Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!
Get Started for Free

bowl

Backwards Outcome Weighted Learning.


Description

Function performs a single step of the bowl method. Multiple decision points can be analyzed by repeated calls, as is done for qLearn() and optimalClass().

Usage

bowl(
  ...,
  moPropen,
  data,
  reward,
  txName,
  regime,
  response,
  BOWLObj = NULL,
  lambdas = 2,
  cvFolds = 0L,
  kernel = "linear",
  kparam = NULL,
  fSet = NULL,
  surrogate = "hinge",
  verbose = 2L
)

Arguments

...

Used primarily to require named input. However, inputs for the optimization methods can be sent through the ellipsis. If surrogate is hinge, the optimization method is dfoptim::hjk(). For all other surrogates, stats::optim() is used.

moPropen

An object of class modelObj or modelObjSubset, which defines the model and R methods to be used to obtain parameter estimates and predictions for the propensity for tx. See ?moPropen for details.

data

A data frame of the covariates and tx histories.

reward

The response vector.

txName

A character object. The column header of data that corresponds to the tx covariate

regime

A formula object or a list of formula objects. The covariates to be included in the decision function/kernel. If a list is provided, this specifies that there is an underlying subset structure – fSet must then be defined. For subsets, the name of each element of the list must correspond to the name of a subset. If a regime is to be estimated using multiple subsets combined, each subset must be included in the name and separated by a comma (no spaces).

response

A numeric vector. The same as reward above. Allows for naming convention followed in most DynTxRegime methods.

BOWLObj

NULL or BOWL-class object returned from previous call to bowl(). If NULL, indicates that the function call is for the first STEP of the BOWL algorithm (i.e., the final decision point). If a BOWL-class object, assumed that the object was returned by the preceding step of the BOWL algorithm.

lambdas

A numeric object or a numeric vector object giving the penalty tuning parameter(s). If more than 1 is provided, the set of tuning parameter values to be considered in the cross-validation algorithm (note that cvFolds must be positive in this case).

cvFolds

If cross-validation is to be used to select the tuning parameters and/or kernel parameters, the number of folds.

kernel

A character object. Must be one of {'linear', 'poly', 'radial'}

kparam

A numeric object.
If kernel = linear, kparam is ignored.
If kernel = poly, kparam is the degree of the polynomial.
If kernel = radial, kparam is the inverse bandwidth of the kernel. If a vector of bandwidth parameters is given, cross-validation will be used to select the parameter (note that cvFolds must be positive in this case).

fSet

A function or NULL defining subset structure. See ?fSet for details.

surrogate

The surrogate 0-1 loss function. Must be one of {'logit', 'exp', 'hinge', 'sqhinge', 'huber'}.

verbose

An integer or logical. If 0, no screen prints are generated. If 1, screen prints are generated with the exception of optimization results obtained in iterative algorithm. If 2, all screen prints are generated.

Value

a BOWL-class object

References

Yingqi Zhao, Donglin Zeng, Eric B. Laber, Michael R. Kosorok (2015) New statistical learning methods for estimating optimal dynamic treatment regimes. Journal of the American Statistical Association, 110:510, 583–598.

See Also

Other statistical methods: earl(), iqLearn, optimalClass(), optimalSeq(), owl(), qLearn(), rwl()

Other weighted learning methods: earl(), owl(), rwl()

Other multiple decision point methods: iqLearn, optimalClass(), optimalSeq(), qLearn()

Examples

# Load and process data set
data(bmiData)

# define the negative 12 month change in BMI from baseline
y12 <- -100*(bmiData[,6L] - bmiData[,4L])/bmiData[,4L]

# define the negative 4 month change in BMI from baseline
y4 <- -100*(bmiData[,5L] - bmiData[,4L])/bmiData[,4L]

# reward for second stage
rewardSS <- y12 - y4

#### Second-stage regression

# Constant propensity model
moPropen <- buildModelObj(model = ~1,
                          solver.method = 'glm',
                          solver.args = list('family'='binomial'),
                          predict.method = 'predict.glm',
                          predict.args = list(type='response'))

fitSS <- bowl(moPropen = moPropen,
              data = bmiData, reward = rewardSS,  txName = 'A2', 
              regime = ~ parentBMI + month4BMI)

##Available methods

  # Coefficients of the propensity score regression
  coef(fitSS)

  # Description of method used to obtain object
  DTRstep(fitSS)

  # Estimated value of the optimal treatment regime for training set
  estimator(fitSS)

  # Value object returned by propensity score regression method
  fitObject(fitSS)

  # Summary of optimization routine
  optimObj(fitSS)

  # Estimated optimal treatment for training data
  optTx(fitSS)

  # Estimated optimal treatment for new data
  optTx(fitSS, bmiData)

  # Plots if defined by propensity regression method
  dev.new()
  par(mfrow = c(2,4))

  plot(fitSS)
  plot(fitSS, suppress = TRUE)

  # Value object returned by propensity score regression method
  propen(fitSS)

  # Parameter estimates for decision function
  regimeCoef(fitSS)

  # Show main results of method
  show(fitSS)

  # Show summary results of method
  summary(fitSS)
 
#### First-stage regression

# Constant propensity model
fitFS <- bowl(moPropen = moPropen,
              data = bmiData, reward = y4,  txName = 'A1', 
              regime = ~ gender + parentBMI,
              BOWLObj = fitSS, lambdas = c(0.5, 1.0), cvFolds = 4L)

##Available methods for fitFS are as shown above for fitSS

  # Results of the cross-validation
  cvInfo(fitFS)

DynTxRegime

Methods for Estimating Optimal Dynamic Treatment Regimes

v4.10
GPL-2
Authors
S. T. Holloway, E. B. Laber, K. A. Linn, B. Zhang, M. Davidian, and A. A. Tsiatis
Initial release
2022-06-05

We don't support your browser anymore

Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.