ilabel feature for site-dependent gmm breaks when not all labels are present in site model

[kejohnso]

When I run job that uses the USA preliminary model and only a single site, it fails with the error printed below. Small tests suggest that this is related to the site_labels set in the job file. In my example, there are four labels (site_labels = SeattleBasin CoastalPlains LosAngeles SanFrancisco) which are indexed in the same order as 1, 2, 3, and 4, but the sites file only include a site of case 2. If I reduce the labels to only the included one (e.g. site_labels = CoastalPlains and change the site file ilabel accordingly to 1, the job completes.

While technically it's possible to run jobs, this isn't ideal because it's better to have consistent labels among various job files.

I will send example jobs privately.

File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/contexts.py", line 1901, in get_cmakers gweights = full_lt.g_weights(all_trt_smrs)[:, -1] # shape Gt File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/logictree.py", line 1164, in g_weights for rlzs in self.get_rlzs_by_gsim(trt_smrs).values(): File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/logictree.py", line 1372, in get_rlzs_by_gsim for gsim, rlzs in self._rlzs_by_gsim(t).items(): File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/logictree.py", line 1336, in _rlzs_by_gsim return self.get_rlzs_by_gsim_dic()[trt_smr] File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/logictree.py", line 1361, in get_rlzs_by_gsim_dic self._rlzs_by = _ddic(trtis, smrs, lambda smr: rlzs[slices[smr]]) File "/Users/kjohnson/GEM/oq-engine/openquake/hazardlib/logictree.py", line 1054, in _ddic rbg[rlz.gsim_rlz.value[trti]].append(rlz.ordinal)

[micheles]

Perhaps we should switch to a dictionary-like syntax? Then the association label->number would be explicit: # first job site_labels = { 'SeattleBasin': 1, 'CoastalPlains': 2, 'LosAngeles': 3, 'SanFrancisco': 4) # second job site_labels = { 'CoastalPlains': 2} It would be the user responsibility to use the same numbering scheme across jobs, i.e. 'CoastalPlains' has to be 2 in both job.ini files in this example.

…

On Tue, Sep 9, 2025 at 4:15 AM Kendra Johnson ***@***.***> wrote: Assigned #10789 <#10789> to @micheles <https://github.com/micheles>. — Reply to this email directly, view it on GitHub <#10789 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AASZQ4CHJTXJC6VWCBSNNL33RYZ2RAVCNFSM6AAAAACF7HLI6SVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJZGU3TONRQGI3DSMA> . You are receiving this because you were assigned.Message ID: ***@***.***>

[micheles]

There is still an error when running job_memphis_breaks.ini:

  File "/home/michele/oq-engine/openquake/hazardlib/logictree.py", line 1336, in _rlzs_by_gsim
    return self.get_rlzs_by_gsim_dic()[trt_smr]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michele/oq-engine/openquake/hazardlib/logictree.py", line 1361, in get_rlzs_by_gsim_dic
    self._rlzs_by = _ddic(trtis, smrs, lambda smr: rlzs[slices[smr]])
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michele/oq-engine/openquake/hazardlib/logictree.py", line 1054, in _ddic
    rbg[rlz.gsim_rlz.value[trti]].append(rlz.ordinal)
        ~~~~~~~~~~~~~~~~~~^^^^^^
IndexError: tuple index out of range

[kejohnso]

@micheles another bug on the ilabel topic...

I created a modified version of case06 (attached) with just one site that should be in the Cascadia region. It seems that (a) the gmms are still chosen from the default logic tree and (b) they are assigned to the wrong TRTs. For example:

| uncertainty_type | uvalue |
|----------------------+-----------------------------------------------------------------|
| sourceModel | ssm.xml |
| Subduction Interface | [NSHMP2014]\ngmpe_name = "BooreEtAl2014"\nsgn = 0 |
| Active Shallow Crust | [AtkinsonMacias2009]\nba08_site_term=true\ncb14_basin_term=true |

job.zip

[micheles]

oq show branches by construction cannot show anything else than the default logic tree, since it has no knowledge of the site being used. The issue is: is the calculator using the right logic tree and TRT or not? It seems correct to me, however we are missing information about what gsims were really used in the calculation for a given site. I will try to add a test for that.

[kejohnso]

@micheles I do not think that the correct logic tree is being selected. Here is a second test with two jobs which differ in the gmmLT and whether ilablel is used :

1. job_memphis_test_ilabel_default.ini: the CoastalPlains logic tree (1 trt with 1 gmm) is the only logic tree, and ilabel is not used
2. job_memphis_test_ilabel_select.ini: there are two gmmLTs (CoastalPlains and Default), and the job file and site model match the site to CoastalPlains

test-ilabel.zip

Running the jobs and then using oq compare hcurves PGA -1 -2 shows they give different results:

| site_id | calc_id    | 0.00500   | 0.00653   | 0.00852   | 0.01112   | 0.01452   | 0.01896   | 0.02475   | 0.03230   | 0.04217   | 0.05505   | 0.07187   | 0.09382   | 0.12247   | 0.15988   | 0.20872   | 0.27247   | 0.35569   | 0.46433   | 0.60615 | 0.79130 | 1.03299 | 1.34850 | 1.76039 | 2.29808 | 3.00000 |
|---------+------------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+---------+---------+---------+---------+---------+---------+---------|
| 0       | 99_962_914 | 8.732E-06 | 8.727E-06 | 8.702E-06 | 8.631E-06 | 8.459E-06 | 8.112E-06 | 7.510E-06 | 6.606E-06 | 5.429E-06 | 4.100E-06 | 2.804E-06 | 1.712E-06 | 9.207E-07 | 4.300E-07 | 1.700E-07 | 5.384E-08 | 1.178E-08 | 0.0       | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     |
| 0       | 99_962_913 | 8.732E-06 | 8.732E-06 | 8.727E-06 | 8.703E-06 | 8.631E-06 | 8.455E-06 | 8.102E-06 | 7.488E-06 | 6.566E-06 | 5.370E-06 | 4.029E-06 | 2.731E-06 | 1.650E-06 | 8.765E-07 | 4.037E-07 | 1.568E-07 | 4.828E-08 | 9.835E-09 | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     | 0.0     |