funKerasTransferLearning: funKerasTransferLearning

Description

Hyperparameter Tuning: Keras TransfewrLearning Test Function.

Usage

funKerasTransferLearning(x, kerasConf = getKerasConf(), data = NULL)

Value

1-column matrix with resulting function values (test loss).

Arguments

x: matrix of hyperparameter values to evaluate with the function. Rows for points and columns for dimension.
kerasConf: List of additional parameters passed to keras as described in getKerasConf. Default: kerasConf = getKerasConf().
data: data

Details

Trains a simple deep NN on the MNIST dataset. Provides a template that can be used for other networks as well. Standard Code from https://tensorflow.rstudio.com/ Modified by T. Bartz-Beielstein (tbb@bartzundbartz.de)

Examples

Run this code

# \donttest{
### These examples require an activated Python environment as described in
### Bartz-Beielstein, T., Rehbach, F., Sen, A., and Zaefferer, M.:
### Surrogate Model Based Hyperparameter Tuning for Deep Learning with SPOT,
### June 2021. http://arxiv.org/abs/2105.14625.

PYTHON_RETICULATE <- FALSE
if(PYTHON_RETICULATE){
library("SPOTMisc")
library("SPOT")
kerasConf <- getKerasConf()

 # Hyperparameters:
 # "dropout" =  x[1],
 # "learning_rate" =  x[2],
 # "epochs" = x[3],
 # "beta_1" =  x[4],
 # "beta_2" =  x[5],
 # "epsilon" = x[6],
 # "optimizer" = x[7]

lower <- c(1e-6, 1e-6, 2, 0.8, 0.8, 1e-9, 1)
upper <- c(0.2, 1e-2, 5, 0.99, 0.9999, 1e-3, 2)
types <- c("numeric",  "numeric",  "integer",  "numeric",  "numeric",
           "integer",  "factor")

## First Example: spot call with extended verbosity. Default objective
## "validationLoss", i.e., validation loss, is used. only 20 function
## evaluations (for testing).
kerasConf$verbose <- 1
res <- spot(x = NULL,
            fun = funKerasTransferLearning,
            lower = lower,
            upper = upper,
            control = list(funEvals=20,
                           model=buildKriging,
                           noise = TRUE,
                           types = types,
                           optimizer=optimDE,
                           plots = TRUE,
                           progress = TRUE,
                           seedFun = 1,
                           seedSPOT = 1,
                           kerasConf = kerasConf)
                           )
 save(res, file = paste0("resKerasTransferLearning", as.numeric(Sys.time()),".RData"))


 ## Example: resKerasTransferLearning04
 ## Default objective function "validationLoss", i.e.,
 ## training loss
library("SPOTMisc")
library("SPOT")
kerasConf <- getKerasConf()

# Hyperparameters:
# "dropout" =  x[1],
# "learning_rate" =  x[2],
# "epochs" = x[3],
# "beta_1" =  x[4],
# "beta_2" =  x[5],
# "epsilon" = x[6],
# "optimizer" = x[7]

lower <- c(1e-6, 1e-6, 2, 0.8, 0.8, 1e-9, 1)
upper <- c(0.2, 1e-2, 5, 0.99, 0.9999, 1e-3, 2)
types <- c("numeric",  "numeric",  "integer",  "numeric",  "numeric",
           "integer",  "factor")

res <- spot(x = NULL,
            fun = funKerasTransferLearning,
            lower = lower,
            upper = upper,
            control = list(funEvals=100,
                           model=buildKriging,
                           noise = TRUE,
                           types = types,
                           optimizer=optimDE,
                           plots = FALSE,
                           progress = TRUE,
                           seedFun = 1,
                           seedSPOT = 1,
                           kerasConf = kerasConf))
save(res,file = paste0("resKerasTransferLearningValidationLoss04",
as.numeric(Sys.time()),".RData"))



 ## Example: resKerasTransferLearning05
 ## objective function "negValidationAccuracy", i.e.,
 ## negative validation accuracy
library("SPOTMisc")
library("SPOT")
kerasConf <- getKerasConf()

# Hyperparameters:
# "dropout" =  x[1],
# "learning_rate" =  x[2],
# "epochs" = x[3],
# "beta_1" =  x[4],
# "beta_2" =  x[5],
# "epsilon" = x[6],
# "optimizer" = x[7]

lower <- c(1e-6, 1e-6, 2, 0.8, 0.8, 1e-9, 1)
upper <- c(0.2, 1e-2, 5, 0.99, 0.9999, 1e-3, 2)
types <- c("numeric",  "numeric",  "integer",  "numeric",  "numeric",
           "integer",  "factor")

kerasConf$returnValue <- "negValidationAccuracy"
res <- spot(x = NULL,
            fun = funKerasTransferLearning,
            lower = lower,
            upper = upper,
            control = list(funEvals=100,
                           model=buildKriging,
                           noise = TRUE,
                           types = types,
                           optimizer=optimDE,
                           plots = FALSE,
                           progress = TRUE,
                           seedFun = 1,
                           seedSPOT = 1,
                           kerasConf = kerasConf))
save(res,file = paste0("resKerasTransferLearningNegValidationAccuracy05",
as.numeric(Sys.time()),".RData"))


 ## Example: resKerasTransferLearning06
 ## objective function "trainingLoss", i.e.,
 ## training loss

library("SPOTMisc")
library("SPOT")
kerasConf <- getKerasConf()

# Hyperparameters:
# "dropout" =  x[1],
# "learning_rate" =  x[2],
# "epochs" = x[3],
# "beta_1" =  x[4],
# "beta_2" =  x[5],
# "epsilon" = x[6],
# "optimizer" = x[7]

lower <- c(1e-6, 1e-6, 2, 0.8, 0.8, 1e-9, 1)
upper <- c(0.2, 1e-2, 5, 0.99, 0.9999, 1e-3, 2)
types <- c("numeric",  "numeric",  "integer",  "numeric",  "numeric",
           "integer",  "factor")

kerasConf$returnValue <- "trainingLoss"
res <- spot(x = NULL,
            fun = funKerasTransferLearning,
            lower = lower,
            upper = upper,
            control = list(funEvals=100,
                           model=buildKriging,
                           noise = TRUE,
                           types = types,
                           optimizer=optimDE,
                           plots = FALSE,
                           progress = TRUE,
                           seedFun = 1,
                           seedSPOT = 1,
                           kerasConf = kerasConf)
)
save(res, file = paste0("resKerasTransferLearningTrainingLoss06",
as.numeric(Sys.time()),".RData"))
 }
# }