Merge pull request #43 from yb6599/master

Derivative.d now handles negative axis arguments

derivative/differentiation.py

@@ -248,7 +248,10 @@ def _restore_axes(dX: NDArray, axis: int, orig_shape: tuple[int, ...]) -> NDArra
         return dX.flatten()
     else:
         # order of operations coupled with _align_axes
-        extra_dims = tuple(length for ax, length in enumerate(orig_shape) if ax != axis)
+        orig_diff_axis = range(len(orig_shape))[axis] # to handle negative axis args
+        extra_dims = tuple(
+            length for ax, length in enumerate(orig_shape) if ax != orig_diff_axis
+        )
         moved_shape = (orig_shape[axis],) + extra_dims
         dX = np.moveaxis(dX.T.reshape((moved_shape)), 0, axis)
         return dX

pyproject.toml

@@ -19,6 +19,7 @@ python = "^3.9"
 numpy = "^1.18.3"
 scipy = "^1.4.1"
 scikit-learn = "^1"
+importlib-metadata = "^7.1.0"
 
 # docs
 sphinx = {version = "^5", optional = true}

tests/test_interface.py

@@ -98,4 +98,15 @@ def test_hyperparam_entrypoint():
     func = utils._load_hyperparam_func("kalman.default")
     expected = 1
     result = func(None, None)
-    assert result == expected
+    assert result == expected
+
+
+def test_negative_axis():
+    t = np.arange(3)
+    x = np.random.random(size=(2, 3, 2))
+    x[1, :, 1] = 1
+    axis = -2
+    expected = np.zeros(3)
+    dx = dxdt(x, t, kind='finite_difference', axis=axis, k=1)
+    assert x.shape == dx.shape
+    np.testing.assert_array_almost_equal(dx[1, :, 1], expected)