When small sample sizes are used, Pearson correlation performs better than Spearman. But for large samples Spearman is preferred. When using GMMs some clusters are small and some are large, but KINC requires a blanket correlation method for all clusters. Perhaps we should consider dynamically choosing the best correlation method given the sample size.
This will probably affect the cmx output file as we should probably keep track of which correlation method was used.
When small sample sizes are used, Pearson correlation performs better than Spearman. But for large samples Spearman is preferred. When using GMMs some clusters are small and some are large, but KINC requires a blanket correlation method for all clusters. Perhaps we should consider dynamically choosing the best correlation method given the sample size.
This will probably affect the cmx output file as we should probably keep track of which correlation method was used.