def _rg(sumstats, args, log, M_annot, ref_ld_cnames, w_ld_cname, i):
'''Run the regressions.'''
n_snp = len(sumstats)
s = lambda x: np.array(x).reshape((n_snp, 1))
if args.chisq_max is not None:
ii = sumstats.Z1**2 * sumstats.Z2**2 < args.chisq_max**2
n_snp = np.sum(ii) # lambdas are late binding, so this works
sumstats = sumstats[ii]
n_blocks = min(args.n_blocks, n_snp)
ref_ld = sumstats.as_matrix(columns=ref_ld_cnames)
intercepts = [
args.intercept_h2[0], args.intercept_h2[i + 1],
args.intercept_gencov[i + 1]
]
rghat = reg.RG(s(sumstats.Z1),
s(sumstats.Z2),
ref_ld,
s(sumstats[w_ld_cname]),
s(sumstats.N1),
s(sumstats.N2),
M_annot,
intercept_hsq1=intercepts[0],
intercept_hsq2=intercepts[1],
intercept_gencov=intercepts[2],
n_blocks=n_blocks,
twostep=args.two_step)
return rghat
def setUp(self):
self.ld = np.abs(np.random.normal(size=100).reshape((50, 2))) + 2
self.z1 = (np.sum(self.ld, axis=1) * 10).reshape((50, 1))
self.w_ld = np.random.normal(size=50).reshape((50, 1))
self.N1 = 9 * np.ones((50, 1))
self.N2 = 7 * np.ones((50, 1))
self.M = np.matrix((700, 222))
self.hsq1 = 0.5
self.hsq2 = 0.6
self.rg = reg.RG(self.z1, -self.z1, self.ld, self.w_ld, self.N1, self.N1,
self.M, 1.0, 1.0, 0, n_blocks=20)
def test_negative_h2(self):
ld = np.arange(50).reshape((50, 1)) + 0.1
z1 = (1 / np.sum(ld, axis=1) * 10).reshape((50, 1))
w_ld = np.ones((50, 1))
N1 = 9 * np.ones((50, 1))
M = np.matrix((-700))
rg = reg.RG(z1, -z1, ld, w_ld, N1, N1, M, 1.0, 1.0, 0, n_blocks=20)
assert rg._negative_hsq
# check no runtime errors when _negative_hsq is True
print rg.summary()
print rg.summary(silly=True)
assert rg.rg_ratio == 'NA'
assert rg.rg_se == 'NA'
assert rg.rg == 'NA'
assert rg.p == 'NA'
assert rg.z == 'NA'