Merge pull request #59 from EducationalTestingService/add-weights-as-…

…attribute

Save weights as an attribute

factor_analyzer/factor_analyzer.py

@@ -3,6 +3,7 @@
 with optional rotation using Varimax or Promax.
 
 :author: Jeremy Biggs ([email protected])
+:author: Nitin Madnani ([email protected])
 :date: 10/25/2017
 :organization: ETS
 """
@@ -288,6 +289,7 @@ def __init__(self,
         self.corr_ = None
         self.loadings_ = None
         self.rotation_matrix_ = None
+        self.weights_ = None
 
     @staticmethod
     def _fit_uls_objective(psi, corr_mtx, n_factors):
@@ -734,13 +736,13 @@ def transform(self, X):
             structure = self.loadings_
 
         try:
-            weights = np.linalg.solve(self.corr_, structure)
+            self.weights_ = np.linalg.solve(self.corr_, structure)
         except Exception as error:
             warnings.warn('Unable to calculate the factor score weights; '
                           'factor loadings used instead: {}'.format(error))
-            weights = self.loadings_
+            self.weights_ = self.loadings_
 
-        scores = np.dot(X_scale, weights)
+        scores = np.dot(X_scale, self.weights_)
         return scores
 
     def get_eigenvalues(self):

tests/test_factor_analyzer.py

@@ -52,6 +52,20 @@ def test_calculate_kmo():
 
 class TestFactorAnalyzer:
 
+    def test_analyze_weights(self):
+
+        data = pd.DataFrame({'A': [2, 4, 5, 6, 8, 9],
+                             'B': [4, 8, 9, 10, 16, 18],
+                             'C': [6, 12, 15, 12, 26, 27]})
+
+        fa = FactorAnalyzer(rotation=None)
+        fa.fit(data)
+        _ = fa.transform(data)
+        expected_weights = np.array(([[0.33536334, -2.72509646, 0],
+                                      [0.33916605, -0.29388849, 0],
+                                      [0.33444588, 3.03060826, 0]]))
+        assert_array_almost_equal(expected_weights, fa.weights_)
+
     def test_analyze_impute_mean(self):
 
         data = pd.DataFrame({'A': [2, 4, 5, 6, 8, 9],