Testing cluster proposals

[andy-iskauskas]

Currently, none of the 'standard' use cases in unit tests lend themselves to testing lhs_gen_cluster (called from generate_new_runs(..., cluster = TRUE). The basic SIR emulators do not differ in active variable selection enough to warrant clustering into two groups, and even if one manually changes the active variables (eg setting SIREmulators$ems$nR$actives = c(FALSE, FALSE, TRUE)) the three emulators still get put into one cluster.

The ideal would be a set of emulators that

• Do not require large computational time to generate and run (eg, the emulators used to test bimodal emulation);
• Admit a non-trivial clustering via active variables;
• Have an interesting structure that would leverage the cluster generation: perhaps 4+ parameters where the split into clusters gives a good spread of cluster 1, cluster 2, and intersection variables.

This could be an artificial set of emulators, or from a relatively simple model that gives rise to this (perhaps a time-series model)

[andy-iskauskas]

Resolved in commit 310cd98.