Elastic Net Regularizer

dask_glm/algorithms.py

@@ -24,7 +24,7 @@
 
 from dask_glm.utils import dot, exp, log1p
 from dask_glm.families import Logistic
-from dask_glm.regularizers import _regularizers
+from dask_glm.regularizers import Regularizer
 
 
 def compute_stepsize_dask(beta, step, Xbeta, Xstep, y, curr_val,
@@ -155,7 +155,7 @@ def admm(X, y, regularizer='l1', lamduh=0.1, rho=1, over_relax=1,
 
     pointwise_loss = family.pointwise_loss
     pointwise_gradient = family.pointwise_gradient
-    regularizer = _regularizers.get(regularizer, regularizer)  # string
+    regularizer = Regularizer.get(regularizer)
 
     def create_local_gradient(func):
         @functools.wraps(func)
@@ -331,7 +331,7 @@ def proximal_grad(X, y, regularizer='l1', lamduh=0.1, family=Logistic,
     recalcRate = 10
     backtrackMult = firstBacktrackMult
     beta = np.zeros(p)
-    regularizer = _regularizers.get(regularizer, regularizer)  # string
+    regularizer = Regularizer.get(regularizer)
 
     if verbose:
         print('#       -f        |df/f|    |dx/x|    step')

dask_glm/regularizers.py

@@ -9,6 +9,7 @@ class Regularizer(object):
     Defines the set of methods required to create a new regularization object. This includes
     the regularization functions itself and it's gradient, hessian, and proximal operator.
     """
+    _name = '_base'
 
     def f(self, beta):
         """Regularization function."""
@@ -44,9 +45,18 @@ def wrapped(beta, *args):
             return hess(beta, *args) + lam * self.hessian(beta)
         return wrapped
 
+    @classmethod
+    def get(cls, obj):
+        if isinstance(obj, cls):
+            return obj
+        elif isinstance(obj, str):
+            return {o._name: o for o in cls.__subclasses__()}[obj]()
+        raise TypeError('Not a valid regularizer object.')
+
 
 class L2(Regularizer):
     """L2 regularization."""
+    _name = 'l2'
 
     def f(self, beta):
         return (beta**2).sum() / 2
@@ -63,6 +73,7 @@ def proximal_operator(self, beta, t):
 
 class L1(Regularizer):
     """L1 regularization."""
+    _name = 'l1'
 
     def f(self, beta):
         return (np.abs(beta)).sum()
@@ -83,6 +94,7 @@ def proximal_operator(self, beta, t):
 
 class ElasticNet(Regularizer):
     """Elastic net regularization."""
+    _name = 'elastic_net'
 
     def __init__(self, weight=0.5):
         self.weight = weight
@@ -111,10 +123,3 @@ def func(b):
                 return 0
             return (b - g * np.sign(b)) / (t - g + 1)
         return beta
-
-
-_regularizers = {
-    'l1': L1(),
-    'l2': L2(),
-    'elastic_net': ElasticNet()
-}

dask_glm/tests/test_algos_families.py

@@ -9,7 +9,7 @@
 from dask_glm.algorithms import (newton, bfgs, proximal_grad,
                                  gradient_descent, admm)
 from dask_glm.families import Logistic, Normal, Poisson
-from dask_glm.regularizers import _regularizers
+from dask_glm.regularizers import Regularizer
 from dask_glm.utils import sigmoid, make_y
 
 
@@ -89,7 +89,7 @@ def test_basic_unreg_descent(func, kwargs, N, nchunks, family):
 @pytest.mark.parametrize('nchunks', [1, 10])
 @pytest.mark.parametrize('family', [Logistic, Normal, Poisson])
 @pytest.mark.parametrize('lam', [0.01, 1.2, 4.05])
-@pytest.mark.parametrize('reg', list(_regularizers.values()))
+@pytest.mark.parametrize('reg', [r() for r in Regularizer.__subclasses__()])
 def test_basic_reg_descent(func, kwargs, N, nchunks, family, lam, reg):
     beta = np.random.normal(size=2)
     M = len(beta)

dask_glm/tests/test_estimators.py

@@ -3,16 +3,16 @@
 from dask_glm.estimators import LogisticRegression, LinearRegression, PoissonRegression
 from dask_glm.datasets import make_classification, make_regression, make_poisson
 from dask_glm.algorithms import _solvers
-from dask_glm.regularizers import _regularizers
+from dask_glm.regularizers import Regularizer
 
 
-@pytest.fixture(params=_solvers.keys())
+@pytest.fixture(params=[r() for r in Regularizer.__subclasses__()])
 def solver(request):
     """Parametrized fixture for all the solver names"""
     return request.param
 
 
-@pytest.fixture(params=_regularizers.keys())
+@pytest.fixture(params=[r() for r in Regularizer.__subclasses__()])
 def regularizer(request):
     """Parametrized fixture for all the regularizer names"""
     return request.param