Merge branch 'release/1.2.15'

.travis.yml

@@ -35,6 +35,7 @@ notifications:
   email: false
 deploy:
   provider: pypi
+  server: https://upload.pypi.org/legacy/
   user: dhorta
   password:
     secure: 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

glimix_core/cov/sum.py

@@ -14,15 +14,16 @@ class SumCov(FunctionReduce):
 
         f(f_0, f_1, \dots) = f_0 + f_1 + \dots
     """
+
     def __init__(self, covariances):
         self._covariances = [c for c in covariances]
         FunctionReduce.__init__(self, self._covariances, 'sum')
 
-    def value_reduce(self, values): # pylint: disable=R0201
+    def value_reduce(self, values):  # pylint: disable=R0201
         r"""Sum covariance function evaluated at `(f_0, f_1, ...)`."""
         return add.reduce(list(values.values()))
 
-    def gradient_reduce(self, _, gradients): # pylint: disable=R0201
+    def gradient_reduce(self, _, gradients):  # pylint: disable=R0201
         r"""Sum of covariance function derivatives.
 
         Returns:

glimix_core/ep/ep_linear_kernel.py

@@ -50,23 +50,21 @@ def _lml(self):
         tQ = sqrt(1 - d) * Q
 
         lml = [
-            -log(L.diagonal()).sum(),  #
-            -0.5 * sum(log(s * S)),  #
-            +0.5 * sum(log(A)),  #
+            -log(L.diagonal()).sum(),
+            -0.5 * sum(log(s * S)),
+            +0.5 * sum(log(A)),
             # lml += 0.5 * sum(log(ttau)),
-            +0.5 * dot(teta * A, dot(tQ, cho_solve(L, dot(tQ.T,
-                                                          teta * A)))),  #!=
-            -0.5 * dot(teta, teta / TS),  #
-            +dot(m, A * teta) - 0.5 * dot(m, A * ttau * m),  #
-            -0.5 * dot(m * A * ttau,
-                       dot(tQ,
-                           cho_solve(L, dot(tQ.T,
-                                            2 * A * teta - A * ttau * m)))),  #
-            +sum(self._moments['log_zeroth']),  #
-            +0.5 * sum(log(TS)),  #
+            +0.5 * dot(teta * A, dot(tQ, cho_solve(L, dot(tQ.T, teta * A)))),
+            -0.5 * dot(teta, teta / TS),
+            +dot(m, A * teta) - 0.5 * dot(m, A * ttau * m),
+            -0.5 * dot(
+                m * A * ttau,
+                dot(tQ, cho_solve(L, dot(tQ.T, 2 * A * teta - A * ttau * m)))),
+            +sum(self._moments['log_zeroth']),
+            +0.5 * sum(log(TS)),
             # lml -= 0.5 * sum(log(ttau)),
-            -0.5 * sum(log(ctau)),  #
-            +0.5 * dot(ceta / TS, ttau * ceta / ctau - 2 * teta),  #
+            -0.5 * sum(log(ctau)),
+            +0.5 * dot(ceta / TS, ttau * ceta / ctau - 2 * teta),
             0.5 * s * d * sum(teta * A * teta)
         ]
         lml = fsum(lml)

glimix_core/ep/posterior_linear_kernel.py

@@ -1,13 +1,13 @@
 from __future__ import division
 
 from numpy import sum as npsum
-from numpy import dot, empty, sqrt, concatenate, zeros
-from scipy.linalg import cho_factor
-
+from numpy import concatenate, dot, empty, sqrt, zeros
 from numpy_sugar.linalg import cho_solve, ddot, dotd, sum2diag
+from scipy.linalg import cho_factor
 
 from .posterior import Posterior
 
+
 class PosteriorLinearKernel(Posterior):
     r"""EP posterior.
 

glimix_core/ep/site.py

@@ -1,9 +1,11 @@
 from __future__ import division
 
-from numpy import zeros, maximum
+from numpy import maximum, zeros
 
-class Site(object): # pylint: disable=R0903
+
+class Site(object):  # pylint: disable=R0903
     r"""EP parameters."""
+
     def __init__(self, n):
         self.tau = zeros(n)
         self.eta = zeros(n)

glimix_core/ep/test/test_ep.py

@@ -2,7 +2,8 @@
 
 from glimix_core.ep import EP
 
+
 def test_ep():
     ep = EP(10)
     with pytest.raises(NotImplementedError):
-        print(ep._compute_moments()) # pylint: disable=W0212
+        print(ep._compute_moments())  # pylint: disable=W0212

glimix_core/glmm/glmm.py

@@ -6,8 +6,8 @@
 from numpy import (asarray, clip, concatenate, dot, exp, inf, log, ndarray,
                    zeros)
 from numpy.linalg import LinAlgError
-
 from numpy_sugar import epsilon
+
 from optimix import Function, Scalar, Vector
 from optimix.optimize import BadSolutionError
 
