unexpected nan numbers in loss curves

[Miriam Colombi]

Once I run seismic risk assessment I got some nan numbers as PoEs in some of the loss curves. In the zip folder "MiriamColombi" attached there are my input files.

[Lars Butler (lars-butler)]

I can reproduce this issue.

Something else I've noticed: when I try to export `loss_curve` artifacts from the DB, I get the following error.

Traceback (most recent call last):
  File "bin/openquake", line 588, in <module>
    main()
  File "bin/openquake", line 572, in main
    export_risk(output_id, expanduser(target_dir))
  File "bin/openquake", line 386, in export_risk
    export(risk_export.export, risk_output_id, target_dir)
  File "bin/openquake", line 399, in export
    files = fn(output_id, target_dir)
  File "/Users/larsbutler/proj/oq-engine/openquake/engine/export/risk.py", line 52, in export
    return [fn(output, os.path.expanduser(target_dir))]
  File "/Users/larsbutler/proj/oq-engine/openquake/engine/export/core.py", line 56, in wrapped
    return fn(output, target_dir)
  File "/Users/larsbutler/proj/oq-engine/openquake/engine/export/risk.py", line 109, in export_loss_curve_xml
    writers.LossCurveXMLWriter(**args).serialize(data)
  File "/Users/larsbutler/proj/oq-nrmllib/openquake/nrmllib/risk/writers.py", line 64, in __init__
    statistics, quantile_value)
  File "/Users/larsbutler/proj/oq-nrmllib/openquake/nrmllib/risk/writers.py", line 898, in validate_hazard_metadata
    _check_logic_tree_metadata(source_model_tree_path, gsim_tree_path)
  File "/Users/larsbutler/proj/oq-nrmllib/openquake/nrmllib/risk/writers.py", line 926, in _check_logic_tree_metadata
    raise ValueError("When specifying a logic tree branch, "
ValueError: When specifying a logic tree branch, both `source_model_tree_path` and `gsim_tree_path` must be specified.

[Lars Butler (lars-butler)]

The pull request which introduced `hazard_metadata`, upon which the serialization relies to get `*_tree_path` information is here: https://github.com/gem/oq-engine/pull/1048/files#L4R1482

It's seems that it does not account for all cases, and thus a bunch of the metadata is missing when we try to serialize.

[Lars Butler (lars-butler)]

First pull request: https://github.com/gem/oq-engine/pull/1104

This stabilizes things so the type of calculation which produced the NaN issue can run without crashing. The NaN issue is still there.

[Lars Butler (lars-butler)]

Some vulnerability model validation: https://github.com/gem/oq-engine/pull/1108