EPIC: hazard calculation performance improvements

[chrisdicaprio]

The following improvements to the hazard calculation pipeline are expected to improve performance (reduce calculation time and storage required)

• [ ] calculate each source separatly when component branches have multiple sources (e.g. separate IFM and DSM sources when running OQ). The DSMs are re-used (e.g. TI and TD branches use the same DSMs), recalculating hazard curves for them is a waste of compute time.
  • https://github.com/GNS-Science/toshi-hazard-post/issues/21
  • https://github.com/GNS-Science/nzshm-runzi/issues/160
• [ ] Apply rate (i.e. N) scaling at THP stage rather than having OQ explicitly calculate realizations. Verify that math works and run numerical test to verify
  • https://github.com/GNS-Science/toshi-hazard-post/issues/20
  • https://github.com/GNS-Science/nzshm-model/issues/21
• [ ] https://github.com/GNS-Science/toshi-hazard-post/issues/19

related:

• https://github.com/GNS-Science/nzshm-runzi/issues/159

[chrisdicaprio]

Rate scaling

• I believe the math is correct that we can apply the scaling to the hazard rates in THP. Numerical experiment comparing component realizations show that for the IFM the difference between scaling before OQ and after is ~1e-5 - 1e-6 relative error. HOWEVER, the DSMs appear to not be scaled exactly by the scaling constant which results in 10-100x more error when using branches with both DSM and IFM sources for experiment. Inquired w/ KT to find out more.