@@ -232,8 +232,7 @@ def gradient(self):  # pylint: disable=W0221
             g = self._lml_derivatives(self._X)
             ev = exp(-v)
             grad = [
-                g['delta'] * (ev/(1 + ev))/(1 + ev),
-                g['scale'] * exp(x),
+                g['delta'] * (ev / (1 + ev)) / (1 + ev), g['scale'] * exp(x),
                 g['mean']
             ]
         except (ValueError, LinAlgError) as e:

glimix_core/glmm/test/test_glmm.py

@@ -1,16 +1,15 @@
 import pytest
-
+from numpy import asarray, ascontiguousarray, dot, ones, sqrt, zeros
 from numpy.random import RandomState
 from numpy.testing import assert_allclose
-from numpy import ascontiguousarray, sqrt, ones, dot, zeros, asarray
 from numpy_sugar.linalg import economic_qs, economic_qs_linear
 
 from glimix_core.example import linear_eye_cov
 from glimix_core.glmm import GLMM
 from glimix_core.random import bernoulli_sample
-
 from optimix import check_grad
 
+
 def test_glmm_precise():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -40,6 +39,7 @@ def test_glmm_precise():
 
     assert_allclose(check_grad(glmm), 0, atol=1e-4)
 
+
 def test_glmm_delta0():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -57,6 +57,7 @@ def test_glmm_delta0():
     assert_allclose(glmm.value(), -294.3289786264443)
     assert_allclose(check_grad(glmm, step=1e-5), 0, atol=1e-2)
 
+
 def test_glmm_delta1():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -74,6 +75,7 @@ def test_glmm_delta1():
     assert_allclose(glmm.value(), -317.9043148331947)
     assert_allclose(check_grad(glmm), 0, atol=1e-4)
 
+
 def test_glmm_wrong_qs():
     random = RandomState(0)
     X = random.randn(10, 15)
@@ -86,6 +88,7 @@ def test_glmm_wrong_qs():
     with pytest.raises(ValueError):
         print(GLMM((nsuc, ntri), 'binomial', X, QS))
 
+
 def test_glmm_optimize():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -116,6 +119,7 @@ def test_glmm_optimize():
 
     assert_allclose(glmm.value(), -159.1688201218538, rtol=1e-06)
 
+
 def test_glmm_optimize_low_rank():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -135,10 +139,11 @@ def test_glmm_optimize_low_rank():
     glmm.feed().maximize(progress=False)
     assert_allclose(glmm.value(), -155.4794212740998, rtol=1e-06)
 
+
 def test_glmm_bernoulli_problematic():
     random = RandomState(1)
     N = 500
-    G = random.randn(N, N+50)
+    G = random.randn(N, N + 50)
     y = bernoulli_sample(0.0, G, random_state=random)
     y = (y, )
 
@@ -160,6 +165,7 @@ def test_glmm_bernoulli_problematic():
     assert_allclose(model.scale, 0.6026005889095781, rtol=1e-5)
     assert_allclose(model.beta, [-0.01806123661347892])
 
+
 def _stdnorm(X, axis=None, out=None):
     X = ascontiguousarray(X)
     if out is None:
@@ -179,6 +185,7 @@ def _stdnorm(X, axis=None, out=None):
 
     return out
 
+
 def test_glmm_binomial_pheno_list():
     random = RandomState(0)
     nsamples = 50
@@ -220,6 +227,7 @@ def test_glmm_scale_very_low():
 
     assert_allclose(check_grad(glmm), 0, atol=1e-2)
 
+
 def test_glmm_scale_very_high():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -237,6 +245,7 @@ def test_glmm_scale_very_high():
 
     assert_allclose(check_grad(glmm), 0, atol=1e-3)
 
+
 def test_glmm_delta_zero():
     random = RandomState(0)
     X = random.randn(100, 5)
@@ -257,6 +266,7 @@ def test_glmm_delta_zero():
     assert_allclose(glmm.value(), -263.56884343483136)
     assert_allclose(glmm.delta, 1)
 
+
 def test_glmm_delta_one():
     random = RandomState(0)
     X = random.randn(100, 5)

glimix_core/lik/test/test_lik.py

@@ -3,9 +3,10 @@
 from numpy.random import RandomState
 from numpy.testing import assert_allclose
 
+from glimix_core.lik import (BernoulliProdLik, BinomialProdLik, DeltaProdLik,
+                             PoissonProdLik)
 from glimix_core.link import ProbitLink
-from glimix_core.lik import (BernoulliProdLik, BinomialProdLik,
-                                 DeltaProdLik, PoissonProdLik)
+
 
 def test_delta_prod_lik():
     random = RandomState(0)
@@ -21,6 +22,7 @@ def test_delta_prod_lik():
     assert_allclose(lik.mean([-1, 0, 0.5]), [-1, 0, 0.5])
     assert_allclose(lik.sample([-10, 0, 0.5], random), [-10, 0, 0.5])
 
+
 def test_bernoulli_prod_lik():
     random = RandomState(0)
 
