I would like to fit a (very) high order regression to a set of data in R, however the poly() function has a limit of order 25.
For this application I need an order on the range of 100 to 120.
model <- lm(noisy.y ~ poly(q,50))
# Error in poly(q, 50) : 'degree' must be less than number of unique points
model <- lm(noisy.y ~ poly(q,30))
# Error in poly(q, 30) : 'degree' must be less than number of unique points
model <- lm(noisy.y ~ poly(q,25))
# OK
Polynomials and orthogonal polynomials
poly(x) has no hard-coded limit for degree. However, there are two numerical constraints in practice.
Basis functions are constructed on unique location of x values. A polynomial of degree k has k + 1 basis and coefficients. poly generates basis without the intercept term, so degree = k implies k basis and k coefficients. If there are n unique x values, it must be satisfied that k <= n, otherwise there is simply insufficient information to construct a polynomial. Inside poly(), the following line checks this condition:
if (degree >= length(unique(x)))
stop("'degree' must be less than number of unique points")
Correlation between x ^ k and x ^ (k+1) is getting closer and closer to 1 as k increases. Such approaching speed is of course dependent on x values. poly first generates ordinary polynomial basis, then performs QR factorization to find orthogonal span. If numerical rank-deficiency occurs between x ^ k and x ^ (k+1), poly will also stop and complain:
if (QR$rank < degree)
stop("'degree' must be less than number of unique points")
But the error message is not informative in this case. Furthermore, this does not have to be an error; it can be a warning then poly can reset degree to rank to proceed. Maybe R core can improve on this bit??
Your trial-and-error shows that you can't construct a polynomial of more than 25 degree. You can first check length(unique(q)). If you have a degree smaller than this but still triggering error, you know for sure it is due to numerical rank-deficiency.
But what I want to say is that a polynomial of more than 3-5 degree is never useful! The critical reason is the Runge's phenomenon. In terms of statistical terminology: a high-order polynomial always badly overfits data!. Don't naively think that because orthogonal polynomials are numerically more stable than raw polynomials, Runge's effect can be eliminated. No, a polynomial of degree k forms a vector space, so whatever basis you use for representation, they have the same span!
Splines: piecewise cubic polynomials and its use in regression
Polynomial regression is indeed helpful, but we often want piecewise polynomials. The most popular choice is cubic spline. Like that there are different representation for polynomials, there are plenty of representation for splines:
truncated power basis
natural cubic spline basis
B-spline basis
B-spline basis is the most numerically stable, as it has compact support. As a result, the covariance matrix X'X is banded, thus solving normal equations (X'X) b = (X'y) are very stable.
In R, we can use bs function from splines package (one of R base packages) to get B-spline basis. For bs(x), The only numerical constraint on degree of freedom df is that we can't have more basis than length(unique(x)).
I am not sure of what your data look like, but perhaps you can try
library(splines)
model <- lm(noisy.y ~ bs(q, df = 10))
Penalized smoothing / regression splines
Regression spline is still likely to overfit your data, if you keep increasing the degree of freedom. The best model seems to be about choosing the best degree of freedom.
A great approach is using penalized smoothing spline or penalized regression spline, so that model estimation and selection of degree of freedom (i.e., "smoothness") are integrated.
The smooth.spline function in stats package can do both. Unlike what its name seems to suggest, for most of time it is just fitting a penalized regression spline rather than smoothing spline. Read ?smooth.spline for more. For your data, you may try
fit <- smooth.spline(q, noisy.y)
(Note, smooth.spline has no formula interface.)
Additive penalized splines and Generalized Additive Models (GAM)
Once we have more than one covariates, we need additive models to overcome the "curse of dimensionality" while being sensible. Depending on representation of smooth functions, GAM can come in various forms. The most popular, in my opinion, is the mgcv package, recommended by R.
You can still fit a univariate penalized regression spline with mgcv:
library(mgcv)
fit <- gam(noisy.y ~ s(q, bs = "cr", k = 10))
Related
I am currently trying to generate a general additive model in R using a response variable and three predictor variables. One of the predictors is linear, and the dataset consists of 298 observations.
