I am trying to use caret package to tune 'df' parameter of a gam model for my cohort analysis.
With the following data:
cohort = 1:60
age = 1:26
grid = data.frame(expand.grid(age = age, cohort = cohort))
size = data.frame(cohort = cohort, N = sample(100:150,length(cohort), replace = TRUE))
df = merge(grid, size, by = "cohort")
log_k = -3 + log(df$N) - 0.5*log(df$age) + df$cohort*(df$cohort-30)*(df$cohort-50)/20000 + runif(nrow(df),min = 0, max = 0.5)
df$conversion = rpois(nrow(df),exp(log_k))
Explanation of the data : Cohort number is the time of arrival of the potential customer. N is the number of potential customer that arrived at that time. Conversion is the number out of those potential customer that 'converted' (bought something). Age is the age (time spent from arrival) of the cohort when conversion took place. For a given cohort there are fewer conversions as age grows. This effect follows a power law.
But the total conversion rate of each cohort can also change slowly in time (cohort number). Thus I want a smoothing spline of the time variable in my model.
I can fit a gam model from package gam
library(gam)
fit = gam(conversion ~ log(N) + log(age) + s(cohort, df = 4), data = df, family = poisson)
fit
> Call:
> gam(formula = conversion ~ log(N) + log(age) + s(cohort, df = 4),
> family = poisson, data = df)
> Degrees of Freedom: 1559 total; 1553 Residual
> Residual Deviance: 1869.943
But if i try to train the model using the CARET package
library(caret)
fitControl = trainControl(verboseIter = TRUE)
fit.crt = train(conversion ~ log(N) + log(age) + s(cohort,df),
data = df, method = "gamSpline",
trControl = fitControl, tune.length = 3, family = poisson)
I get this error :
+ Resample01: df=1
model fit failed for Resample01: df=1 Error in as.matrix(x) : object 'N' not found
- Resample01: df=1
+ Resample01: df=2
model fit failed for Resample01: df=2 Error in as.matrix(x) : object 'N' not found .....
Please does anyone know what I'm doing wrong here?
Thanks
There are a two things wrong with your code.
The train function can be a bit tedious depending on the method you used (as you have noticed). In the case of method = "gamSpline", the train function adds a smooth term to every independent term in the formula. So it converts your variables to s(log(N), df), s(log(age) df) and to s(s(cohort, df), df).
Wait s(s(cohort, df), df) does not really makes sense. So you must change s(cohort, df) to cohort.
I am not sure why, but the train with method = "gamSpline" does not like it when you put functions (e.g. log) in the formula. I think this is due to the fact that this method already applies the s() functions to your variables. This problem can be solved by applying the log earlier to your variables. Such as df$N <- log(df$N) or logN <- log(df$N) and use logN as variable. And of course, do the same for age.
My guess is that you don't want this method to apply a smoothing term to all your independent variables based on the code you provided. I am not sure if this is possible and how to do it, if it is possible.
Hope this helps.
EDIT: If you want a more elegant solution than the one I provided at point 2, make sure to read the comment of #topepo. This suggestion also allows you to apply s() function to the variables you want if I understand it correctly.
Related
I have an R coding question.
This is my first time asking a question here, so apologies if I am unclear or do something wrong.
I am trying to use a Generalized Linear Mixed Model (GLMM) with Poisson error family to test for any significant effect on a count response variable by three separate dichotomous variables (AGE = ADULT or JUVENILE, SEX = MALE or FEMALE and MEDICATION = NEW or OLD) and an interaction between AGE and MEDICATION (AGE:MEDICATION).
There is some dependency in my data in that the data was collected from a total of 22 different sites (coded as SITE vector with 33 distinct levels), and the data was collected over a total of 21 different years (coded as YEAR vector with 21 distinct levels, and treated as a categorical variable). Unfortunately, every SITE was not sampled for each YEAR, with some being sampled for a greater number of years than others.
The data is also quite sparse, in that I do not have a great number of measurements of the response variable (coded as COUNT and an integer vector) per SITE per YEAR.
