Starting on v0.2 (python3 only)

Makefile

@@ -4,11 +4,8 @@ all:
 develop:
 	python setup.py develop
 
-test2:
-	python2 /usr/local/bin/nosetests --logging-level=INFO
-
-test3:
-	nosetests -v --with-coverage --cover-package=descent --logging-level=INFO
+test:
+	pytest --cov=descent --cov-report=html tests/
 
 clean:
 	rm -rf htmlcov/

descent/__init__.py

@@ -3,21 +3,18 @@
 =======
 
 A Python package for performing first-order optimization
-
-For more information, see the accompanying README.md
-
 """
 
-__all__ = [
-    'algorithms',
-    'proxops',
-    'utils',
-    'main',
-]
+__version__ = '0.2.0'
 
 from .algorithms import *
 from .proxops import *
 from .utils import *
 from .main import *
 
-__version__ = '0.1.5'
+__all__ = [
+    *algorithms.__all__,
+    *proxops.__all__,
+    *utils.__all__,
+    *main.__all__,
+]

descent/algorithms.py

@@ -8,7 +8,8 @@
 from future.utils import with_metaclass
 import numpy as np
 from collections import deque
-from abc import ABCMeta, abstractmethod
+from abc import abstractmethod
+from custom_inherit import DocInheritMeta
 
 __all__ = ['sgd', 'StochasticGradientDescent',
            'nag', 'NesterovAcceleratedGradient',
@@ -17,24 +18,49 @@
            'adam', 'ADAM']
 
 
-class Algorithm(object, with_metaclass(ABCMeta)):
+class Algorithm(object, with_metaclass(DocInheritMeta(style="numpy", abstract_base_class=True))):
 
     def __init__(self, xinit):
-        self.k = 0.
+        """Initializes an optimizer.
+
+        Parameters
+        ----------
+        xinit : array_like
+        """
+        self.k = 0
         self.xk = xinit.copy()
 
     def __next__(self):
         """Called to update every iteration"""
-        self.k += 1.0
+        self.k += 1
 
     @abstractmethod
     def __call__(self, gradient):
+        """Applies the gradient.
+
+        Parameters
+        ----------
+        gradient : array_like
+
+        Returns
+        -------
+        parameters : array_like
+        """
         raise NotImplementedError
 
 
 class StochasticGradientDescent(Algorithm):
 
     def __init__(self, xinit, lr=1e-3, momentum=0., decay=0.):
+        """Stochastic gradient descent.
+
+        Optional Parameters
+        -------------------
+        lr : float
+            The learning rate
+        momentum : float
+        decay : float
+        """
 
         super().__init__(xinit)
         self.vk = np.zeros_like(xinit)
@@ -87,7 +113,7 @@ def __call__(self, gradient):
 
 class RMSProp(Algorithm):
 
-    def __init__(self, xinit, lr=1e-3, damping=0.1, decay=0.9):
+    def __init__(self, xinit, lr=1e-3, damping=1e-12, decay=0.9):
         """
         RMSProp
 
@@ -99,7 +125,7 @@ def __init__(self, xinit, lr=1e-3, damping=0.1, decay=0.9):
             Learning rate (Default: 1e-3)
 
         damping : float, optional
-            Damping term (Default: 0)
+            Damping term (Default: 1e-12)
 
         decay : float, optional
             Decay of the learning rate (Default: 0)
@@ -169,7 +195,7 @@ def __init__(self, xinit, nterms=10, lr=1e-3, epsilon=1e-8):
         self.mem = np.ones_like(self.xk)
         self.g = np.zeros_like(self.xk)
         self.g2 = np.zeros_like(self.xk)
-        self.epsilon = 1e-8
+        self.epsilon = epsilon
 
     def __call__(self, gradient):
 

descent/main.py

@@ -14,6 +14,8 @@
 except ImportError:
     from time import time as perf_counter
 
+__all__ = ['GradientDescent', 'Consensus']
+
 
 class Optimizer(object):
     def __init__(self, theta_init, display=sys.stdout):

tests/test_rosenbrock.py

@@ -11,10 +11,8 @@
 def rosenbrock(theta):
     """Objective and gradient for the rosenbrock function"""
 
-    x = theta[0]
-    y = theta[1]
-
     # Rosenbrock's banana function
+    x, y = theta
     obj = (1 - x)**2 + 100 * (y - x**2)**2
 
     # gradient for the Rosenbrock function