// Perform the "Cluster-Rank-Modeling" method of Wolverton and Meredig
//  See: http://pubs.acs.org/doi/abs/10.1021/cm403727z

// Load in a dataset of compound properties
data = new data.materials.ElementDataset
data import ./datasets/zirconia_sln_energies.txt

// Define where to find elemental property data
data attributes properties directory ./lookup-data/

// Select which set of elemental properties to use for attributes
//  NOTE: This is a different set than those used by Meredig & Wolverton
data attributes properties add set general

// Define solution energy as the target property
data target sln_energy

// Generate new attributes
data attributes generate

// Define clusterer to use k-means++ based on a subset of attributes
clusterer = new cluster.WekaClusterer SimpleKMeans -N 4 -I 500

// Build a model that splits data using the clusterer
model = new models.regression.SplitRegression
model splitter ClustererSplitter $clusterer

// Use the the best-fitting quadratic model of a single attribute
submodel = new models.regression.PolynomialRegression 2
selector = new attributes.selectors.BestIndividualSelector 1 PolynomialFitEvaluator 2
submodel set selector $selector
model submodel set generic $submodel

// Train the model
model train $data
print model training stats
print model model

exit