FTRLProx_cv: FTRL-Proximal Linear Model Cross Validation

In yanyachen/rFTRLProximal: FTRL-Proximal Algorithm

Description

An advanced interface for FTRL-Proximal online learning model cross validation.

Usage

FTRLProx_cv(data, model = NULL, family = c("gaussian", "binomial",
  "poisson"), params = list(alpha = 0.1, beta = 1, l1 = 1, l2 = 1),
  epoch = 1, folds, eval = NULL, patience = 0, maximize = NULL,
  nthread = 1, verbose = TRUE)

Arguments

data	a object of class ftrl.Dataset.
model	a previously built model object to continue the training from.
family	link function to be used in the model. "gaussian", "binomial" and "poisson" are avaliable.
params	a list of parameters of FTRL-Proximal Algorithm. alpha alpha in the per-coordinate learning rate beta beta in the per-coordinate learning rate l1 L1 regularization parameter l2 L2 regularization parameter dropout percentage of the input features to drop from each sample
epoch	The number of iterations over training data to train the model.
folds	list provides a possibility of using a list of pre-defined CV folds (each element must be a vector of fold's indices).
eval	a custimized evaluation function, the first argument shoule be prediction, the second argument shoule be label.
patience	The number of rounds with no improvement in the evaluation metric in order to stop the training. User can specify 0 to disable early stopping.
maximize	whether to maximize the evaluation metric.
nthread	number of parallel threads used to run ftrl. Please set to 1 if your feature set is not sparse enough.
verbose	logical value. Indicating if the progress bar is displayed or not.

Value

a list with the following elements is returned:

• dt a data.table with each mean and standard deviation stat for training set and test set

• pred a numerical vector with predictions for each CV-fold for the model having been trained on the data in all other folds.

References

H. B. McMahan, G. Holt, D. Sculley, et al. "Ad click prediction: a view from the trenches". In: _The 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2013, Chicago, IL, USA, August 11-14, 2013_. Ed. by I. S.Dhillon, Y. Koren, R. Ghani, T. E. Senator, P. Bradley, R. Parekh, J. He, R. L. Grossman and R. Uthurusamy. ACM, 2013, pp. 1222-1230. DOI: 10.1145/2487575.2488200. <URL: http://doi.acm.org/10.1145/2487575.2488200>.

Examples

library(FeatureHashing)
library(data.table)
library(rBayesianOptimization)
library(MLmetrics)
data(ipinyou)
m.train <- FTRLProx_Hashing(~ 0 + ., ipinyou.train[, -"IsClick", with = FALSE],
                            hash.size = 2^13, signed.hash = FALSE, verbose = TRUE,
                            label = as.numeric(ipinyou.train$IsClick))
ftrl_model_cv <- FTRLProx_cv(data = m.train, model = NULL,
                             family = "binomial",
                             params = list(alpha = 0.01, beta = 0.1,
                                           l1 = 1.0, l2 = 1.0, dropout = 0), epoch = 50,
                             folds = KFold(as.numeric(ipinyou.train$IsClick), nfolds = 5,
                                           stratified = FALSE, seed = 0),
                             eval = AUC, patience = 5, maximize = TRUE,
                             nthread = 1, verbose = TRUE)

yanyachen/rFTRLProximal documentation built on May 4, 2019, 2:30 p.m.