I have run the following code to generate a basic GAM:
GAM <- gam(response~ linearpredictor+ s(predictor2) + s(predictor3), data = data[2:5])
This produces a model with 18 degrees of freedom and seems to substantially overfit the data. I'm wondering how I might generate a GAM that maximizes smoothness and predictive error. I realize that each of these features is going to come at the expense of the other, but is there good a way to find the optimal model that doesn't overfit?
Additionally, I need to perform leave one out cross validation (LOOCV), and I am not sure how to make sure that gam() does this in the MGCV package. Any help on either of these problems uld be greatly appreciated. Thank you.
I've run this to generate a GAM, but it overfits the data.
GAM <- gam(response~ linearpredictor+ s(predictor2) + s(predictor3), data = data[2:5])
I have also generated 1,000,000 GAMs with varying combinations of smoothing parameters and ranged the maximum degrees of freedom allowed from 10 (as shown in the code below) to 19. The variable "combinations2" is a list of all 1,000,000 combinations of smoothers I selected. This code is designed to try and balance degrees of freedom and AIC score. It does function, but I'm not sure that I'm actually going to be able to find the optimal model from this. I also cannot tell how to make sure that it uses LOOCV.
BestGAM <- gam(response~ linearpredictor+ predictor2+ predictor3, data = data[2:5])
for(i in 1:100000){
PotentialGAM <- gam(response~ linearpredictor+ s(predictor2) + s(predictor3), data = data[2:5], sp=c(combinations2[i,]$Var1,combinations2[i,]$Var2))
if (AIC(PotentialGAM,BestGAM)$df[1] <= 10 & AIC(PotentialGAM,BestGAM)$AIC[1] < AIC(PotentialGAM,BestGAM)$AIC[2]){
BestGAM <<- PotentialGAM
listNumber <- i
}
}
You are fitting your GAM using generalised cross validation (GCV) smoothness selection. GCV is a way to get around the invariance problem of ordinary cross validation (OCV; what you also call LOOCV) when estimating GAMs. Note that GCV is the same as OCV on a rotated version of the fitting problem (rotating y - Xβ by Q, any orthogonal matrix), and while when fitting with GCV {mgcv} doesn't actually need to do the rotation and the expected GCV score isn't affected by the rotation, GCV is just OCV (wood 2017, p. 260)
It has been shown that GCV can undersmooth (resulting in more wiggly models) as the objective function (GCV profile) can become flat around the optimum. Instead it is preferred to estimate GAMs (with penalized smooths) using REML or ML smoothness selection; add method = "REML" (or "ML") to your gam() call.
If the REML or ML fit is as wiggly as the GCV one with your data, then I'd be likely to presume gam() is not overfitting, but that there is something about your response data that hasn't been explained here (are the data ordered in time, for example?)
As to your question
how I might generate a GAM that maximizes smoothness and [minimize?] predictive error,
you are already doing that using GCV smoothness selection and for a particular definition of "smoothness" (in this case it is squared second derivatives of the estimated smooths, integrated over the range of the covariates, and summed over smooths).
If you want GCV but smoother models, you can increase the gamma argument above 1; gamma 1.4 is often used for example, which means that each EDF costs 40% more in the GCV criterion.
FWIW, you can get the LOOCV (OCV) score for your model without actually fitting 288 GAMs through the use of the influence matrix A. Here's a reproducible example using my {gratia} package:
library("gratia")
library("mgcv")
df <- data_sim("eg1", seed = 1)
m <- gam(y ~ s(x0) + s(x1) + s(x2) + s(x3), data = df, method = "REML")
A <- influence(m)
r <- residuals(m, type = "response")
ocv_score <- mean(r^2 / (1 - A))
I am developing a COX regression model in R.
The model I am currently using is as follows
fh <- cph(S ~ rcs(MPV,4) + rcs(age,3) + BMI + smoking + hyperten + gender +
rcs(FVCPP,3) + TLcoPP, x=TRUE, y=TRUE, surv=TRUE, time.inc=2*52)
If I then want to look at this with
print(fh, latex = TRUE)
I get 3 coefs/SE/Wald etc for MPV (MVP, MVP' and MVP'') and 2 for age (age, age').
