[Bug Report] Installation of pyRTE+RRTMGP fails tests

[ankurmahesh]

Describe the bug
I am attempting to install pyRTE+RRTMGMP. When I follow the build instructions on the readme for installation via conda, I get this error, when I run

conda build conda.recipe

=================================== FAILURES ===================================
________________________________ test_lw_solver ________________________________

request = <FixtureRequest for <Function test_lw_solver>>

    def test_lw_solver(request):
        path = os.path.dirname(request.path)
        with open(f'{path}/lw_solver_input.json', 'r') as f:
            data = json.load(f)
    
        for key in data:
            values = data[key]
            if isinstance(values, list):
                values = np.array(values)
                data[key] = values
    
        args = list(data.values())
    
        py.rte_lw_solver_noscat(*args)
    
        test_data = np.load(f"{path}/lw_solver_output.npy")
    
>       assert(np.allclose(test_data - data['broadband_up'], 0.))
E       assert False
E        +  where False = <function allclose at 0x7fc2724e6fb0>((array([345.43212164, 204.42786215,  40.22612771,   8.80242324,\n       483.53738262, 286.1590084 ,  56.30870811,  12.32...374.10298773, 374.10298773, 374.10298773, 374.10298773,\n       523.67098538, 523.67098538, 523.67098538, 523.67098538]) - array([172.71606082, 102.21393107,  20.11306386,   4.40121162,\n       241.76869131, 143.0795042 ,  28.15435405,   6.16...187.05149387, 187.05149387, 187.05149387, 187.05149387,\n       261.83549269, 261.83549269, 261.83549269, 261.83549269])), 0.0)
E        +    where <function allclose at 0x7fc2724e6fb0> = np.allclose

tests/lw_solver_test/noscat_test/lw_solver_noscat_test.py:24: AssertionError
____________________________ test_lw_solver_noscat _____________________________

    def test_lw_solver_noscat():
        rrtmgp_gas_optics = kdist.gas_optics.load_atmosferic_conditions(rfmip)
    
        _, solver_flux_up, solver_flux_down, _, _ = lw_solver_noscat(
            tau=rrtmgp_gas_optics.tau,
            lay_source=rrtmgp_gas_optics.lay_src,
            lev_source=rrtmgp_gas_optics.lev_src,
            sfc_emis=rfmip["surface_emissivity"].data,
            sfc_src=rrtmgp_gas_optics.sfc_src,
            sfc_src_jac=rrtmgp_gas_optics.sfc_src_jac,
        )
    
>       assert np.isclose(solver_flux_up, ref_flux_up, atol=ERROR_TOLERANCE).all()
E       assert np.False_
E        +  where np.False_ = <built-in method all of numpy.ndarray object at 0x7fc2581774b0>()
E        +    where <built-in method all of numpy.ndarray object at 0x7fc2581774b0> = array([[False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False],\n      ...alse],\n       [False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False]]).all
E        +      where array([[False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False],\n      ...alse],\n       [False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False]]) = <function isclose at 0x7fc2724eb130>(array([[144.78141419, 144.8159722 , 144.84800098, ..., 228.09748602,\n        230.91920492, 239.10500159],\n       [150....3.2440728 ],\n       [143.73562813, 143.77591885, 143.80751672, ..., 217.92762   ,\n        218.58066063, 219.72328903]]), array([[289.56284, 289.63196, 289.696  , ..., 456.19498, 461.8384 ,\n        478.21   ],\n       [300.46246, 300.51285, ...    446.48816],\n       [287.47125, 287.55185, 287.61502, ..., 435.85526, 437.16132,\n        439.44656]], dtype=float32), atol=0.0001)
E        +        where <function isclose at 0x7fc2724eb130> = np.isclose

tests/test_python_frontend/test_lw_solver.py:43: AssertionError
=========================== short test summary info ============================
FAILED tests/lw_solver_test/noscat_test/lw_solver_noscat_test.py::test_lw_solver
FAILED tests/test_python_frontend/test_lw_solver.py::test_lw_solver_noscat - ...
======================== 2 failed, 54 passed in 15.09s =========================
Tests failed for pyrte_rrtmgp-0.0.6-py39h3fd9d12_0.tar.bz2 - moving package to /global/homes/a/amahesh/conda-bld/broken
WARNING:conda_build.build:Tests failed for pyrte_rrtmgp-0.0.6-py39h3fd9d12_0.tar.bz2 - moving package to /global/homes/a/amahesh/conda-bld/broken
TESTS FAILED: pyrte_rrtmgp-0.0.6-py39h3fd9d12_0.tar.bz2

To Reproduce
I ran git clone git@github.com:earth-system-radiation/pyRTE-RRTMGP.git
cd pyRTE-RRTMPG/
conda install conda-build conda-verify
conda build conda.recipe

Expected behavior
I'd expect the build tests to succeed so that I can use the package in conda

Software Versions:

• OS: Linux on NERSC Perlmutter: AMD EPYC 7763 (which is based on x86-64 I think)
• Version (or most recent commit hash) of RTE+RRTMGP: 1f9205c
  Also, I get this same failure when I try to build locally, on a Mac with Apple Silicon (which I know is not officially supported, according to the README). The macOS is Sonoma 14.4. (Note on the Mac with apple silicon I had to change gfortran_osx-64 in the conda.recipe/meta.yaml to gfortran_impl_osx-arm64)

[ankurmahesh]

After searching the Github repo, I also came across this pull request: #52. I tried cloning into this branch. I had to change the tests/test_python_frontend/test_sw_solver.py to use the rsu and rsd files from rrtgmp-data (not from files stored at this repo), just as was done for test_lw_solver.py in the branch. However, this still leads to errors with passing the lw test cases (see below).

