emulation_algorithm_settings: Initialise machine-learning algorithms settings for emulation...
In spartan: Simulation Parameter Analysis R Toolkit ApplicatioN: 'spartan'

Some of the machine learning algorithms incorporated have their own specific settings (neural network and random forest, for example). To keep the methods generic to multiple algorithms and to save passing multiple objects around, these are declared in this function and returned as an algorithmSettings object. Where the user wishes to use the default values, there is no need to run this function: this will be created during emulator generation. However, should the user wish to change any of the settings, such as the number of generations in the neural network or the number of trees in random forest, they should run this method with the new values. In addition, this object will also contain two other settings: whether graphs should be plotted of the accuracy of the emulator against the training and test sets, and whether the emulator object that is created should be stored in the working directory. Parameters this object stores are detailed in the arguments section. However, for neural network emulation, the user is required to initialise this object with a list of neural network hidden layer structures to evaluate. Should this not be done, an error message will be produced and the call will terminate.

emulation_algorithm_settings(num_trees = 500,
  num_of_generations = 8e+05, num_of_folds = 10,
  network_structures = NULL, save_emulators = TRUE,
  save_ensemble = TRUE, plot_test_accuracy = TRUE)

`num_trees`	Number of trees used to generate a random forest model. If a random forest is not selected as the emulation technique, this argument does not need to be provided. Defaults to 500
`num_of_generations`	Used in neural network generation, as the maximum number of steps used in training the network. If this is reached, then training is stopped. If a neural network is not selected as the emulation technique, this argument does not need to be provided. Defaults to 800,000
`num_of_folds`	Number of folds (subsets of training data) used in k-fold cross validation when developing a neural network emulator. If a neural network is not selected as the emulation technique, this argument does not need to be provided. Defaults to 10.
`network_structures`	List of neural network structures to examine using k-fold cross validation when developing a neural network emulator. Should be a list in the form of number of nodes in each hidden layer. For example, c(c(4),c(4,3)) would consider two structures, one for a single hidden layer of 4 nodes, and another with two hidden layers of 4 and 3 nodes. If a neural network is not selected as the emulation technique, this argument does not need to be provided.
`save_emulators`	Boolean indicating whether the generated emulator for each simulation output should be saved to file (as an Rda object). This is stored in the current working directory. Defaults to TRUE
`save_ensemble`	Used in Technique 7, to state whether an ensemble generated from multiple emulators should be saved to file. Defaults to TRUE
`plot_test_accuracy`	Boolean indicating whether a plot showing a comparison against the emulator predicted values to those observed in the test set should be created. This ggplot is stored as a PDF in the current working directory. Defaults to FALSE