Scoping-related (?): anova() on list of created mixed-effects models - r

In a project where I'm performing mixed-effects modelling using lme, I'm trying to compare models with different correlation structures and equal fixed parts. As I'll be building a lot of these models (for different dependent variables), I tried to write a function to generate a list of models with different correlation structures, as in the example below (I really tried to keep it to a minimum working example).
If I run an anova() on the elements of this list, this works, but only if fixedPart is in my global environment. Why is this the case? Is there a way to circumvent this problem, so that I can just keep m and re-use/delete fixedPart?
I presume this problem is related to the (lexical) scoping in R, but I cannot find a way to actually fix it.
Thanks in advance!
#Dependencies
library(multilevel)
library(multcomp)
#Generate sample data
nVals = 100
sData = rnorm(nVals, mean = 1, sd = 1)
dF <- data.frame(nSubject = 1:nVals,
v1data = sData + rnorm(nVals, mean = 0, sd = 0.1),
v2data = sData + rnorm(nVals, mean = 0, sd = 0.1),
v3data = sData + rnorm(nVals, mean = 0, sd = 0.4))
dLongF = reshape(data=dF, varying=c("v1data","v2data","v3data"), v.names='data', direction="long", idvar="nSubject", times=1:3)
#Define function to assess different covariance structures
doAllCorrModels <- function(dataF, subjVarName, visitVarName, fixedPart){
mList <- vector("list",2)
mList[[1]] <- lme(fixedPart, #Random intercept, homogeneous variance
random=as.formula(paste("~1|", subjVarName)),
data=dataF,
weights=NULL)
mList[[2]] <- lme(fixedPart, #Random intercept, heterogeneous variance
random=as.formula(paste("~1|", subjVarName)),
data=dataF,
weights=varIdent(form = as.formula(paste("~1|", visitVarName)))
)
mList
}
#Get different covariance structures
dataF <- dLongF
subjVarName <- "nSubject"
visitVarName <- "time"
fixedPart <- data ~ time
m <- doAllCorrModels(dataF, subjVarName, visitVarName, fixedPart)
#This works:
a1 <- anova(m[[1]], m[[2]])
#But this does not:
rm(fixedPart)
a2 <- anova(m[[1]], m[[2]])

You can avoid this by using do.call:
doAllCorrModels <- function(dataF, subjVarName, visitVarName, fixedPart){
mList <- vector("list",2)
mList[[1]] <- do.call(lme, list(fixed = fixedPart,
random=as.formula(paste("~1|", subjVarName)),
data=dataF,
weights=NULL))
mList[[2]] <- do.call(lme, list(fixed = fixedPart,
random=as.formula(paste("~1|", subjVarName)),
data=dataF,
weights=varIdent(form = as.formula(paste("~1|", visitVarName)))))
mList
}

Related

Does caret::train() in r have a standardized output across different fit methods/models?

I'm working with the train() function from the caret package to fit multiple regression and ML models to test their fit. I'd like to write a function that iterates through all model types and enters the best fit into a dataframe. Biggest issue is that caret doesn't provide all the model fit statistics that I'd like so they need to be derived from the raw output. Based on my exploration there doesn't seem to be a standardized way caret outputs each models fit.
Another post (sorry don't have a link) created this function which pulls from fit$results and fit$bestTune to get pre calculated RMSE, R^2, etc.
get_best_result <- function(caret_fit) {
best = which(rownames(caret_fit$results) == rownames(caret_fit$bestTune))
best_result = caret_fit$results[best, ]
rownames(best_result) = NULL
best_result
}
One example of another fit statistic I need to calculate using raw output is BIC. The two functions below do that. The residuals (y_actual - y_predicted) are needed along with the number of x variables (k) and the number of rows used in the prediction (n). k and n must be derived from the output not the original dataset due to the models dropping x variables (feature selection) or rows (omitting NAs) based on its algorithm.
calculate_MSE <- function(residuals){
# residuals can be replaced with y_actual-y_predicted
mse <- mean(residuals^2)
return(mse)
}
calculate_BIC <- function(n, mse, k){
BIC <- n*log(mse)+k*log(n)
return(BIC)
}
The real question is is there a standardized output of caret::train() for x variables or either y_actual, y_predicted, or residuals?
I tried fit$finalModel$model and other methods but to no avail.
Here is a reproducible example along with the function I'm using. Please consider the functions above a part of this reproducible example.
library(rlist)
library(data.table)
# data
df <- data.frame(y1 = rnorm(50, 0, 1),
y2 = rnorm(50, .25, 1.5),
x1 = rnorm(50, .4, .9),
x2 = rnorm(50, 0, 1.1),
x3 = rnorm(50, 1, .75))
missing_index <- sample(1:50, 7, replace = F)
df[missing_index,] <- NA
# function to fit models and pull results
fitModels <- function(df, Ys, Xs, models){
# empty list
results <- list()
# number of for loops
loops_counter <- 0
# for every y
for(y in 1:length(Ys)){
# for every model
for(m in 1:length(models)){
# track loops
loops_counter <- loops_counter + 1
# fit the model
set.seed(1) # seed for reproducability
fit <- tryCatch(train(as.formula(paste(Ys[y], paste(Xs, collapse = ' + '),
sep = ' ~ ')),
data = df,
method = models[m],
na.action = na.omit,
tuneLength = 10),
error = function(e) {return(NA)})
# pull results
results[[loops_counter]] <- c(Y = Ys[y],
model = models[m],
sample_size = nrow(fit$finalModel$model),
RMSE = get_best_result(fit)[[2]],
R2 = get_best_result(fit)[[3]],
MAE = get_best_result(fit)[[4]],
BIC = calculate_BIC(n = length(fit$finalModel),
mse = calculate_MSE(fit$finalModel$residuals),
k = length(fit$finalModel$xNames)))
}
}
# list bind
results_df <- list.rbind(results)
return(results_df)
}
linear_models <- c('lm', 'glmnet', 'ridge', 'lars', 'enet')
fits <- fitModels(df, c(y1, y2), c(x1,x2,x3), linear_models)

