Test failure on an M2 Apple silicon Mac #8
Description
At least one test fails when inflow is installed under an arm64 architecture on a Mac with an M2 Max chip. On the same M2 Max Mac, all inflow tests pass when installed under an x86_64 (Intel-based) architecture. The failure appears to stem from very minor floating point precision differences in one of the weight tables being tested, but pytest will report a failure nonetheless. This does not appear to impact the integrity of inflow for use, but I wanted to document the test failure for reference.
For performance consistent with Intel-based Macs, inflow can be installed under an x86_64-based conda environment on Apple silicon Macs. Further reading on how to do this: https://taylorreiter.github.io/2022-04-05-Managing-multiple-architecture-specific-installations-of-conda-on-apple-M1/.
Comparing the benchmark and output weight tables:
$ diff tests/data/weight_table/weight_lis_no_intersect.csv tests/output/weight_table_test_lis_no_intersect.csv 32,33c32,33 < 45379,770746.6012,19,70,27,-2.0350,51.6550,4537900700019 < 45379,770746.6012,20,70,27,-2.0250,51.6550,4537900700020 --- > 45379,770746.6013,19,70,27,-2.0350,51.6550,4537900700019 > 45379,770746.6013,20,70,27,-2.0250,51.6550,4537900700020
The failure report from pytest:
========================================== FAILURES ==========================================
______________________ test_generate_weight_table_missing_intersections ______________________
def test_generate_weight_table_missing_intersections():
"""
Verify that the main routine in weight.py produces a weight table
consistent with benchmark using a land-sea mask.
"""
catchment_shapefile = os.path.join(DATA_DIR, 'catchment',
'thames_near_cirencester_uk',
'Catchment_thames_drainID45390.shp')
connectivity_file = os.path.join(DATA_DIR, 'connectivity',
'rapid_connect_45390.csv')
lsm_file = os.path.join(DATA_DIR, 'lsm_grids', 'lis',
'LIS_HIST_201101210000.d01.nc')
lsm_lat_variable = 'lat'
lsm_lon_variable = 'lon'
out_weight_table_file = os.path.join(OUTPUT_DIR,
'weight_table_test_lis_no_intersect.csv')
benchmark_file = os.path.join(DATA_DIR, 'weight_table',
'weight_lis_no_intersect.csv')
weight.generate_weight_table(lsm_file, catchment_shapefile,
connectivity_file,
out_weight_table_file,
lsm_lat_variable=lsm_lat_variable,
lsm_lon_variable=lsm_lon_variable,
catchment_id_field_name='DrainLnID')
> compare_weight_table_files(out_weight_table_file, benchmark_file)
tests/test_weight.py:366:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
tests/test_weight.py:60: in compare_weight_table_files
assert_almost_equal(output, benchmark)
../../miniforge3/envs/inflow/lib/python3.11/contextlib.py:81: in inner
return func(*args, **kwds)
../../miniforge3/envs/inflow/lib/python3.11/contextlib.py:81: in inner
return func(*args, **kwds)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
args = (<function assert_array_almost_equal.<locals>.compare at 0x143c62480>, array([[ 4.52770000e+04, 0.00000000e+00, -9.99...0.00000000e+00, -9.99900000e+03,
-9.99900000e+03, -9.99900000e+03, -9.99900000e+03,
-9.99900000e+03]]))
kwds = {'err_msg': '', 'header': 'Arrays are not almost equal to 7 decimals', 'precision': 7, 'verbose': True}
@wraps(func)
def inner(*args, **kwds):
with self._recreate_cm():
> return func(*args, **kwds)
E AssertionError:
E Arrays are not almost equal to 7 decimals
E
E Mismatched elements: 2 / 329 (0.608%)
E Max absolute difference: 9.99999465e-05
E Max relative difference: 1.29744259e-10
E x: array([[ 4.5277000e+04, 0.0000000e+00, -9.9990000e+03, -9.9990000e+03,
E -9.9990000e+03, -9.9990000e+03, -9.9990000e+03],
E [ 4.5335000e+04, 5.5984600e+01, 2.1000000e+01, 7.0000000e+01,...
E y: array([[ 4.5277000e+04, 0.0000000e+00, -9.9990000e+03, -9.9990000e+03,
E -9.9990000e+03, -9.9990000e+03, -9.9990000e+03],
E [ 4.5335000e+04, 5.5984600e+01, 2.1000000e+01, 7.0000000e+01,...
../../miniforge3/envs/inflow/lib/python3.11/contextlib.py:81: AssertionError