def compute_corr(x, y, xlim=None, ylim=None, coeff=None):
"""
Wrapper function to compute the correct correlation coefficient.
"""
# set up the correct function for computing the requested correlation
# coefficient
if coeff == 'spearmanr':
# pass to scipy function
return _spearmanr(x, y)
elif coeff == 'kendallt':
# pass to our kendall tau function, in case we need to handle
# censored data
return kendall(x, y, xlim=xlim, ylim=ylim)
elif coeff == 'pearsonr':
# pass to scipy function
return _pearsonr(x, y)
def correlation(x, y):
rho = _pearsonr(x, y)[0]
return rho
def run_tests():
"""
Test output of pymcspearman against tabulated values from MCSpearman
"""
from tempfile import NamedTemporaryFile as ntf
from urllib.request import urlretrieve
# get test data
tfile = ntf()
urlretrieve(
"https://raw.githubusercontent.com/PACurran/MCSpearman/master/test.data",
tfile.name)
# open temporary file
data = _np.genfromtxt(tfile,
usecols=(0, 1, 2, 3),
dtype=[('x', float), ('dx', float), ('y', float),
('dy', float)])
# tabulated results from a MCSpearman run with 10000 iterations
MCSres = [
(0.8308, 0.001), # spearman only
(0.8213, 0.0470), # bootstrap only
(0.7764, 0.0356), # perturbation only
(0.7654, 0.0584)
] # bootstrapping and perturbation
# spearman only
res = pymccorrelation(data['x'],
data['y'],
dx=data['dx'],
dy=data['dy'],
coeff='spearmanr',
Nboot=None,
Nperturb=None,
return_dist=True)
try:
assert _np.isclose(MCSres[0][0], res[0], atol=MCSres[0][1])
_sys.stdout.write("Passed spearman check.\n")
except AssertionError:
_sys.stderr.write("Spearman comparison failed.\n")
# bootstrap only
res = pymccorrelation(data['x'],
data['y'],
dx=data['dx'],
dy=data['dy'],
Nboot=10000,
coeff='spearmanr',
Nperturb=None,
return_dist=True)
try:
assert _np.isclose(MCSres[1][0], _np.mean(res[2]), atol=MCSres[1][1])
_sys.stdout.write("Passed bootstrap only method check.\n")
except AssertionError:
_sys.stderr.write("Bootstrap only method comparison failed.\n")
# perturbation only
res = pymccorrelation(data['x'],
data['y'],
dx=data['dx'],
dy=data['dy'],
coeff='spearmanr',
Nboot=None,
Nperturb=10000,
return_dist=True)
try:
assert _np.isclose(MCSres[2][0], _np.mean(res[2]), atol=MCSres[2][1])
_sys.stdout.write("Passed perturbation only method check.\n")
except AssertionError:
_sys.stderr.write("Perturbation only method comparison failed.\n")
# composite method
res = pymccorrelation(data['x'],
data['y'],
dx=data['dx'],
dy=data['dy'],
coeff='spearmanr',
Nboot=10000,
Nperturb=10000,
return_dist=True)
try:
assert _np.isclose(MCSres[3][0], _np.mean(res[2]), atol=MCSres[3][1])
_sys.stdout.write("Passed composite method check.\n")
except AssertionError:
_sys.stderr.write("Composite method comparison failed.\n")
# test Kendall tau IFN86 for consistency with scipy
sres = _kendalltau(data['x'], data['y'])
IFN86res = kendall_IFN86(data['x'],
data['y'],
xlim=_np.zeros(len(data)),
ylim=_np.zeros(len(data)))
kt_wrap_res = pymccorrelation(data['x'],
data['y'],
xlim=_np.zeros(len(data)),
ylim=_np.zeros(len(data)),
coeff='kendallt')
try:
assert _np.isclose(sres[0], IFN86res[0])
assert _np.isclose(sres[1], IFN86res[1])
_sys.stdout.write("Passed Kendall tau comparison with scipy.\n")
except AssertionError:
_sys.stderr.write("Kendall tau comparison with scipy failed.\n")
try:
assert _np.isclose(kt_wrap_res[0], IFN86res[0])
assert _np.isclose(kt_wrap_res[1], IFN86res[1])
_sys.stdout.write("Passed internal Kendall tau comparison.\n")
except AssertionError:
_sys.stderr.write("Internal Kendall tau comparison failed.\n")
# test pearson r wrapper
wrap_res = pymccorrelation(data['x'],
data['y'],
coeff='pearsonr',
return_dist=False)
res = _pearsonr(data['x'], data['y'])
try:
assert _np.isclose(wrap_res[0], res[0])
assert _np.isclose(wrap_res[1], res[1])
_sys.stdout.write("Passed Pearson r wrapper check.\n")
except AssertionError:
_sys.stderr.write("Pearson r wrapper check failed.\n")
def pearsonr(y_true, y_pred, multioutput="uniform_average"):
if multioutput == "uniform_average":
return _pearsonr(y_true, y_pred)[0]
else:
return np.array(
[_pearsonr(yt, yp)[0] for yt, yp in zip(y_true.T, y_pred.T)])
def correlation(x, y):
rho = _pearsonr(x, y)[0]
if _np.isnan(rho):
rho = 0.
return rho