def cramerphi_chi2(data0, data1):
n = len(data0)
assert n == len(data1)
if n == 0:
return float('NaN'), 0, 0
index0 = dict((x, i) for i, x in enumerate(sorted(set(data0))))
index1 = dict((x, i) for i, x in enumerate(sorted(set(data1))))
data0 = numpy.array([index0[d] for d in data0])
data1 = numpy.array([index1[d] for d in data1])
assert data0.ndim == 1
assert data1.ndim == 1
unique0 = numpy.unique(data0)
unique1 = numpy.unique(data1)
n0 = len(unique0)
n1 = len(unique1)
min_levels = min(n0, n1)
if min_levels == 1:
# No variation in at least one column, so no notion of
# correlation.
return float('NaN'), n0, n1
ct = numpy.zeros((n0, n1), dtype=int)
for i0, x0 in enumerate(unique0):
for i1, x1 in enumerate(unique1):
matches0 = numpy.array(data0 == x0, dtype=int)
matches1 = numpy.array(data1 == x1, dtype=int)
ct[i0][i1] = numpy.dot(matches0, matches1)
# Compute observed chi^2 statistic.
chi2 = stats.chi2_contingency(ct)
return chi2, n0, n1
def cramerphi_chi2(data0, data1):
n = len(data0)
assert n == len(data1)
if n == 0:
return float('NaN'), 0, 0
index0 = dict((x, i) for i, x in enumerate(sorted(set(data0))))
index1 = dict((x, i) for i, x in enumerate(sorted(set(data1))))
data0 = numpy.array([index0[d] for d in data0])
data1 = numpy.array([index1[d] for d in data1])
assert data0.ndim == 1
assert data1.ndim == 1
unique0 = numpy.unique(data0)
unique1 = numpy.unique(data1)
n0 = len(unique0)
n1 = len(unique1)
min_levels = min(n0, n1)
if min_levels == 1:
# No variation in at least one column, so no notion of
# correlation.
return float('NaN'), n0, n1
ct = numpy.zeros((n0, n1), dtype=int)
for i0, x0 in enumerate(unique0):
for i1, x1 in enumerate(unique1):
matches0 = numpy.array(data0 == x0, dtype=int)
matches1 = numpy.array(data1 == x1, dtype=int)
ct[i0][i1] = numpy.dot(matches0, matches1)
# Compute observed chi^2 statistic.
chi2 = stats.chi2_contingency(ct)
return chi2, n0, n1
def test_chi2_contingency():
assert stats.chi2_contingency([[42]]) == 0.
assert relerr(7.66, stats.chi2_contingency([[4,2,3], [3,16,2]])) < 0.01