Simulating a mixed linear model and evaluating it with lmerTest in R

I am trying to understand how to use mixed linear models to analyse my data by simulating a model, but I can't reproduce the input parameters. What am I missing?
I want to start simulating a model with a random intercept for each subject. Here is the formula of what I want to simulate and reproduce:
If beta1 (<11) is small I find gamma00 as the intercept in fixed section, but I am completedly unaable to retrieve the slope (beta1). Also, the linear effect is not significant. Where is my conceptual mistake?
library(lmerTest)
# Generating data set
# General values and variables
numObj <- 20
numSub <- 100
e <- rnorm(numObj * numSub, mean = 0, sd = 0.1)
x <- scale(runif(numObj * numSub, min = -100, max = 100))
y <- c()
index <- 1
# Coefficients
gamma00 <- 18
gamma01 <- 0.5
beta1 <- -100
w <- runif(numSub, min = -3, max = 3)
uo <- rnorm(numSub, mean = 0, sd = 0.1)
meanBeta0 <- mean(gamma00 + gamma01*w + uo) # I should be able to retrieve that parameter.
for(j in 1:numSub){
for(i in 1:numObj){
y[index] <- gamma00 + gamma01*w[j]+ uo[j] + beta1*x[i] + e[index]
index <- index + 1
}
}
dataFrame2 <- data.frame(y = y, x = x, subNo = factor(rep(1:numSub, each = numObj)), objNum = factor(rep(1:numObj, numSub)))
model2 <- lmer(y ~ x +
(1 | subNo), data = dataFrame2)
summary(model2)
anova(model2)
No conceptual mistake here, just a mixed up index value: you should be using index rather than i to index x in your data generation loop.
Basically due to the mix-up you were using the first subject's x values for generating data for all the subjects, but using the individual x values in the model.

simulation study using AIC

I have to code a simulation study in R. So, I have X1,...,X15~N(0,1) explanatory variables and Y~N(2+2*X1+0.8*X2­1.2*X15, 1) and I need to simulate n=100 values and repeat that for iter=100 times. Then, for each linear model created I have to calculate the AIC­values and, finally, find the best model. The problem is that I can't figure out how to do that for item=100 times. I wrote the code for 1 simulation, which is the following:
set.seed(123)
n<‐100
p<‐15
iter<‐100 X<‐matrix(rep(NA,n*p),ncol=p) for (j in 1:p) {
X[,j]<‐rnorm(n = 100, mean = 0, sd = 1) }
mu<‐(2+2*X[,1])+(0.8*X[,2])‐(1.2*X[,15]) Y<‐rnorm(n = 100, mean = mu , sd = 1)
sim<‐data.frame(Y,X)
d<‐lm(Y~X, data = sim)
But how I do the rest I have to do, i.e.the 100 simulations and the calculations of AIC? I'm really new to R, so I am quite confused.
How about this
nsim <- 100
nobs <- 100
nvar <- 15
results <- lapply(1:nsim, function(i) {
X <- matrix(rnorm(nobs*nvar),nrow=nobs)
y <- rnorm(nobs, mean=2 + X[,c(1,2,15)]%*% c(2, .8,-1.2))
DF <- data.frame(y, X)
lm(y ~ X, data=DF)})
That should give you your simulations. Now find the "best"
findbest <- which.min(sapply(results, function(i) { AIC(i) }))
results[[findbest]]
Since all data are simulated using the same underlying data-generating process any variation in AIC is essentially random variation.