My Poisson GLMM is constructed using the following code:
model <- glmer(data = mydata,
family = poisson(link = "log"),
formula = COUNT ~ SEX + SEX:MEDICATION + AGE + AGE:SEX + MEDICATION + AGE:MEDICATION + (1|SITE/YEAR),
offset = log(COUNT.SAMPLE.SIZE),
nAGQ = 0)
In order to try and obtain more reliable estimates for the fixed effect coefficients (particularly given the sparse nature of my data), I am trying to obtain 95% confidence intervals for the fixed effect coefficients through non-parametric bootstrapping.
I have come across the "glmmboot" package which can be used to conduct non-parametric bootstrapping of GLMMs, however when I try to run the non-parametric bootstrapping using the following code:
library(glmmboot)
bootstrap_model(base_model = model,
base_data = mydata,
resamples = 1000)
When I run this code, I receive the following message:
Performing case resampling (no random effects)
Naturally, though, my model does have random effects, namely (1|SITE/YEAR).
If I try to tell the function to resample from a specific block, by adding in the "reample_specific_blocks" argument, i.e.:
library(glmmboot)
bootstrap_model(base_model = model,
base_data = mydata,
resamples = 1000,
resample_specific_blocks = "YEAR")
Then I get the following error message:
Performing block resampling, over SITE
Error: Invalid grouping factor specification, YEAR:SITE
I get a similar error message if I try set 'resample_specific_blocks' to "SITE".
If I then try to set 'resample_specific_blocks' to "SITE:YEAR" or "SITE/YEAR" I get the following error message:
Error in bootstrap_model(base_model = model, base_data = mydata, resamples = 1000, :
No random columns from formula found in resample_specific_blocks
I have tried explicitly nesting YEAR within SITE and then adapting the model accordingly using the code:
mydata <- within(mydata, SAMPLE <- factor(SITE:YEAR))
model.refit <- glmer(data = mydata,
family = poisson(link = "log"),
formula = COUNT ~ SEX + AGE + MEDICATION + AGE:MEDICATION + (1|SITE) + (1|SAMPLE),
offset = log(COUNT.SAMPLE.SIZE),
nAGQ = 0)
bootstrap_model(base_model = model.refit,
base_data = mydata,
resamples = 1000,
resample_specific_blocks = "SAMPLE")
But unfortunately I just get this error message:
Error: Invalid grouping factor specification, SITE
The same error message comes up if I set resample_specific_blocks argument to SITE, or if I just remove the resample_specific_blocks argument.
I believe that the case_bootstrap() function found in the lmeresampler package could potentially be another option, but when I look into the help for it it looks like I would need to create a function and I unfortunately have no experience with creating my own functions within R.
If anyone has any suggestions on how I can get the bootstrap_model() function in the glmmboot package to recognise the random effects in my model/dataframe, or any suggestions for alternative methods on conducting non-parametric bootstrapping to create 95% confidence intervals for the coefficients of the fixed effects in my model, it would be greatly appreciated! Many thanks in advance, and for reading such a lengthy question!
