Use log-log interpolation for hazard map and disaggregation calculations

[Damiano Monelli]

When deriving a hazard map or when doing a disaggregation calculation, the ground motion value associated to a probability of exceedance value, is computed by performing a liner interpolation between the two closest probability of exceedance values (see https://github.com/gem/oq-engine/blob/master/openquake/engine/calculators/hazard/classical/post_processing.py#L77, https://github.com/gem/oq-engine/blob/master/openquake/engine/calculators/hazard/disaggregation/core.py#L137.

For more precise results the interpolation should be done in log space, that is instead of:

numpy.interp(poes, curve[::-1], imls)

we should use:

10 ** (numpy.interp(numpy.log10(poes), numpy.log10(curve[::-1]),
numpy.log10(imls)))

[Michele Simionato (michele-simionato)]

If a curve has zero values, or extremely small values, we use a cutoff to avoid issues with the logarithm. We choose 1e-30 as a reasonable value for the cutoff (notice that in postgres poes are stored as 32 bit floats and the minimum possible float is 1e-38).

See https://github.com/gem/oq-engine/pull/1338 for the implementation.