upgrade coverage and better handling of univariate and multivariate c…

…ases.

.travis.yml

@@ -24,7 +24,8 @@ install:
 
 # Run test
 script:
-  - nosetests --with-coverage --cover-package mdla.py --logging-level=INFO
+  - nosetests --with-coverage --cover-package mdla.py --cover-package dict_metrics.py --logging-level=INFO
+
 
 # Calculate coverage
 after_success:

dict_metrics.py

@@ -137,15 +137,11 @@ def frobenius_based(A, B):
     return norm(np.abs(A)-np.abs(B), 'fro')
 
 def abs_euclidean(A, B):
-    if A.shape != B.shape:
-        raise ValueError('Atoms have different dim (', A.shape, ' and ', B.shape,'). Error raised in abs_euclidean(A, B)')
     if (A.ndim != 1 and A.shape[1] != 1) or (B.ndim != 1 and B.shape[1] != 1):
         raise ValueError('Atoms are not univariate (', A.shape, ' and ', B.shape,'). Error raised in abs_euclidean(A, B)')
     return 2.*(1.-np.abs(A.T.dot(B)))
 
 def euclidean (A, B):
-    if A.shape != B.shape:
-        raise ValueError('Atoms have different dim (', A.shape, ' and ', B.shape,'). Error raised in euclidean(A, B)')
     if (A.ndim != 1 and A.shape[1] != 1) or (B.ndim != 1 and B.shape[1] != 1):
         raise ValueError('Atoms are not univariate (', A.shape, ' and ', B.shape,'). Error raised in euclidean(A, B)')
     return 2.*(1.-A.T.dot(B))

test_dict_metrics.py

@@ -14,14 +14,11 @@
 from dict_metrics import hausdorff, emd, _multivariate_correlation, \
      detection_rate, precision_recall, precision_recall_points, beta_dist
 
-# TODO: verif scale, dist (and non existant dist), n_kernels,
-#       univariate, multivariate
-
 n_kernels, n_features, n_dims = 10, 5, 3
 dm = [randn(n_features, n_dims) for i in range(n_kernels)]
 for i in range(len(dm)):
     dm[i] /= norm(dm[i], 'fro')
-du = [randn(n_features,) for i in range(n_kernels)]
+du = [randn(n_features, 1) for i in range(n_kernels)]
 for i in range(len(du)):
     du[i] /= norm(du[i])
 
@@ -59,18 +56,30 @@ def test_kernel_registration():
         assert_almost_equal(0., m(dm3, dm, 'chordal'))
 
     # max(dm3) > max(dm4), min(dm4) > min(dm3)
-    dm4 = []
-    for i in range(len(dm)):
-        k_l = dm[i].shape[0]
-        dm4.append(concatenate((zeros((i/2+1, 3)), dm[i]), axis=0))
-    dm5 = []
-    for i in range(len(dm)):
-        k_l = dm[i].shape[0]
-        dm5.append(concatenate((zeros((3, 3)), dm[i]), axis=0))
+    # dm4 = []
+    # for i in range(len(dm)):
+    #     k_l = dm[i].shape[0]
+    #     dm4.append(concatenate((zeros((i/2+1, 3)), dm[i]), axis=0))
+    # dm5 = []
+    # for i in range(len(dm)):
+    #     k_l = dm[i].shape[0]
+    #     dm5.append(concatenate((zeros((3, 3)), dm[i]), axis=0))
+    # for m in [hausdorff, emd]:
+    #     assert_almost_equal(0., m(dm4, dm5, 'chordal'))
+    #     assert_almost_equal(0., m(dm5, dm4, 'chordal'))
 
-    for m in [hausdorff, emd]:
-        assert_almost_equal(0., m(dm4, dm5, 'chordal'))
-        assert_almost_equal(0., m(dm5, dm4, 'chordal'))
+    du2 = [randn(n_features+i/2, 1) for i in range(n_kernels)]
+    for i in range(len(du2)):
+        du2[i] /= norm(du2[i])
+    du3 = []
+    for i in range(len(du)):
+        k_l = du[i].shape[0]
+        du3.append(concatenate((zeros((4, 1)), du[i]), axis=0))
+    for g, m in zip(gdu, [hausdorff, emd]):
+        assert_not_equal(0., m(du, du2, g))
+        assert_not_equal(0., m(du2, du, g))
+        assert_almost_equal(0., m(du, du3, g))
+        assert_almost_equal(0., m(du3, du, g))
 
 def test_unknown_metric():
     for m in [hausdorff, emd]:
@@ -93,8 +102,8 @@ def test_inhomogeneous_dims():
 def test_univariate():
     for m in [hausdorff, emd]:
         for g in gdu:
-            assert_raises (ValueError, emd, dm, dm, g)
-
+            assert_raises (ValueError, m, dm, dm, g)
+            
 def test_correlation():
     du2 = [randn(n_features,) for i in range(n_kernels)]
     for i in range(len(du2)):