In some implementations, a computer-implemented method for generating computer-readable data models includes receiving time series data; applying a plurality of variable transformations to the time series data to generate a variable matrix with first and second dimensions; partitioning the variable matrix along a first one of the first and second dimensions to generate a plurality of data sets; partitioning the plurality of data sets along a second one of the first and second dimensions to generate a plurality data subsets; providing each of the plurality of data subsets to a respective computational unit in a distributed computing environment for evaluation; receiving, from the respective computational units, scores for a plurality of variables as determined by the respective computational units from the plurality of data subsets; and selecting a portion of the plurality of variables as having at least a threshold level of accuracy in modeling the time series data.