The updated branch appears to have made the test_lw_solver_noscat closer to the true values (though still not within the tolerance). However, the test_lw_solver is still off, seemingly by a factor of 2, just as in the failure above.

________________________________ test_lw_solver ________________________________

request = <FixtureRequest for <Function test_lw_solver>>

    def test_lw_solver(request):
        path = os.path.dirname(request.path)
        with open(f'{path}/lw_solver_input.json', 'r') as f:
            data = json.load(f)
    
        for key in data:
            values = data[key]
            if isinstance(values, list):
                values = np.array(values)
                data[key] = values
    
        args = list(data.values())
    
        py.rte_lw_solver_noscat(*args)
    
        test_data = np.load(f"{path}/lw_solver_output.npy")
    
>       assert(np.allclose(test_data - data['broadband_up'], 0.))
E       assert False
E        +  where False = <function allclose at 0x10229ccb0>((array([345.43212164, 204.42786215,  40.22612771,   8.80242324,\n       483.53738262, 286.1590084 ,  56.30870811,  12.32...374.10298773, 374.10298773, 374.10298773, 374.10298773,\n       523.67098538, 523.67098538, 523.67098538, 523.67098538]) - array([172.71606082, 102.21393107,  20.11306386,   4.40121162,\n       241.76869131, 143.0795042 ,  28.15435405,   6.16...187.05149387, 187.05149387, 187.05149387, 187.05149387,\n       261.83549269, 261.83549269, 261.83549269, 261.83549269])), 0.0)
E        +    where <function allclose at 0x10229ccb0> = np.allclose

tests/lw_solver_test/noscat_test/lw_solver_noscat_test.py:24: AssertionError
____________________________ test_lw_solver_noscat _____________________________

    def test_lw_solver_noscat():
        rrtmgp_gas_optics = kdist.gas_optics.load_atmosferic_conditions(rfmip)
    
        _, solver_flux_up, solver_flux_down, _, _ = lw_solver_noscat(
            tau=rrtmgp_gas_optics.tau,
            lay_source=rrtmgp_gas_optics.lay_src,
            lev_source=rrtmgp_gas_optics.lev_src,
            sfc_emis=rfmip["surface_emissivity"].data,
            sfc_src=rrtmgp_gas_optics.sfc_src,
            sfc_src_jac=rrtmgp_gas_optics.sfc_src_jac,
        )
    
>       assert np.isclose(solver_flux_up, ref_flux_up, atol=ERROR_TOLERANCE).all()
E       assert False
E        +  where False = <built-in method all of numpy.ndarray object at 0x16e27fb10>()
E        +    where <built-in method all of numpy.ndarray object at 0x16e27fb10> = array([[False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False],\n      ...alse],\n       [False, False, False, ...,  True,  True, False],\n       [False, False, False, ...,  True,  True, False]]).all
E        +      where array([[False, False, False, ..., False, False, False],\n       [False, False, False, ..., False, False, False],\n      ...alse],\n       [False, False, False, ...,  True,  True, False],\n       [False, False, False, ...,  True,  True, False]]) = <function isclose at 0x10229e1b0>(array([[289.95776422, 290.02647951, 290.09010084, ..., 456.24771837,\n        461.88459126, 478.1998788 ],\n       [300....6.47711207],\n       [287.8319806 , 287.91209465, 287.97485138, ..., 435.85481284,\n        437.15826435, 439.43587803]]), array([[289.56674, 289.63577, 289.69986, ..., 456.19098, 461.83554,\n        478.21   ],\n       [300.46246, 300.51276, ...    446.48813],\n       [287.47003, 287.55054, 287.6137 , ..., 435.85275, 437.15918,\n        439.44656]], dtype=float32), atol=0.0001)
E        +        where <function isclose at 0x10229e1b0> = np.isclose

tests/test_python_frontend/test_lw_solver.py:44: AssertionError

[ankurmahesh]

The second test can be passed by changing the TAG at this line to 1.8.1, to reflect the updated files in the rrtmpg-data github repository. I will submit a PR with this change.

However, the first test (test_lw_solver), where the results appear to be off by a factor of 2, is still not fixed.

[tcmetzger]

@ankurmahesh Thank you for filing this bug report! Things have been a bit slower than usual over the past weeks, but we are planning to get back to this soon!

[sehnem]

They were fixed, but found that there is an inconsistency in the results of the LW problem in Apple Silicon.