For reference, I include links to the RDocumentation and GitHub for the glmmboot package:
https://www.rdocumentation.org/packages/glmmboot/versions/0.6.0
https://github.com/ColmanHumphrey/glmmboot
The following is code that will allow for creation of a reproducible example using the data set from lme4::grouseticks
#Load in required packages
library(tidyverse)
library(lme4)
library(glmmboot)
library(psych)
#Load in the grouseticks dataframe
data("grouseticks")
tibble(grouseticks)
#Create dummy vectors for SEX, AGE and MEDICATION
set.seed(1)
SEX <-sample(1:2, size = 403, replace = TRUE)
SEX <- as.factor(ifelse(SEX == 1, "MALE", "FEMALE"))
set.seed(2)
AGE <- sample(1:2, size = 403, replace = TRUE)
AGE <- as.factor(ifelse(AGE == 1, "ADULT", "JUVENILE"))
set.seed(3)
MEDICATION <- sample(1:2, size = 403, replace = TRUE)
MEDICATION <- as.factor(ifelse(MEDICATION == 1, "OLD", "NEW"))
grouseticks$SEX <- SEX
grouseticks$AGE <- AGE
grouseticks$MEDICATION <- MEDICATION
#Use the INDEX vector to create a vector of sample sizes per LOCATION
#per YEAR
grouseticks$INDEX <- 1
sample.sizes <- grouseticks %>%
group_by(LOCATION, YEAR) %>%
summarise(SAMPLE.SIZE = sum(INDEX))
#Combine the dataframes together into the dataframe to be used in the
#model
mydata$SAMPLE.SIZE <- as.integer(mydata$SAMPLE.SIZE)
#Create the Poisson GLMM model
model <- glmer(data = mydata,
family = poisson(link = "log"),
formula = TICKS ~ SEX + SEX + AGE + MEDICATION + AGE:MEDICATION + (1|LOCATION/YEAR),
nAGQ = 0)
#Attempt non-parametric bootstrapping on the model to get 95%
#confidence intervals for the coefficients of the fixed effects
set.seed(1)
Model.bootstrap <- bootstrap_model(base_model = model,
base_data = mydata,
resamples = 1000)
Model.bootstrap
I would like to ask for help with my project. My goal is to get ROC curve from existing logistic regression.
First of all, here is what I'm analyzing.
glm.fit <- glm(Severity_Binary ~ Side + State + Timezone + Temperature.F. + Wind_Chill.F. + Humidity... + Pressure.in. + Visibility.mi. + Wind_Direction + Wind_Speed.mph. + Precipitation.in. + Amenity + Bump + Crossing + Give_Way + Junction + No_Exit + Railway + Station + Stop + Traffic_Calming + Traffic_Signal + Sunrise_Sunset , data = train_data, family = binomial)
glm.probs <- predict(glm.fit,type = "response")
glm.probs = predict(glm.fit, newdata = test_data, type = "response")
glm.pred = ifelse(glm.probs > 0.5, "1", "0")
This part works fine, I am able to show a table of prediction and mean result. But here comes the problem for me, I'm using pROC library, but I am open to use anything else which you can help me with. I'm using test_data with approximately 975 rows, but variable proc has only 3 sensitivities/specificities values.
library(pROC)
proc <- roc(test_data$Severity_Binary,glm.probs)
test_data$sens <- proc$sensitivities[1:975]
test_data$spec <- proc$specificities[1:975]
ggplot(test_data, aes(x=spec, y=sens)) + geom_line()
Here´s what I have as a result:
With Warning message:
Removed 972 row(s) containing missing values (geom_path).
As I found out, proc has only 3 values as I said.
You can't (and shouldn't) assign the sensitivity and specificity to the data. They are summary data and exist in a different dimension than your data.
Specifically, these two lines are wrong and make no sense at all:
test_data$sens <- proc$sensitivities[1:975]
test_data$spec <- proc$specificities[1:975]
Instead you must either save them to a new data.frame, or use some of the existing functions like ggroc:
ggroc(proc)
If you consider what the ROC curve does, there is no reason to expect it to have the same dimensions as your dataframe. It provides summary statistics of your model performance (sensitivity, specificity) evaluated on your dataset for different thresholds in your prediction.
Usually you would expect some more nuance on the curve (more than the 3 datapoints at thresholds -Inf, 0.5, Inf). You can look at the distribution of your glm.probs - this ROC curve indicates that all predictions are either 0 or 1, with very little inbetween (hence only one threshold at 0.5 on your curve). [This could also mean that you unintentially used your binary glm.pred for calculating the ROC curve, and not glm.probs as shown in the question (?)]
This seems to be more an issue with your model than with your code - here an example from a random different dataset, using the same steps you took (glm(..., family = binomial, predict(, type = "response"). This produces a ROC curve with 333 steps for ~1300 datapoints.
