals from a linear regression. The package also provides a number of plot and test functions for typical model misspecification problems, such as over/underdispersion, zero-inflation, and residual spatial and temporal autocorrelation.
+#' @description The 'DHARMa' package uses a simulation-based approach to create readily interpretable scaled (quantile) residuals for fitted (generalized) linear mixed models. Currently supported are (generalized) linear mixed models from 'lme4' (classes 'lmerMod', 'glmerMod'), 'glmmTMB' and 'spaMM', generalized additive models ('gam' from 'mgcv'), 'glm' (including 'negbin' from 'MASS', but excluding quasi-distributions) and 'lm' model classes. Moreover, externally created simulations, e.g. posterior predictive simulations from Bayesian software such as 'JAGS', 'STAN', or 'BUGS' can be processed as well. The resulting residuals are standardized to values between 0 and 1 and can be interpreted as intuitively as residuals from a linear regression. The package also provides a number of plot and test functions for typical model misspecification problems, such as over/underdispersion, zero-inflation, and residual spatial and temporal autocorrelation.
 #' @details See index / vignette for details
 #' @seealso \code{\link{simulateResiduals}}
 #' @examples