Family: Gradient Boosting Families

View source: R/family.R

FamilyR Documentation

Gradient Boosting Families

Description

boost_family objects provide a convenient way to specify loss functions and corresponding risk functions to be optimized by one of the boosting algorithms implemented in this package.

Usage

Family(ngradient, loss = NULL, risk = NULL,
       offset = function(y, w)
           optimize(risk, interval = range(y),
                    y = y, w = w)$minimum,
       check_y = function(y) y,
       weights = c("any", "none", "zeroone", "case"),
       nuisance = function() return(NA),
       name = "user-specified", fW = NULL,
       response = function(f) NA,
       rclass = function(f) NA)
AdaExp()
AUC()
Binomial(type = c("adaboost", "glm"),
         link = c("logit", "probit", "cloglog", "cauchit", "log"), ...)
GaussClass()
GaussReg()
Gaussian()
Huber(d = NULL)
Laplace()
Poisson()
GammaReg(nuirange = c(0, 100))
CoxPH()
QuantReg(tau = 0.5, qoffset = 0.5)
ExpectReg(tau = 0.5)
NBinomial(nuirange = c(0, 100))
PropOdds(nuirange = c(-0.5, -1), offrange = c(-5, 5))
Weibull(nuirange = c(0, 100))
Loglog(nuirange = c(0, 100))
Lognormal(nuirange = c(0, 100))
Gehan()
Hurdle(nuirange = c(0, 100))
Multinomial()
Cindex(sigma = 0.1, ipcw = 1)
RCG(nuirange = c(0, 1), offrange = c(-5, 5))

Arguments

ngradient

a function with arguments y, f and w implementing the negative gradient of the loss function (which is to be minimized).

loss

an optional loss function with arguments y and f.

risk

an optional risk function with arguments y, f and w to be minimized (!), the weighted mean of the loss function by default.

offset

a function with argument y and w (weights) for computing a scalar offset.

fW

transformation of the fit for the diagonal weights matrix for an approximation of the boosting hat matrix for loss functions other than squared error.

response

inverse link function of a GLM or any other transformation on the scale of the response.

rclass

function to derive class predictions from conditional class probabilities (for models with factor response variable).

check_y

a function for checking and transforming the class / mode of a response variable.

nuisance

a function for extracting nuisance parameters from the family.

weights

a character indicating what type of weights are allowed. These can be either arbitrary (non-negative) weights code"any", only zero and one weights "zeroone", (non-negative) interger weights "case", or no weights are allowed "none".

name

a character giving the name of the loss function for pretty printing.

type

which parameterization of Binomial shoule be used?

b

link

link function. For possible values see Usage section.

d

delta parameter for Huber loss function. If omitted, it is chosen adaptively.

tau

the quantile or expectile to be estimated, a number strictly between 0 and 1.

qoffset

quantile of response distribution to be used as offset, i.e., starting values for the intercept. Per default the median of the response is used, which is in general a good choice (see Fenske et al. 2011, for details).

nuirange

a vector containing the end-points of the interval to be searched for the minimum risk w.r.t. the nuisance parameter. In case of PropOdds, the starting values for the nuisance parameters.

offrange

interval to search in for offset.

sigma

smoothness parameter for sigmoid functions inside Cindex.

ipcw

vector containing inverse probability of censoring weights for all observations. If omitted, it is estimated inside Cindex family.

...

additional arguments to link functions.

Details

The boosting algorithm implemented in mboost minimizes the (weighted) empirical risk function risk(y, f, w) with respect to f. By default, the risk function is the weighted sum of the loss function loss(y, f) but can be chosen arbitrarily. The ngradient(y, f) function is the negative gradient of loss(y, f) with respect to f.

Pre-fabricated functions for the most commonly used loss functions are available as well. Buehlmann and Hothorn (2007) give a detailed overview of the available loss functions. An updated overview can be found in Hofner et al (2014).

The offset function returns the population minimizers evaluated at the response, i.e., 1/2 \log(p / (1 - p)) for Binomial() or AdaExp() and (\sum w_i)^{-1} \sum w_i y_i for Gaussian() and the median for Huber() and Laplace(). The offset is used as starting value for the boosting algorithm.