PS: (Ingore the fact that this is evaluated on training data, the point is the code looks alright up to the point of generating the ROC curve)
m1 <- glm(survived ~ passengerClass + sex + age, data = dftitanic, family = binomial)
myroc <- roc(dftitanic$survived,predict(m1, dftitanic, type = "response"))
plot(myroc)
I am trying to run the following code on R:
m <- gam(Flp_pop ~ s(Flp_CO, bs = "cr", k = 30), data = data, family = poisson, method = "REML")
My dataset is like this:
enter image description here
But when I try to execute, I get this error message:
"Error in if (abs(old.score - score) > score.scale * conv.tol) { :
missing value where TRUE/FALSE needed
In addition: There were 50 or more warnings (use warnings() to see the first 50)"
I am very new to R, maybe it is a very basic question. But does anyone know why this is happening?
Thanks!
The Poisson distribution has support on the non-negative integers and you are passing a continuous variable as the response. Here's an example with simulated data
library("mgcv")
library("gratia")
library("dplyr")
df <- data_sim("eg1", seed = 2) %>% # simulate Gaussian response
mutate(yabs = abs(y)) # make y non negative
mp <- gam(yabs ~ s(x2, bs = "cr"), data = df,
family = poisson, method = "REML")
# fails
which reproduces the error you saw
Error in if (abs(old.score - score) > score.scale * conv.tol) { :
missing value where TRUE/FALSE needed
In addition: There were 50 or more warnings (use warnings() to see the first 50)
The warnings are of the form:
$> warnings()[1]
Warning message:
In dpois(y, y, log = TRUE) : non-integer x = 7.384012
Indicating the problem; the model is evaluating the probability mass for your response data given the estimated model and you're evaluating this at the indicated non-integer value, which returns a 0 mass plus the warning.
If we'd passed the original Gaussian variable as the response, which includes negative values, the function would have errored out earlier:
mp <- gam(y ~ s(x2, bs = "cr"), data = df,
family = poisson, method = "REML")
which raises this error:
r$> mp <- gam(y ~ s(x2, bs = "cr"), data = df,
family = poisson, method = "REML")
Error in eval(family$initialize) :
negative values not allowed for the 'Poisson' family
An immediate but not necessarily advisable solution is just to use the quasipoisson family
mq <- gam(yabs ~ s(x2, bs = "cr"), data = df,
family = quasipoisson, method = "REML")
which uses the same mean variance relationship as the Poisson distribution but not the actual distribution so we can get away with abusing it.
Better would be to ask yourself why you were trying to fit a model that is ostensibly for counts to a response that is a continuous (non-negative) variable?
If the answer is you had a count but then normalised it in some way (say by dividing by some measure of effort like area surveyed or length of observation time) then you should use an offset of the form + offset(log(effort_var)) added to the model formula, and use the original non-normalised integer variable as the response.
If you really have a continuous response and the poisson was an over sight, try fitting with family = Gamma(link = "log")) or family = tw().
If it's something else, you should edit your question to include that info and perhaps we here can help or the question could be migrated to CrossValidated if the issue is more statistical in nature.
not sure where can I get help, since this exact post was considered off-topic on StackExchange.
I want to run some regressions based on a balanced panel with electoral data from Brazil focusing on 2 time periods. I want to understand if after a change in legislation that prohibited firm donations to candidates, those individuals that depended most on these resources had a lower probability of getting elected.
I have already ran a regression like this on R:
model_continuous <- plm(percentage_of_votes ~ time +
treatment + time*treatment, data = dataset, model = 'fd')
On this model I have used a continuous variable (% of votes) as my dependent variable. My treatment units or those that in time = 0 had no campaign contributions coming from corporations.
Now I want to change my dependent variable so that it is a binary variable indicating if the candidate was elected on that year. All of my units were elected on time = 0. How can I estimate a logit or probit model using fixed effects? I have tried using the pglm package in R.
model_binary <- pglm(dummy_elected ~ time + treatment + time*treatment,
data = dataset,
effects = 'twoways',
model = 'within',
family = 'binomial',
start = NULL)
However, I got this error:
Error in maxRoutine(fn = logLik, grad = grad, hess = hess, start = start, :
argument "start" is missing, with no default
Why is that happening? What is wrong with my model? Is it conceptually correct?
