Progress

tests/test_numerics_arrays.py

@@ -44,7 +44,12 @@ def get_rtol(matrix):
     return min(10 * cond * machine_eps,100*cond * machine_eps if cond > 1e8 else 1e-9)
 
 def check_inv(matrix, rtol=None):
-    cond = np.linalg.cond(matrix)
+    just_return = False
+    try:
+        # This will fail for bad matrix (inconsistent size) inputs
+        cond = np.linalg.cond(matrix)
+    except:
+        just_return = True
     py_fail = False
     numpy_fail = False
     try:
@@ -56,7 +61,7 @@ def check_inv(matrix, rtol=None):
     except:
         numpy_fail = True
     if py_fail and not numpy_fail:
-        if cond > 1e14:
+        if not just_return and cond > 1e14:
             # Let ill conditioned matrices pass
             return 
         raise ValueError(f"Inconsistent failure states: Python Fail: {py_fail}, Numpy Fail: {numpy_fail}")
@@ -65,7 +70,8 @@ def check_inv(matrix, rtol=None):
     if not py_fail and numpy_fail:
         # We'll allow our inv to work with numbers closer to
         return
-
+    if just_return:
+        return
 
     # Convert result to numpy array if it isn't already
     result = np.array(result)
@@ -98,9 +104,15 @@ def check_inv(matrix, rtol=None):
     # to zero out anything too close to "zero" relative to the values used in the matrix
     # This is very necessary, and was needed when testing on different CPU architectures
     inv_norm = np.max(np.sum(np.abs(result), axis=1))
-    trivial_relative_to_norm = np.where(np.abs(result)/inv_norm < 100*thresh)
+    if cond < 1e10:
+        zero_thresh = 100*thresh
+    elif cond < 1e14:
+        zero_thresh = 1000*thresh
+    else:
+        zero_thresh = 10000*thresh
+    trivial_relative_to_norm = np.where(np.abs(result)/inv_norm < zero_thresh)
     result[trivial_relative_to_norm] = 0.0
-    trivial_relative_to_norm = np.where(np.abs(expected)/inv_norm < 100*thresh)
+    trivial_relative_to_norm = np.where(np.abs(expected)/inv_norm < zero_thresh)
     expected[trivial_relative_to_norm] = 0.0
 
     if rtol is None:
@@ -506,7 +518,7 @@ def matrix_info(matrix):
     [[1.0, 2.0, 3.0, 4.0],
      [5.0, 6.0, 7.0, 8.0],
      [9.0, 10.0, 11.0, 12.0],
-     [13.0, 14.0, 15.0, 16.0]],  # Nearly singular
+     [13.0, 14.0, 15.0, 16.0]],  # Singular
 
     [[1.0, 0.1, 0.1, 0.1],
      [0.1, 2.0, 0.2, 0.2],