Where emulations have already been created, this method combines these to form one ensemble. This takes as input a list of the emulator objects, the simulation parameters and output response labels, and a set of test data from which the performance weights will be evolved. We would recommend providing the testing set of the output from the partition_dataset method. An option is given, by setting these within emulation_algorithm_settings, to save the ensemble object to file, as well as produce plots showing the accuracy of the generated ensemble for the test data set
generate_ensemble_from_existing_emulations(existing_emulations, parameters,
measures, observed_data, algorithm_settings = NULL,
normalise = FALSE, timepoint = NULL, output_formats = c("pdf"))Vector of emulator objects created by method
generate_requested_emulations
Array containing the names of the parameters for which values are input into each emulation
Array containing the names of the output measures predicted by each emulation
Dataset to train the new ensemble on. We recommend using the test data in the set generated by partition_data method, and not the training set, as the emulators themselves have been trained on that data and the ensemble could thus overfit.
Object output from the function emulation_algorithm_settings, containing the settings of the machine learning algorithms used in emulation creation. Here this is needed to decide whether any accuracy plots should be produced during ensemble creation, whether or not the ensemble should be saved to file, and to specify the number of generations for which the neural network that is generating the algorithm weightings should run.
Whether the predictions generated when testing the ensemble should be normalised for presenting test results
If using multiple timepoints, the timepoint for which this ensemble is being created
File formats in which result graphs should be produced
A list containing the ensemble, the time taken to generate it, and the sampling mins and maxes used in its creation such that unseen data used by and predictions generated by the ensemble can be scaled and rescaled correctly