I want the second regression to be as similar as possible to the first one.
I have read that clogit function from the survival package could do the job, but I dont know how to do it.
Edit:
this is what a sample dataset could look like:
dataset <- data.frame(individual = c(1,1,2,2,3,3,4,4,5,5),
time = c(0,1,0,1,0,1,0,1,0,1),
treatment = c(0,0,1,1,0,0,1,1,0,0),
corporate = c(0,0,0.1,0,0,0,0.5,0,0,0))
Based on the comments, I believe the logistic regression reduces to treatment and dummy_elected. Accordingly I have fabricated the following dataset:
dataset <- data.frame("treatment" = c(rep(1,1000),rep(0,1000)),
"dummy_elected" = c(rep(1, 700), rep(0, 300), rep(1, 500), rep(0, 500)))
I then ran the GLM model:
library(MASS)
model_binary <- glm(dummy_elected ~ treatment, family = binomial(), data = dataset)
summary(model_binary)
Note that the treatment coefficient is significant and the coefficients are given. The resulting probabilities are thus
Probability(dummy_elected) = 1 => 1 / (1 + Exp(-(1.37674342264577E-16 + 0.847297860386033 * :treatment)))
Probability(dummy_elected) = 0 => 1 - 1 / (1 + Exp(-(1.37674342264577E-16 + 0.847297860386033 * :treatment)))
Note that these probabilities are consistent with the frequencies I generated the data.
So for each row, take the max probability across the two equations above and that's the value for dummy_elected.
I have built a survival cox-model, which includes a covariate * time interaction (non-proportionality detected).
I am now wondering how could I most easily get survival predictions from my model.
My model was specified:
coxph(formula = Surv(event_time_mod, event_indicator_mod) ~ Sex +
ageC + HHcat_alt + Main_Branch + Acute_seizure + TreatmentType_binary +
ICH + IVH_dummy + IVH_dummy:log(event_time_mod)
And now I was hoping to get a prediction using survfit and providing new.data for the combination of variables I am doing the predictions:
survfit(cox, new.data=new)
Now as I have event_time_mod in the right-hand side in my model I need to specify it in the new data frame passed on to survfit. This event_time would need to be set at individual times of the predictions. Is there an easy way to specify event_time_mod to be the correct time to survfit?
Or are there any other options for achieving predictions from my model?
Of course I could create as many rows in the new data frame as there are distinct times in the predictions and setting to event_time_mod to correct values but it feels really cumbersome and I thought that there must be a better way.
You have done what is refereed to as
An obvious but incorrect approach ...
as stated in Using Time Dependent Covariates and Time Dependent Coefficients in the Cox Model vignette in version 2.41-3 of the R survival package. Instead, you should use the time-transform functionality, i.e., the tt function as stated in the same vignette. The code would be something similar to the example in the vignette
> library(survival)
> vfit3 <- coxph(Surv(time, status) ~ trt + prior + karno + tt(karno),
+ data=veteran,
+ tt = function(x, t, ...) x * log(t+20))
>
> vfit3
Call:
coxph(formula = Surv(time, status) ~ trt + prior + karno + tt(karno),
data = veteran, tt = function(x, t, ...) x * log(t + 20))
coef exp(coef) se(coef) z p
trt 0.01648 1.01661 0.19071 0.09 0.9311
prior -0.00932 0.99073 0.02030 -0.46 0.6462
karno -0.12466 0.88279 0.02879 -4.33 1.5e-05
tt(karno) 0.02131 1.02154 0.00661 3.23 0.0013
Likelihood ratio test=53.8 on 4 df, p=5.7e-11
n= 137, number of events= 128
The survfit though does not work when you have a tt term
> survfit(vfit3, veteran[1, ])
Error in survfit.coxph(vfit3, veteran[1, ]) :
The survfit function can not yet process coxph models with a tt term
However, you can easily get out the terms, linear predictor or mean response with predict. Further, you can create the term over time for the tt term using the answer here.