def test_ensemble():
tcrecs = ensemble.tcrecs_generate('cmip5', dist='lognorm', n=100,
seed=40000)
# check first 100 lines
filepath = os.path.join(os.path.dirname(__file__),
'tcrecs/tcrecs_output.txt')
tcrecs_compare = np.loadtxt(filepath)
assert np.allclose(tcrecs,tcrecs_compare)
def test_ensemble_generator():
"""This test determines whether the ensemble generator is behaving as
expected."""
# load up the CMIP5 TCR and ECS values
infile = os.path.join(os.path.dirname(__file__),
'../../fair/tools/tcrecs/cmip5tcrecs.csv')
cmip5_tcrecs = np.loadtxt(infile, skiprows=3, delimiter=',')
cmip5_tcr_mean = np.mean(cmip5_tcrecs[:, 0])
cmip5_ecs_mean = np.mean(cmip5_tcrecs[:, 1])
cmip5_tcr_std = np.std(cmip5_tcrecs[:, 0])
cmip5_ecs_std = np.std(cmip5_tcrecs[:, 1])
cmip5_corrcoef = np.corrcoef(cmip5_tcrecs[:, 0], cmip5_tcrecs[:, 1])[1, 0]
# generate 1000 random values with these inputs. Set strip to False
# so that overall statistics are not affected and more comparable with
# input data
ensgen_tcrecs = ensemble.tcrecs_generate(seed=0, strip_ecs_lt_tcr=False)
ensgen_tcr_mean = np.mean(ensgen_tcrecs[:, 0])
ensgen_ecs_mean = np.mean(ensgen_tcrecs[:, 1])
ensgen_tcr_std = np.std(ensgen_tcrecs[:, 0])
ensgen_ecs_std = np.std(ensgen_tcrecs[:, 1])
ensgen_corrcoef = np.corrcoef(ensgen_tcrecs[:, 0], ensgen_tcrecs[:, 1])[1,
0]
# check generated values match cmip5 stats to within 5%
# possibly too generous but should catch variance v stdev errors
assert 0.95 < ensgen_tcr_mean / cmip5_tcr_mean < 1.05
assert 0.95 < ensgen_tcr_std / cmip5_tcr_std < 1.05
assert 0.95 < ensgen_ecs_mean / cmip5_ecs_mean < 1.05
assert 0.95 < ensgen_ecs_std / cmip5_ecs_std < 1.05
assert 0.95 < ensgen_corrcoef / cmip5_corrcoef < 1.05
# check changing seed changes distribution
ensgen_tcrecs1 = ensemble.tcrecs_generate(seed=1, strip_ecs_lt_tcr=False)
assert np.any(ensgen_tcrecs1 != ensgen_tcrecs)
# check uncorrelated distribution
ensgen_tcrecs = ensemble.tcrecs_generate(seed=0, correlated=False)
assert 0.95 < ensgen_tcr_mean / cmip5_tcr_mean < 1.05
assert 0.95 < ensgen_tcr_std / cmip5_tcr_std < 1.05
assert 0.95 < ensgen_ecs_mean / cmip5_ecs_mean < 1.05
assert 0.95 < ensgen_ecs_std / cmip5_ecs_std < 1.05
assert (-0.10 < np.corrcoef(ensgen_tcrecs[:, 0], ensgen_tcrecs[:, 1])[1, 0]
< 0.10)
# check normal distribution assumption
ensgen_tcrecs = ensemble.tcrecs_generate(seed=0,
strip_ecs_lt_tcr=False,
dist='norm')
ensgen_tcr_mean = np.mean(ensgen_tcrecs[:, 0])
ensgen_ecs_mean = np.mean(ensgen_tcrecs[:, 1])
ensgen_tcr_std = np.std(ensgen_tcrecs[:, 0])
ensgen_ecs_std = np.std(ensgen_tcrecs[:, 1])
assert 0.95 < ensgen_tcr_mean / cmip5_tcr_mean < 1.05
assert 0.95 < ensgen_tcr_std / cmip5_tcr_std < 1.05
assert 0.95 < ensgen_ecs_mean / cmip5_ecs_mean < 1.05
assert 0.95 < ensgen_ecs_std / cmip5_ecs_std < 1.05
assert 0.95 < ensgen_corrcoef / cmip5_corrcoef < 1.05
# check normal, correlated
ensgen_tcrecs = ensemble.tcrecs_generate(seed=0,
correlated=False,
dist='norm')
assert 0.95 < ensgen_tcr_mean / cmip5_tcr_mean < 1.05
assert 0.95 < ensgen_tcr_std / cmip5_tcr_std < 1.05
assert 0.95 < ensgen_ecs_mean / cmip5_ecs_mean < 1.05
assert 0.95 < ensgen_ecs_std / cmip5_ecs_std < 1.05
assert (-0.10 < np.corrcoef(ensgen_tcrecs[:, 0], ensgen_tcrecs[:, 1])[1, 0]
< 0.10)