Note that all families are functions and thus need to be specified either with empty brackets (e.g., family = Gaussian() for Gaussian regression) or with additional arguments if these are supported by the respective family (e.g., family = QuantReg(tau = 0.2) for quantile regression for the 20% quantile).

A short summary of the available families is given in the following paragraphs:

AdaExp(), Binomial() and AUC() implement families for binary classification. AdaExp() uses the exponential loss, which essentially leads to the AdaBoost algorithm of Freund and Schapire (1996). Binomial() implements the negative binomial log-likelihood of a logistic regression model as loss function. Thus, using Binomial family closely corresponds to fitting a logistic model. Alternative link functions can be specified.

However, the coefficients resulting from boosting with family Binomial(link = "logit") are 1/2 of the coefficients of a logit model obtained via glm. Buehlmann and Hothorn (2007) argue that the family Binomial is the preferred choice for binary classification. For binary classification problems the response y has to be a factor. Internally y is re-coded to -1 and +1 (Buehlmann and Hothorn 2007).

Binomial(type = "glm") is an alternative to Binomial() leading to coefficients of the same size as coefficients from a classical logit model via glm. Additionally, it works not only with a two-level factor but also with a two-column matrix containing the number of successes and number of failures (again, similar to glm).

AUC() uses 1-AUC(y, f) as the loss function. The area under the ROC curve (AUC) is defined as AUC = (n_{-1} n_1)^{-1} \sum_{i: y_i = 1} \sum_{j: y_j = -1} I(f_i > f_j). Since this is not differentiable in f, we approximate the jump function I((f_i - f_j) > 0) by the distribution function of the triangular distribution on [-1, 1] with mean 0, similar to the logistic distribution approximation used in Ma and Huang (2005).

Gaussian() is the default family in mboost. It implements L_2Boosting for continuous response. Note that families GaussReg() and GaussClass() (for regression and classification) are deprecated now. Huber() implements a robust version for boosting with continuous response, where the Huber-loss is used. Laplace() implements another strategy for continuous outcomes and uses the L_1-loss instead of the L_2-loss as used by Gaussian().

Poisson() implements a family for fitting count data with boosting methods. The implemented loss function is the negative Poisson log-likelihood. Note that the natural link function \log(\mu) = \eta is assumed. The default step-site nu = 0.1 is probably too large for this family (leading to infinite residuals) and smaller values are more appropriate.

GammaReg() implements a family for fitting nonnegative response variables. The implemented loss function is the negative Gamma log-likelihood with logarithmic link function (instead of the natural link).

CoxPH() implements the negative partial log-likelihood for Cox models. Hence, survival models can be boosted using this family.

QuantReg() implements boosting for quantile regression, which is introduced in Fenske et al. (2009). ExpectReg works in analogy, only for expectiles, which were introduced to regression by Newey and Powell (1987).

Families with an additional scale parameter can be used for fitting models as well: PropOdds() leads to proportional odds models for ordinal outcome variables (Schmid et al., 2011). When using this family, an ordered set of threshold parameters is re-estimated in each boosting iteration. An example is given below which also shows how to obtain the thresholds. NBinomial() leads to regression models with a negative binomial conditional distribution of the response. Weibull(), Loglog(), and Lognormal() implement the negative log-likelihood functions of accelerated failure time models with Weibull, log-logistic, and lognormal distributed outcomes, respectively. Hence, parametric survival models can be boosted using these families. For details see Schmid and Hothorn (2008) and Schmid et al. (2010).

Gehan() implements rank-based estimation of survival data in an accelerated failure time model. The loss function is defined as the sum of the pairwise absolute differences of residuals. The response needs to be defined as Surv(y, delta), where y is the observed survial time (subject to censoring) and delta is the non-censoring indicator (see Surv for details). For details on Gehan() see Johnson and Long (2011).