@@ -35,6 +37,7 @@ def test_bernoulli_prod_lik():
     assert_allclose(lik.mean([-1, 0, 0.5]), [0.15865525, 0.5, 0.69146246])
     assert_allclose(lik.sample([-10, 0, 0.5], random), [0, 1, 1])
 
+
 def test_binomial_prod_lik():
     random = RandomState(0)
 
@@ -50,6 +53,7 @@ def test_binomial_prod_lik():
     assert_allclose(lik.mean([-1, 0, 0.5]), [0.15865525, 0.5, 0.69146246])
     assert_allclose(lik.sample([-10, 0, 0.5], random), [0, 1, 2])
 
+
 def test_poisson_prod_lik():
     random = RandomState(0)
 

glimix_core/link/test/test_link.py

@@ -1,26 +1,29 @@
 import pytest
-
 from numpy.testing import assert_allclose
 
-from glimix_core.link import ProbitLink, LogitLink, LogLink
+from glimix_core.link import LogitLink, LogLink, ProbitLink
 from glimix_core.link.link import Link
 
+
 def test_probit_link():
     link = ProbitLink()
     assert_allclose(link.value(link.inv(3.2)), 3.2)
     assert_allclose(link.latent_variance, 1.0)
 
+
 def test_logit_link():
     link = LogitLink()
     assert_allclose(link.value(link.inv(3.2)), 3.2)
     assert_allclose(link.latent_variance, 3.289868133696453)
 
+
 def test_loglink_link():
     link = LogLink()
     assert_allclose(link.value(link.inv(3.2)), 3.2)
     with pytest.raises(NotImplementedError):
         print(link.latent_variance)
 
+
 def test_link_interface():
     link = Link()
 

glimix_core/lmm/scan.py

@@ -1,15 +1,14 @@
 from __future__ import division
 
-from numpy import append, dot, empty, log, full
 from numpy import sum as npsum
-
-from tqdm import tqdm
-
+from numpy import append, dot, empty, full, log
 from numpy_sugar.linalg import solve
+from tqdm import tqdm
 
 LOG2PI = 1.837877066409345339081937709124758839607238769531250
 
-class FastScanner(object): # pylint: disable=R0903
+
+class FastScanner(object):  # pylint: disable=R0903
     r"""Fast inference over multiple covariates.
 
     Let :math:`\tilde{\mathrm M}_i` be a column-matrix of fixed-effect
@@ -40,7 +39,7 @@ def __init__(self, y, M, QS, delta):
         self._yTQdiag = [l / r for (l, r) in zip(yTQ, self._diags)]
 
         self._a = [(i**2 / j).sum() for (i, j) in zip(yTQ, self._diags)]
-        self._b = [dot(i, j.T) for(i, j) in zip(self._yTQdiag, MTQ)]
+        self._b = [dot(i, j.T) for (i, j) in zip(self._yTQdiag, MTQ)]
 
         self._MTQdiag = [i / j for (i, j) in zip(MTQ, self._diags)]
 
@@ -119,8 +118,10 @@ def _fast_scan_chunk(self, markers):
             beta = solve(self._C[1] - self._C[0], b11m - b00m)
             effect_sizes[i] = beta[-1]
 
-            p0 = self._a[1] - 2 * b11m.dot(beta) + beta.dot(self._C[1].dot(beta))
-            p1 = self._a[0] - 2 * b00m.dot(beta) + beta.dot(self._C[0]).dot(beta)
+            p0 = self._a[1] - 2 * b11m.dot(beta) + beta.dot(
+                self._C[1].dot(beta))
+            p1 = self._a[0] - 2 * b00m.dot(beta) + beta.dot(
+                self._C[0]).dot(beta)
 
             scale = (p0 + p1) / markers.shape[0]
 

glimix_core/mean/__init__.py

@@ -26,8 +26,8 @@
   :members:
 """
 
-from .offset import OffsetMean
 from .linear import LinearMean
+from .offset import OffsetMean
 from .sum import SumMean
 
 __all__ = ['OffsetMean', 'LinearMean', 'SumMean']

glimix_core/mean/linear.py

@@ -1,10 +1,8 @@
 from __future__ import division
 
-from numpy import zeros, dot
-from numpy import ascontiguousarray
+from numpy import ascontiguousarray, dot, zeros
 
-from optimix import Function
-from optimix import Vector
+from optimix import Function, Vector
 
 
 class LinearMean(Function):
@@ -18,6 +16,7 @@ class LinearMean(Function):
 
     where :math:`\boldsymbol\alpha` is a vector of effect sizes.
     """
+
     def __init__(self, size):
         Function.__init__(self, effsizes=Vector(zeros(size)))
 
@@ -32,7 +31,7 @@ def value(self, x):
         """
         return dot(x, self.variables().get('effsizes').value)
 
-    def gradient(self, x): # pylint: disable=R0201
+    def gradient(self, x):  # pylint: disable=R0201
         r"""Gradient of the linear mean function.
 
         Args: