def test_benchmark():
a, b, c, d = _gen_rand_abcd()
n = 100
res = np.zeros(n)
for i in range(n):
startT = time.time()
ORs, pvalues = fishersapi.fishers_vec(a,
b,
c,
d,
alternative='two-sided')
res[i] = n / (time.time() - startT)
print('Imported test: %1.2f tests per second' % np.mean(res))
n = 1
res_scipy = np.zeros(n)
for i in range(n):
startT = time.time()
ORs, pvalues = fishersapi._scipy_fishers_vec(a,
b,
c,
d,
alternative='two-sided')
res_scipy[i] = n / (time.time() - startT)
print('scipy test: %1.2f tests per second' % np.mean(res_scipy))
def test_fishers_vec():
"""Testing the vectorized version against scipy on random data."""
a, b, c, d = _gen_rand_abcd()
n = len(a)
for alt in ['two-sided', 'less', 'greater']:
ORs, pvalues = fishersapi.fishers_vec(a, b, c, d, alternative=alt)
scipy_pvalues, scipy_ORs = np.zeros(n), np.zeros(n)
for i in range(n):
scipy_ORs[i], scipy_pvalues[i] = stats.fisher_exact(
[[a[i], b[i]], [c[i], d[i]]], alternative=alt)
npt.assert_allclose(ORs, scipy_ORs, rtol=1e-4)
npt.assert_allclose(pvalues, scipy_pvalues, rtol=1e-4)
def test_fishers_vec_minn():
"""Testing the vectorized version against scipy on random data."""
a, b, c, d = _gen_rand_abcd()
n = len(a)
counts = a + b + c + d
gtmin = np.sum(counts >= np.median(counts))
for alt in ['two-sided', 'less', 'greater']:
ORs, pvalues = fishersapi.fishers_vec(a,
b,
c,
d,
alternative=alt,
min_n=np.median(counts))
npt.assert_equal(gtmin, (~np.isnan(pvalues)).sum())
def _fisherNBR(res_df, ct_cols):
"""Applies a Fisher's exact test to every row of res_df using the 4 columns provided
in count_cols. For each row, the vector of counts in count_cols can
be reshaped into a 2 x 2 contingency table.
The count_cols should be in the following order:
a X+/MEM+
b X+/MEM-
c X-/MEM+
d X-/MEM-
where X+ indicates the second level of x_col (e.g. 1 for [0, 1]). The result is that
that an OR > 1 = [(a / c) / (b / d)] indicates enrichment of X
within the cluster.
Relative-rate of X in vs. out of the cluster is also provided.
Parameters
----------
res_df : pd.DataFrame [ntests x 4]
Each row contains a set of 4 counts to be tested.
count_cols : list
Columns containing the counts in a "flattened" order such that
it can be reshaped into a 2 x 2 contingency table
Returns
-------
res : dict
A dict of three numpy vectors containing the OR, the RR and the p-value.
Vectors will have length ntests, same as res_df.shape[0]"""
a = res_df[ct_cols[0]].values
b = res_df[ct_cols[1]].values
c = res_df[ct_cols[2]].values
d = res_df[ct_cols[3]].values
OR, p = fishers_vec(a, b, c, d, alternative='two-sided')
RR = (a / (a + c)) / (b / (b + d))
return {'RR':RR, 'OR':OR, 'pvalue':p}
def test_integers():
OR, pvalue = fishersapi.fishers_vec(10, 2, 15, 3)
assert np.isscalar(OR)
assert np.isscalar(pvalue)