Cindex() optimizes the concordance-index for survival data (often denoted as Harrell's C or C-index). The concordance index evaluates the rank-based concordance probability between the model and the outcome. The C-index measures whether large values of the model are associated with short survival times and vice versa. The interpretation is similar to the AUC: A C-index of 1 represents a perfect discrimination while a C-index of 0.5 will be achieved by a completely non-informative marker. The Cindex() family is based on an estimator by Uno et al. (2011), which incorporates inverse probability of censoring weighting ipcw. To make the estimator differentiable, sigmoid functions are applied; the corresponding smoothness can be controlled via sigma. For details on Cindex() see Mayr and Schmid (2014).

Hurdle models for zero-inflated count data can be fitted by using a combination of the Binomial() and Hurdle() families. While the Binomial() family allows for fitting the zero-generating process of the Hurdle model, Hurdle() fits a negative binomial regression model to the non-zero counts. Note that the specification of the Hurdle model allows for using Binomial() and Hurdle() independently of each other.

Linear or additive multinomial logit models can be fitted using Multinomial(); although is family requires some extra effort for model specification (see example). More specifically, the predictor must be in the form of a linear array model (see %O%). Note that this family does not work with tree-based base-learners at the moment. The class corresponding to the last level of the factor coding of the response is used as reference class.

RCG() implements the ratio of correlated gammas (RCG) model proposed by Weinhold et al. (2016).

Value

An object of class boost_family.

Warning

The coefficients resulting from boosting with family Binomial(link = "logit") are 1/2 of the coefficients of a logit model obtained via glm (see above).

For AUC(), variables should be centered and scaled and observations with weight > 0 must not contain missing values. The estimated coefficients for AUC() have no probabilistic interpretation.

Author(s)

ExpectReg() was donated by Fabian Sobotka. AUC() was donated by Fabian Scheipl.

References

Peter Buehlmann and Torsten Hothorn (2007), Boosting algorithms: regularization, prediction and model fitting. Statistical Science, 22(4), 477–505.

Nora Fenske, Thomas Kneib, and Torsten Hothorn (2011), Identifying risk factors for severe childhood malnutrition by boosting additive quantile regression. Journal of the American Statistical Association, 106:494-510.

Yoav Freund and Robert E. Schapire (1996), Experiments with a new boosting algorithm. In Machine Learning: Proc. Thirteenth International Conference, 148–156.

Shuangge Ma and Jian Huang (2005), Regularized ROC method for disease classification and biomarker selection with microarray data. Bioinformatics, 21(24), 4356–4362.

Andreas Mayr and Matthias Schmid (2014). Boosting the concordance index for survival data – a unified framework to derive and evaluate biomarker combination. PloS ONE, 9(1):84483.

Whitney K. Newey and James L. Powell (1987), Asymmetric least squares estimation and testing. Econometrika, 55, 819–847.

Matthias Schmid and Torsten Hothorn (2008), Flexible boosting of accelerated failure time models. BMC Bioinformatics, 9(269).

Matthias Schmid, Sergej Potapov, Annette Pfahlberg, and Torsten Hothorn (2010). Estimation and regularization techniques for regression models with multidimensional prediction functions. Statistics and Computing, 20, 139–150.

Schmid, M., T. Hothorn, K. O. Maloney, D. E. Weller and S. Potapov (2011): Geoadditive regression modeling of stream biological condition. Environmental and Ecological Statistics, 18(4), 709–733.

Uno H, Cai T, Pencina MJ, D Agostino RB and Wei LJ (2011). On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data. Statistics in Medicine, 30(10), 1105–17.

Benjamin Hofner, Andreas Mayr, Nikolay Robinzonov and Matthias Schmid (2014). Model-based Boosting in R: A Hands-on Tutorial Using the R Package mboost. Computational Statistics, 29, 3–35.
\Sexpr[results=rd]{tools:::Rd_expr_doi("10.1007/s00180-012-0382-5")}

Available as vignette via: vignette(package = "mboost", "mboost_tutorial")

Brent A. Johnson and Qi Long (2011) Survival ensembles by the sum of pairwise differences with application to lung cancer microarray studies. Annals of Applied Statistics, 5, 1081–1101.

