[5/N] Modernize FFX: approach namedtuple (#42)

ffx/api.py

@@ -1,6 +1,6 @@
-"""api.py defines user interfaces to FFX. run() runs the complete method.
+'''api.py defines user interfaces to FFX. run() runs the complete method.
 FFXRegressor is a Scikit-learn style regressor.
-"""
+'''
 
 from sklearn.base import BaseEstimator, RegressorMixin
 
@@ -12,7 +12,7 @@ def run(train_X, train_y, test_X, test_y, varnames=None, verbose=False):
 
 
 class FFXRegressor(BaseEstimator, RegressorMixin):
-    """This class provides a Scikit-learn style estimator."""
+    '''This class provides a Scikit-learn style estimator.'''
 
     def fit(self, X, y):
         # if X is a Pandas DataFrame, we don't have to pass in varnames.

ffx/core/__init__.py

@@ -11,6 +11,7 @@
 from sklearn.base import RegressorMixin
 from sklearn.linear_model import ElasticNet
 
+from .approach import Approach
 from .constants import (
     CONSIDER_DENOM,
     CONSIDER_EXPON,
@@ -27,19 +28,6 @@
     OP_MIN0,
 )
 
-
-def _approachStr(approach):
-    assert len(approach) == 5
-    assert set(approach).issubset([0, 1])
-    return 'inter%d denom%d expon%d nonlin%d thresh%d' % (
-        approach[0],
-        approach[1],
-        approach[2],
-        approach[3],
-        approach[4],
-    )
-
-
 # =========================================================================
 # strategy
 
@@ -52,11 +40,9 @@ def __init__(self, approach):
         @arguments
           approach -- 5-d list of [use_inter, use_denom, use_expon, use_nonlin, use_thresh]
         """
-        assert len(approach) == 5
-        assert set(approach).issubset([0, 1])
         self.approach = approach
 
-        self.num_alphas = 1000
+        self._num_alphas = 1000
 
         # final round will stop if either of these is hit
         self.final_target_train_nmse = 0.01  # 0.01 = 1%
@@ -81,28 +67,19 @@ def __init__(self, approach):
         self.all_expr_exponents = [-1.0, -0.5, +0.5, +1.0]
 
     def includeInteractions(self):
-        return bool(self.approach[0])
+        return bool(self.approach.use_inter)
 
     def includeDenominator(self):
-        return bool(self.approach[1])
+        return bool(self.approach.use_denom)
 
     def exprExponents(self):
-        if self.approach[2]:
-            return self.all_expr_exponents
-        else:
-            return [1.0]
+        return self.all_expr_exponents if self.approach.use_expon else [1.0]
 
     def nonlinOps(self):
-        if self.approach[3]:
-            return self.all_nonlin_ops
-        else:
-            return []
+        return self.all_nonlin_ops if self.approach.use_nonlin else []
 
     def thresholdOps(self):
-        if self.approach[4]:
-            return self.all_threshold_ops
-        else:
-            return []
+        return self.all_threshold_ops if self.approach.use_thresh else []
 
     def eps(self):
         return self._eps
@@ -111,7 +88,7 @@ def l1_ratio(self):
         return self._l1_ratio
 
     def numAlphas(self):
-        return self.num_alphas
+        return self._num_alphas
 
 
 # =========================================================================
@@ -506,20 +483,19 @@ def build(self, train_X, train_y, test_X, test_y, varnames=None, verbose=False):
                         for thresh in [0, 1]:
                             if thresh == 1 and not CONSIDER_THRESH:
                                 continue
-                            approach = [inter, denom, expon, nonlin, thresh]
-                            if sum(approach) >= 4:
+                            approach = Approach(inter, denom, expon, nonlin, thresh)
+                            if approach.num_feature_types >= 4:
                                 continue  # not too many features at once
                             approaches.append(approach)
                             if verbose:
-                                # "  ", _approachStr(approach)
-                                print('\t'.join(['Yes' if a else 'No' for a in approach]))
+                                print(approach)
 
         for (i, approach) in enumerate(approaches):
             if verbose:
                 print('-' * 200)
                 print(
                     'Build with approach %d/%d (%s): begin'
-                    % (i + 1, len(approaches), _approachStr(approach))
+                    % (i + 1, len(approaches), str(approach))
                 )
             ss = FFXBuildStrategy(approach)
 
@@ -540,7 +516,7 @@ def build(self, train_X, train_y, test_X, test_y, varnames=None, verbose=False):
             if verbose:
                 print(
                     'Build with approach %d/%d (%s): done.  %d model(s).'
-                    % (i + 1, len(approaches), _approachStr(approach), len(next_models))
+                    % (i + 1, len(approaches), str(approach), len(next_models))
                 )
                 print('Models:')
                 for model in next_models:
@@ -564,15 +540,6 @@ def build(self, train_X, train_y, test_X, test_y, varnames=None, verbose=False):
 
         return models
 
-    def _FFXapproach(self, inter, denom, expon, nonlin, thresh):
-        return 'FFX inter%d denom%d expon%d nonlin%d thresh%d' % (
-            inter,
-            denom,
-            expon,
-            nonlin,
-            thresh,
-        )
-
     def _nondominatedModels(self, models):
         test_nmses = [model.test_nmse for model in models]
         num_bases = [model.numBases() for model in models]

ffx/core/approach.py

@@ -0,0 +1,24 @@
+from collections import namedtuple
+
+
+class Approach(namedtuple('_Approach', 'use_inter use_denom use_expon use_nonlin use_thresh')):
+    def __new__(cls, use_inter, use_denom, use_expon, use_nonlin, use_thresh):
+        assert set([use_inter, use_denom, use_expon, use_nonlin, use_thresh]).issubset([0, 1])
+        return super(Approach, cls).__new__(
+            cls, use_inter, use_denom, use_expon, use_nonlin, use_thresh
+        )
+
+    @property
+    def num_feature_types(self):
+        '''How many types of features does this approach consider?
+        '''
+        return sum(self._asdict().values())
+
+    def __repr__(self):
+        return 'inter%d denom%d expon%d nonlin%d thresh%d' % (
+            self.use_inter,
+            self.use_denom,
+            self.use_expon,
+            self.use_nonlin,
+            self.use_thresh,
+        )

ffx/core/constants.py

@@ -1,4 +1,3 @@
-# user-changeable constants
 CONSIDER_INTER = True  # consider interactions?
 CONSIDER_DENOM = True  # consider denominator?
 CONSIDER_EXPON = True  # consider exponents?

ffx_tests/core_tests/test_approach.py

@@ -0,0 +1,16 @@
+import pytest
+
+from ffx.core.approach import Approach
+
+
+def test_approach():
+    a = Approach(use_inter=0, use_denom=0, use_expon=0, use_nonlin=0, use_thresh=0)
+    assert str(a) == 'inter0 denom0 expon0 nonlin0 thresh0'
+    assert a.num_feature_types == 0
+
+    b = Approach(use_inter=1, use_denom=1, use_expon=1, use_nonlin=1, use_thresh=1)
+    assert str(b) == 'inter1 denom1 expon1 nonlin1 thresh1'
+    assert b.num_feature_types == 5
+
+    with pytest.raises(AssertionError):
+        Approach(use_inter='not an integer', use_denom=1, use_expon=1, use_nonlin=1, use_thresh=1)