gridisl

Machine Learning Toolkit with Grid-Search Discrete SuperLearner for Longitudinal Data. Provides access to machine learning algorithms implemented in xgboost or h2o (RandomForests, Gradient Boosting Machines, Deep Neural Nets). Simple syntax for specifying large grids of tuning parameters, including random grid search over parameter space. Model selection can be performed via V-fold cross-validation or random validation splits.

Installation

To install the development version (requires the devtools package):

devtools::install_github('osofr/gridisl', build_vignettes = FALSE)

Perform Simultaneuous Model Selection with xgboost and h2o

Initialize h2o cluster:

  require("h2o")
  h2o::h2o.init(nthreads = -1)
  options(gridisl.verbose = TRUE)
  # options(gridisl.verbose = FALSE)
  data(cpp)
  cpp <- cpp[!is.na(cpp[, "haz"]), ]
  covars <- c("apgar1", "apgar5", "parity", "gagebrth", "mage", "meducyrs", "sexn")

Specify the models:

## Single GBM w/ h2o vs. xgboost with (roughly) equivalent parameter settings as evaluated by holdout MSE & CV-MSE

GRIDparams <- 
    defModel(estimator = "h2o__gbm", family = "gaussian",
               ntrees = 500,
               learn_rate = 0.01,
               max_depth = 5,
               min_rows = 10, # [default=10]
               col_sample_rate_per_tree = 0.3,  # [default=1]
               stopping_rounds = 10, stopping_metric = "MSE", score_each_iteration = TRUE, score_tree_interval = 1,
               seed = 23) +

    defModel(estimator = "xgboost__gbm", family = "gaussian",
               nrounds = 500,
               learning_rate = 0.01, 
               max_depth = 5,
               min_child_weight = 10,
               colsample_bytree = 0.3,
               alpha = 0.5
               early_stopping_rounds = 10,
               seed = 23)

Fit all models at once and evaluate perfomance based on random holdout observations:

## SuperLearner with random holdout:
cpp_holdout <- add_holdout_ind(data = cpp, ID = "subjid", hold_column = "hold", random = TRUE, seed = 12345)
mfit_hold <- fit(GRIDparams, ID = "subjid", t_name = "agedays", x = c("agedays", covars), y = "haz",
                data = cpp_holdout, method = "holdout", hold_column = "hold")

Fit all models at once and evaluate perfomance based on V-fold cross-validation:

## SuperLearner with CV:
cpp_folds <- add_CVfolds_ind(cpp, ID = "subjid", nfolds = 5, seed = 23)
mfit_cv <- fit(GRIDparams, ID = "subjid", t_name = "agedays", x = c("agedays", covars), y = "haz",
                data = cpp_folds, method = "cv", fold_column = "fold")

Specifying large ensembles of models with grids of hyper-parameters:

  GRIDparams <- 
    defModel(estimator = "h2o__gbm", family = "gaussian",
              ntrees = 500,
              param_grid = list(
                learn_rate = c(0.01, 0.02, 0.5, 0.3),
                max_depth = 5,
                sample_rate = c(0.3, 0.5, 0.8, 0.9, 1),
                col_sample_rate_per_tree = c(0.3, 0.4, 0.5, 0.7, 0.9, 1.0)
              ),
              stopping_rounds = 10, stopping_metric = "MSE", score_each_iteration = TRUE, score_tree_interval = 1,
              seed = 23) +
    defModel(estimator = "xgboost__gbm", family = "gaussian",
              nrounds = 500,
              param_grid = list(
                eta = c(0.01, 0.02, 0.5, 0.3),
                max_depth = 5,
                max_delta_step = c(0,1),
                subsample = c(0.3, 0.5, 0.8, 0.9, 1),
                colsample_bytree = c(0.3, 0.4, 0.5, 0.7, 0.9, 1.0)
                ),
              early_stopping_rounds = 50,
              seed = 23)

## SL with random holdout:
cpp_holdout <- add_holdout_ind(data = cpp, ID = "subjid", hold_column = "hold", random = TRUE, seed = 12345)
mfit_hold <- fit(GRIDparams, ID = "subjid", t_name = "agedays", x = c("agedays", covars), y = "haz",
                data = cpp_holdout, method = "holdout", hold_column = "hold")

## SL with CV:
cpp_folds <- add_CVfolds_ind(cpp, ID = "subjid", nfolds = 5, seed = 23)
mfit_cv <- fit(GRIDparams, ID = "subjid", t_name = "agedays", x = c("agedays", covars), y = "haz",
                data = cpp_folds, method = "cv", fold_column = "fold")

Copyright

The contents of this repository are distributed under the MIT license.

The MIT License (MIT)

Copyright (c) 2016-2017 Oleg Sofrygin 

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.