Weinhold, L., S. Pechlivanis, S. Wahl, P. Hoffmann and M. Schmid (2016) A Statistical Model for the Analysis of Bounded Response Variables in DNA Methylation Studies. BMC Bioinformatics. 2016; 17: 480. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1186/s12859-016-1347-4")}

See Also

mboost for the usage of Familys. See boost_family-class for objects resulting from a call to Family.

Examples

### Define a new family
MyGaussian <- function(){
       Family(ngradient = function(y, f, w = 1) y - f,
       loss = function(y, f) (y - f)^2,
       name = "My Gauss Variant")
}
# Now use the new family
data(bodyfat, package = "TH.data")
mod <- mboost(DEXfat ~ ., data = bodyfat, family = MyGaussian())
# N.B. that the family needs to be called with empty brackets


### Proportional odds model
data(iris)
iris$Species <- factor(iris$Species, ordered = TRUE)
if (require("MASS")) {
    (mod.polr <- polr(Species  ~ Sepal.Length, data = iris))
}
mod.PropOdds <- glmboost(Species  ~ Sepal.Length, data = iris,
                         family = PropOdds(nuirange = c(-0.5, 3)))
mstop(mod.PropOdds) <- 1000
## thresholds are treated as nuisance parameters, to extract these use
nuisance(mod.PropOdds)
## effect estimate
coef(mod.PropOdds)["Sepal.Length"]
## make thresholds comparable to a model without intercept
nuisance(mod.PropOdds) - coef(mod.PropOdds)["(Intercept)"] -
    attr(coef(mod.PropOdds), "offset")

### Multinomial logit model via a linear array model
## One needs to convert the data to a list
myiris <- as.list(iris)
## ... and define a dummy vector with one factor level less
## than the outcome, which is used as reference category.
myiris$class <- factor(levels(iris$Species)[-nlevels(iris$Species)])
## Now fit the linear array model
mlm <- mboost(Species ~ bols(Sepal.Length, df = 2) %O%
                        bols(class, df = 2, contrasts.arg = "contr.dummy"),
              data = myiris,
              family = Multinomial())
coef(mlm) ## one should use more boosting iterations.
head(round(pred <- predict(mlm, type = "response"), 2))

## Prediction with new data:
newdata <- as.list(iris[1,])
## One always needs to keep the dummy vector class as above!
newdata$class <- factor(levels(iris$Species)[-nlevels(iris$Species)])
pred2 <- predict(mlm, type = "response", newdata = newdata)
## check results
pred[1, ]
pred2

## Not run: ############################################################
## Do not run and check these examples automatically as
## they take some time

## Compare results with nnet::multinom
if (require("nnet")) {
    mlmn <- multinom(Species ~ Sepal.Length, data = iris)
    max(abs(fitted(mlm[1000], type = "response") -
            fitted(mlmn, type = "prob")))
}

## End(Not run and test)

## End(Not run)


### Example for RCG model
## generate covariate values
set.seed(12345)
x1 <- rnorm(500)
x2 <- rnorm(500)
## generate linear predictors
zetaM <- 0.1 + 0.3 * x1 - 0.5 * x2 
zetaU <- 0.1 - 0.1 * x1 + 0.2 * x2
## generate beta values
M <- rgamma(500, shape = 2, rate = exp(zetaM))
U <- rgamma(500, shape = 2, rate = exp(zetaU))
y <- M / (M + U)

## fit RCG model
data <- data.frame(y, x1, x2)
RCGmodel <- glmboost(y ~ x1 + x2, data = data, family = RCG(),
                     control = boost_control(mstop = 1000,
                     trace = TRUE, nu = 0.01))
## true coefficients: gamma = (0.0, 0.4, -0.7),
##                    alpha (= shape) = 2,
##                    rho = 0
## compare to coefficient estimates
coef(RCGmodel)
nuisance(RCGmodel)

## compute downstream tests 
## (only suitable without early stopping, i.e., if likelihood based model converged)
downstream.test(RCGmodel)

## compute conditional expectations
predictions <- predict(RCGmodel, type = "response")
plot(predictions, y)
abline(0,1)


mboost documentation built on Sept. 8, 2023, 6:15 p.m.