neglected to reshape sigma

Author: pavelkomarov

pynumdiff/tests/test_utils.py

@@ -26,7 +26,7 @@ def test_estimate_integration_constant():
             (np.ones(5)*10, np.ones(5)*5 + 0.01*np.random.randn(5), 5), # with some noise
             (np.array([0]), np.array([1]), -1), # singleton case
             (np.vstack([np.arange(5)]*5), np.vstack([np.arange(5) + c for c in range(5)]), -np.arange(5).reshape(-1,1)), # multidimensional case
-            (np.ones((5,5)), np.vstack([np.arange(5) + c for c in range(5)]), -np.arange(1,6).reshape(-1,1))]:
+            (np.ones((7,5)), np.vstack([np.arange(5) + c for c in range(7)]), -np.arange(1,8).reshape(-1,1))]: # nonsquare case
         x0 = utility.estimate_integration_constant(x, x_hat, axis=-1)
         assert np.allclose(x0, c, rtol=1e-3)
 

pynumdiff/utils/utility.py

@@ -58,7 +58,7 @@ def estimate_integration_constant(x, x_hat, M=6, axis=0):
              :math:`\\mathbf{\\hat{x}}` with :math:`\\mathbf{x}`
     """
     s = list(x_hat.shape); s[axis] = 1; s = tuple(s) # proper shape for multidimensional integration constants
-    sigma = median_abs_deviation(x - x_hat, axis=axis, scale='normal') # M is in units of this robust scatter metric
+    sigma = median_abs_deviation(x - x_hat, axis=axis, scale='normal').reshape(s) # M is in units of this robust scatter metric
     if M == float('inf') or np.all(sigma < 1e-3): # If no scatter, then no outliers, so use L2
         return np.mean(x - x_hat, axis=axis).reshape(s) # Solves the l2 distance minimization, argmin_c ||x_hat + c - x||_2^2
     elif M < 1e-3: # small M looks like l1 loss, and Huber gets too flat to work well

pyproject.toml

@@ -7,7 +7,7 @@ name = "pynumdiff"
 dynamic = ["version"]
 description = "pynumdiff: numerical derivatives in python"
 readme = "README.md"
-license = {text = "MIT"}
+license = "MIT"
 maintainers = [
   {name = "Floris van Breugel", email = "fvanbreugel@unr.edu"},
   {name = "Pavel Komarov", email = "pvlkmrv@uw.edu"},