Mean and quantile loss curves computation

[matley]

Currently, the mean/quantile loss curve computation fails when there is at least one degenerate individual loss curve (e.g. a loss curve where all the losses are set to 0).

Once that risklib supports the new event based algorithm ( https://bugs.launchpad.net/oq-risklib/+bug/1168446 ) we will be ready to compute mean/quantile loss curves with the following strategy:

Given a set of loss curves (for the same asset)

0) Let L(i) the losses of the i-th curve, P(i) the poes of the i-th curve, M(i) = max(L(i))

1) let i* be the index of the loss curve having the maximum M(i) and L'(i) a subarray of L(i) where each loss <= M(i*).

2) For each i compute P'(i) by interpolating each L'(i) on the reference losses L(i*). If some L(i) is out of bounds, set the corresponding P'(i) to 0.

3) Compute the mean/quantile curves (as done previously) by considering the curves with losses L'(i) and poes P'(i)

[matley (matley)]

The condition "if zero does not belong to L(i*) add 0 with P(0) = 1" never holds as of https://bugs.launchpad.net/oq-risklib/+bug/1168446

[Michele Simionato (michele-simionato)]

Implemented in https://github.com/gem/oq-engine/pull/1096 and https://github.com/gem/oq-risklib/pull/52

[matley (matley)]

Also related: https://github.com/gem/oq-engine/pull/1140

[matley (matley)]

Description updated. Now we pick up the loss curve with the maximum loss as the reference curve.

[matley (matley)]

https://github.com/gem/oq-engine/pull/1145