Merge 194a3d1 into b85dd66

convoys/regression.py

@@ -1,4 +1,5 @@
 import autograd
+from autograd_gamma import gammainc
 import emcee
 import numpy
 from scipy.special import gammaincinv
@@ -7,7 +8,6 @@
 import scipy.optimize
 import sys
 import warnings
-from convoys.gamma import gammainc
 
 
 __all__ = ['Exponential',

docs/index.rst

@@ -26,7 +26,4 @@ Full API documentation
 .. automodule:: convoys.utils
    :members:
 
-.. automodule:: convoys.gamma
-   :members:
-
 .. include:: afterwords.rst

setup.py

@@ -20,6 +20,7 @@
       packages=find_packages(),
       install_requires=[
           'autograd',
+          'autograd-gamma>=0.2.0',
           'emcee',
           'matplotlib>=2.0.0',
           'pandas',

test_convoys.py

@@ -6,11 +6,9 @@
 import pandas
 import pytest
 import random
-import scipy.special
 import scipy.stats
 matplotlib.use('Agg')  # Needed for matplotlib to run in Travis
 import convoys
-import convoys.gamma
 import convoys.plotting
 import convoys.regression
 import convoys.single
@@ -29,43 +27,6 @@ def generate_censored_data(N, E, C):
     return B, T
 
 
-def test_gammainc(k=2.5, x=4.2, g_eps=1e-7):
-    # Verify that function values are correct
-    assert convoys.gamma.gammainc(k, x) == pytest.approx(scipy.special.gammainc(k, x))
-
-    # Verify that it handles vectors
-    assert convoys.gamma.gammainc(k, numpy.array([1, 2, 3])) == \
-        pytest.approx(scipy.special.gammainc(k, numpy.array([1, 2, 3])))
-
-    # Verify the derivative wrt k
-    f_grad_k = autograd.grad(
-        lambda k: convoys.gamma.gammainc(k, x))
-    f_grad_k_numeric = (scipy.special.gammainc(k + g_eps, x) -
-                        scipy.special.gammainc(k, x)) / g_eps
-    assert f_grad_k(k) == pytest.approx(f_grad_k_numeric)
-
-    # Verify the derivative wrt x
-    f_grad_x = autograd.grad(
-        lambda x: convoys.gamma.gammainc(k, x))
-    f_grad_x_numeric = (scipy.special.gammainc(k, x + g_eps) -
-                        scipy.special.gammainc(k, x)) / g_eps
-    assert f_grad_x(x) == pytest.approx(f_grad_x_numeric)
-
-    # Verify the derivative wrt x when x is a vector
-    f_grad_x = autograd.grad(
-        lambda x: autograd.numpy.sum(convoys.gamma.gammainc(1.0, x)))
-    f_grad_x_correct = autograd.grad(
-        lambda x: autograd.numpy.sum(1.0 - autograd.numpy.exp(-x)))
-    xs = numpy.array([1., 2., 3.])
-    assert f_grad_x(xs) == pytest.approx(f_grad_x_correct(xs))
-
-    # Verify the derivative wrt k when x is a vector
-    xs = numpy.array([1., 2., 3.])
-    f_grad_k = autograd.grad(
-        lambda k: autograd.numpy.sum(convoys.gamma.gammainc(k, xs)))
-    assert f_grad_k(xs).shape == (3,)
-
-
 def test_kaplan_meier_model():
     data = [
             (2, 0),