How to estimate the Kalman Filter with 'KFAS' R package, with an AR(1) transition equation?

I am using 'KFAS' package from R to estimate a state-space model with the Kalman filter. My measurement and transition equations are:
y_t = Z_t * x_t + \eps_t (measurement)
x_t = T_t * x_{t-1} + R_t * \eta_t (transition),
with \eps_t ~ N(0,H_t) and \eta_t ~ N(0,Q_t).
So, I want to estimate the variances H_t and Q_t, but also T_t, the AR(1) coefficient. My code is as follows:
library(KFAS)
set.seed(100)
eps <- rt(200, 4, 1)
meas <- as.matrix((arima.sim(n=200, list(ar=0.6), innov = rnorm(200)*sqrt(0.5)) + eps),
ncol=1)
Zt <- 1
Ht <- matrix(NA)
Tt <- matrix(NA)
Rt <- 1
Qt <- matrix(NA)
ss_model <- SSModel(meas ~ -1 + SSMcustom(Z = Zt, T = Tt, R = Rt,
Q = Qt), H = Ht)
fit <- fitSSM(ss_model, inits = c(0,0.6,0), method = 'L-BFGS-B')
But it returns: "Error in is.SSModel(do.call(updatefn, args = c(list(inits, model), update_args)),: System matrices (excluding Z) contain NA or infinite values, covariance matrices contain values larger than 1e+07"
The NA definitions for the variances works well, as documented in the package's paper. However, it seems this cannot be done for the AR coefficients. Does anyone know how can I do this?
Note that I am aware of the SSMarima function, which eases the definition of the transition equation as ARIMA models. Although I am able to estimate the AR(1) coef. and Q_t this way, I still cannot estimate the \eps_t variance (H_t). Moreover, I am migrating my Kalman filter codes from EViews to R, so I need to learn SSMcustom for other models that are more complicated.
Thanks!
It seems that you are missing something in your example, as your error message comes from the function fitSSM. If you want to use fitSSM for estimating general state space models, you need to provide your own model updating function. The default behaviour can only handle NA's in covariance matrices H and Q. The main goal of fitSSM is just to get started with simple stuff. For complex models and/or large data, I would recommend using your self-written objective function (with help of logLik method) and your favourite numerical optimization routines manually for maximum performance. Something like this:
library(KFAS)
set.seed(100)
eps <- rt(200, 4, 1)
meas <- as.matrix((arima.sim(n=200, list(ar=0.6), innov = rnorm(200)*sqrt(0.5)) + eps),
ncol=1)
Zt <- 1
Ht <- matrix(NA)
Tt <- matrix(NA)
Rt <- 1
Qt <- matrix(NA)
ss_model <- SSModel(meas ~ -1 + SSMcustom(Z = Zt, T = Tt, R = Rt,
Q = Qt), H = Ht)
objf <- function(pars, model, estimate = TRUE) {
model$H[1] <- pars[1]
model$T[1] <- pars[2]
model$Q[1] <- pars[3]
if (estimate) {
-logLik(model)
} else {
model
}
}
opt <- optim(c(1, 0.5, 1), objf, method = "L-BFGS-B",
lower = c(0, -0.99, 0), upper = c(100, 0.99, 100), model = ss_model)
ss_model_opt <- objf(opt$par, ss_model, estimate = FALSE)
Same with fitSSM:
updatefn <- function(pars, model) {
model$H[1] <- pars[1]
model$T[1] <- pars[2]
model$Q[1] <- pars[3]
model
}
fit <- fitSSM(ss_model, c(1, 0.5, 1), updatefn, method = "L-BFGS-B",
lower = c(0, -0.99, 0), upper = c(100, 0.99, 100))
identical(ss_model_opt, fit$model)

binning continuous variables by IV value in R

I am building a logistic regression model in R. I want to bin continuous predictors in an optimal way in relationship to the target variable. There are two things that I know of:
the continuous variables are binned such that its IV (information value) is maximized
maximize the chi-square in the two way contingency table -- the target has two values 0 and 1, and the binned continuous variable has the binned buckets
Does anyone know of any functions in R that can perform such binning?
Your help will be greatly appreciated.
For the first point, you could bin using the weight of evidence (woe) with the package woebinning which optimizes the number of bins for the IV
library(woeBinning)
# get the bin cut points from your dataframe
cutpoints <- woe.binning(dataset, "target_name", "Variable_name")
woe.binning.plot(cutpoints)
# apply the cutpoints to your dataframe
dataset_woe <- woe.binning.deploy(dataset, cutpoint, add.woe.or.dum.var = "woe")
It returns your dataset with two extra columns
Variable_name.binned which is the labels
Variable_name.woe.binned which is the replaced values that you can then parse into your regression instead of Variable_name
For the second point, on chi2, the package discretization seems to handle it but I haven't tested it.
The methods used by regression splines to set knot locations might be considered. The rpart package probably has relevant code. You do need to penalize the inferential statistics because this results in an implicit hiding of the degrees of freedom expended in the process of moving the breaks around to get the best fit. Another common method is to specify breaks at equally spaced quantiles (quartiles or quintiles) within the subset with IV=1. Something like this untested code:
cont.var.vec <- # names of all your continuous variables
breaks <- function(var,n) quantiles( dfrm[[var]],
probs=seq(0,1,length.out=n),
na.rm=TRUE)
lapply(dfrm[ dfrm$IV == 1 , cont.var.vec] , breaks, n=5)
s
etwd("D:")
rm(list=ls())
options (scipen = 999)
read.csv("dummy_data.txt") -> dt
head(dt)
summary(dt)
mydata <- dt
head(mydata)
summary(mydata)
##Capping
for(i in 1:ncol(mydata)){
if(is.numeric(mydata[,i])){
val.quant <- unname(quantile(mydata[,i],probs = 0.75))
mydata[,i] = sapply(mydata[,i],function(x){if(x > (1.5*val.quant+1)){1.5*val.quant+1}else{x}})
}
}
library(randomForest)
x <- mydata[,!names(mydata) %in% c("Cust_Key","Y")]
y <- as.factor(mydata$Y)
set.seed(21)
fit <- randomForest(x,y,importance=T,ntree = 70)
mydata2 <- mydata[,!names(mydata) %in% c("Cust_Key")]
mydata2$Y <- as.factor(mydata2$Y)
fit$importance
####var reduction#####
vartoremove <- ncol(mydata2) - 20
library(rminer)
#####
for(i in 1:vartoremove){
rf <- fit(Y~.,data=mydata2,model = "randomForest", mtry = 10 ,ntree = 100)
varImportance <- Importance(rf,mydata2,method="sensg")
Z <- order(varImportance$imp,decreasing = FALSE)
IND <- Z[2]
var_to_remove <- names(mydata2[IND])
mydata2[IND] = NULL
print(i)
}
###########
library(smbinning)
as.data.frame(mydata2) -> inp
summary(inp)
attach(inp)
rm(result)
str(inp)
inp$target <- as.numeric(inp$Y) *1
table(inp$target)
ftable(inp$Y,inp$target)
inp$target <- inp$target -1
result= smbinning(df=inp, y="target", x="X37", p=0.0005)
result$ivtable
smbinning.plot(result,option="badrate",sub="test")
summary(inp)
result$ivtable
boxplot(inp$X2~inp$Y,horizontal=T, frame=F, col="red",main="Distribution")
###Sample
require(caTools)
inp$Y <- NULL
sample = sample.split(inp$target, SplitRatio = .7)
train = subset(inp, sample == TRUE)
test = subset(inp, sample == FALSE)
head(train)
nrow(train)
fit1 <- glm(train$target~.,data=train,family = binomial)
summary(rf)
prediction1 <- data.frame(actual = test$target, predicted = predict(fit1,test ,type="response") )
result= smbinning(df=prediction1, y="actual", x="predicted", p=0.005)
result$ivtable
smbinning.plot(result,option="badrate",sub="test")
tail(prediction1)
write.csv(prediction1 , "test_pred_logistic.csv")
predict_train <- data.frame(actual = train$target, predicted = predict(fit1,train ,type="response") )
write.csv(predict_train , "train_pred_logistic.csv")
result= smbinning(df=predict_train, y="actual", x="predicted", p=0.005)
result$ivtable
smbinning.plot(result,option="badrate",sub="train")
####random forest
rf <- fit(target~.,data=train,model = "randomForest", mtry = 10 ,ntree = 200)
prediction2 <- data.frame(actual = test$target, predicted = predict(rf,train))
result= smbinning(df=prediction2, y="actual", x="predicted", p=0.005)
result$ivtable
smbinning.plot(result,option="badrate",sub="train")
###########IV
library(devtools)
install_github("riv","tomasgreif")
library(woe)
##### K-fold Validation ########
library(caret)
cv_fold_count = 2
folds = createFolds(mydata2$Y,cv_fold_count,list=T);
smpl = folds[[i]];
g_train = mydata2[-smpl,!names(mydata2) %in% c("Y")];
g_test = mydata2[smpl,!names(mydata2) %in% c("Y")];
cost_train = mydata2[-smpl,"Y"];
cost_test = mydata2[smpl,"Y"];
rf <- randomForest(g_train,cost_train)
logit.data <- cbind(cost_train,g_train)
logit.fit <- glm(cost_train~.,data=logit.data,family = binomial)
prediction <- data.f
rame(actual = test$Y, predicted = predict(rf,test))

Resources