Could someone please explain to me what these outputs are? i.e. I believe they are to do with the restricted cubic splines I have added.
When you write rcs(MPV,4), you define the number of knots to use in the spline; in this case 4. Similarly, rcs(age,3) defines a spline with 3 knots. Due to identifiability constraints, 1 knot from each spline is subtracted out. You can think of this as defining an intercept for each spline. So rcs(Age,3) is a linear combination of 2 nonlinear basis functions and an intercept, while rcs(MPV,4) is a linear combination of 3 nonlinear basis functions and an intercept, i.e.,
and
In the notation above, what you get out from the print statement are the regression coefficients and , with corresponding standard errors, p-values etc. The intercepts and are typically set to zero, but they are important, because without them, the model fitting routine how have no idea of where on the y-axis to constrain the splines.
As a final note, you might actually be more interested in the output of summary(fh).
I am conducting an analysis of where on the landscape a predator encounters potential prey. My response data is binary with an Encounter location = 1 and a Random location = 0 and my independent variables are continuous but have been rescaled.
I originally used a GLM structure
glm_global <- glm(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs,
data=Data_scaled, family=binomial)
but realized that this failed to account for potential spatial-autocorrelation in the data (a spline correlogram showed high residual correlation up to ~1000m).
Correlog_glm_global <- spline.correlog (x = Data_scaled[, "Y"],
y = Data_scaled[, "X"],
z = residuals(glm_global,
type = "pearson"), xmax = 1000)
I attempted to account for this by implementing a GLMM (in lme4) with the predator group as the random effect.
glmm_global <- glmer(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs+(1|Group),
data=Data_scaled, family=binomial)
When comparing AIC of the global GLMM (1144.7) to the global GLM (1149.2) I get a Delta AIC value >2 which suggests that the GLMM fits the data better. However I am still getting essentially the same correlation in the residuals, as shown on the spline correlogram for the GLMM model).
Correlog_glmm_global <- spline.correlog (x = Data_scaled[, "Y"],
y = Data_scaled[, "X"],
z = residuals(glmm_global,
type = "pearson"), xmax = 10000)
I also tried explicitly including the Lat*Long of all the locations as an independent variable but results are the same.
After reading up on options, I tried running Generalized Estimating Equations (GEEs) in “geepack” thinking this would allow me more flexibility with regards to explicitly defining the correlation structure (as in GLS models for normally distributed response data) instead of being limited to compound symmetry (which is what we get with GLMM). However I realized that my data still demanded the use of compound symmetry (or “exchangeable” in geepack) since I didn’t have temporal sequence in the data. When I ran the global model
gee_global <- geeglm(Encounter ~ Dist_water_cs+coverMN_cs+I(coverMN_cs^2)+
Prey_bio_stand_cs+Prey_freq_stand_cs+Dist_centre_cs,
id=Pride, corstr="exchangeable", data=Data_scaled, family=binomial)
(using scaled or unscaled data made no difference so this is with scaled data for consistency)
suddenly none of my covariates were significant. However, being a novice with GEE modelling I don’t know a) if this is a valid approach for this data or b) whether this has even accounted for the residual autocorrelation that has been evident throughout.
I would be most appreciative for some constructive feedback as to 1) which direction to go once I realized that the GLMM model (with predator group as a random effect) still showed spatially autocorrelated Pearson residuals (up to ~1000m), 2) if indeed GEE models make sense at this point and 3) if I have missed something in my GEE modelling. Many thanks.
Taking the spatial autocorrelation into account in your model can be done is many ways. I will restrain my response to R main packages that deal with random effects.
First, you could go with the package nlme, and specify a correlation structure in your residuals (many are available : corGaus, corLin, CorSpher ...). You should try many of them and keep the best model. In this case the spatial autocorrelation in considered as continous and could be approximated by a global function.
Second, you could go with the package mgcv, and add a bivariate spline (spatial coordinates) to your model. This way, you could capture a spatial pattern and even map it. In a strict sens, this method doesn't take into account the spatial autocorrelation, but it may solve the problem. If the space is discret in your case, you could go with a random markov field smooth. This website is very helpfull to find some examples : https://www.fromthebottomoftheheap.net
Third, you could go with the package brms. This allows you to specify very complex models with other correlation structure in your residuals (CAR and SAR). The package use a bayesian approach.
I hope this help. Good luck
I want to determine the marginal effects of each dependent variable in a probit regression as follows:
predict the (base) probability with the mean of each variable
for each variable, predict the change in probability compared to the base probability if the variable takes the value of mean + 1x standard deviation of the variable
In one of my regressions, I have a multiplicative variable, as follows:
my_probit <- glm(a ~ b + c + I(b*c), family = binomial(link = "probit"), data=data)
Two questions:
When I determine the marginal effects using the approach above, will the value of the multiplicative term reflect the value of b or c taking the value mean + 1x standard deviation of the variable?
Same question, but with an interaction term (* and no I()) instead of a multiplicative term.
Many thanks
When interpreting the results of models involving interaction terms, the general rule is DO NOT interpret coefficients. The very presence of interactions means that the meaning of coefficients for terms will vary depending on the other variate values being used for prediction. The right way to go about looking at the results is to construct a "prediction grid", i.e. a set of values that are spaced across the range of interest (hopefully within the domain of data support). The two essential functions for this process are expand.grid and predict.
dgrid <- expand.grid(b=fivenum(data$b)[2:4], c=fivenum(data$c)[2:4]
# A grid with the upper and lower hinges and the medians for `a` and `b`.
predict(my_probit, newdata=dgrid)
You may want to have the predictions on a scale other than the default (which is to return the linear predictor), so perhaps this would be easier to interpret if it were:
predict(my_probit, newdata=dgrid, type ="response")
Be sure to read ?predict and ?predict.glm and work with some simple examples to make sure you are getting what you intended.
Predictions from models containing interactions (at least those involving 2 covariates) should be thought of as being surfaces or 2-d manifolds in three dimensions. (And for 3-covariate interactions as being iso-value envelopes.) The reason that non-interaction models can be decomposed into separate term "effects" is that the slopes of the planar prediction surfaces remain constant across all levels of input. Such is not the case with interactions, especially those with multiplicative and non-linear model structures. The graphical tools and insights that one picks up in a differential equations course can be productively applied here.
I have two types of individuals, say M and F, each described with six variables (forming a 6D space S). I would like to identify the regions in S where the densities of M and F differ maximally. I first tried a logistic binomial model linking F/ M to the six variables but the result of this GLM model is very hard to interpret (in part due to the numerous significant interaction terms). Thus I am thinking to an “spatial” analysis where I would separately estimate the density of M and F individuals everywhere in S, then calculating the difference in densities. Eventually I would manually look for the largest difference in densities, and extract the values at the 6 variables.
I found the function sm.density in the package sm that can estimate densities in a 3d space, but I find nothing for a space with n>3. Would you know something that would manage to do this in R? Alternatively, would have a more elegant method to answer my first question (2nd sentence)?
In advance,
Thanks a lot for your help
The function kde of the package ks performs kernel density estimation for multinomial data with dimensions ranging from 1 to 6.
pdfCluster and np packages propose functions to perform kernel density estimation in higher dimension.
If you prefer parametric techniques, you look at R packages doing gaussian mixture estimation like mclust or mixtools.
The ability to do this with GLM models may be constrained both by interpretablity issues that you already encountered as well as by numerical stability issues. Furthermore, you don't describe the GLM models, so it's not possible to see whether you include consideration of non-linearity. If you have lots of data, you might consider using 2D crossed spline terms. (These are not really density estimates.) If I were doing initial exploration with facilities in the rms/Hmisc packages in five dimensions it might look like:
library(rms)
dd <- datadist(dat)
options(datadist="dd")
big.mod <- lrm( MF ~ ( rcs(var1, 3) + # `lrm` is logistic regression in rms
rcs(var2, 3) +
rcs(var3, 3) +
rcs(var4, 3) +
rcs(var5, 3) )^2,# all 2way interactions
data=dat,
max.iter=50) # these fits may take longer times
bplot( Predict(bid.mod, var1,var2, n=10) )
That should show the simultaneous functional form of var1's and var2's contribution to the "5 dimensional" model estimates at 10 points each and at the median value of the three other variables.