[GitHub] [incubator-mxnet] sjtuWangDing commented on a change in pull request #16970: [numpy] add op tensorinv

[GitBox <gi...@apache.org>]

sjtuWangDing commented on a change in pull request #16970: [numpy] add op tensorinv
URL: https://github.com/apache/incubator-mxnet/pull/16970#discussion_r354119962
 
 

 ##########
 File path: tests/python/unittest/test_numpy_op.py
 ##########
 @@ -3412,6 +3412,113 @@ def check_inv(A_inv, data_np):
         check_inv(A_inv, data_np)
 
 
+@with_seed()
+@use_np
+def test_np_linalg_tensorinv():
+    class TestTensorinv(HybridBlock):
+        def __init__(self, ind=2):
+            super(TestTensorinv, self).__init__()
+            self._ind = ind
+        
+        def hybrid_forward(self, F, a):
+            return F.np.linalg.tensorinv(a, ind=self._ind)
+
+    def check_tensorinv(inv_a, a_np, ind):
+        try:
+            inv_a_expected = _np.linalg.tensorinv(a_np, ind=ind)
+        except Exception as e:
+            print(a_np)
+            print(a_np.shape)
+            print(e)
+        else:
+            assert inv_a.shape == inv_a_expected.shape
+            assert_almost_equal(inv_a.asnumpy(), inv_a_expected, rtol=rtol, atol=atol)
+
+    def newInvertibleMatrix(shape):
+        # generate well-conditioned matrices with small eigenvalues
+        n = int(np.prod(np.array(shape[:-2]))) if len(shape) > 2 else 1
+        # eigenvalues
+        D = _np.array([_np.diag(_np.random.uniform(-10., 10., shape[-1])) \
+                        for i in range(n)]).reshape(shape)
+        # orthogonal matrix through householder transformation
+        I = _np.array([_np.eye(shape[-1]) for i in range(n)]).reshape(shape)
+        v = _np.random.uniform(-10, 10,
+            int(np.prod(np.array(shape[:-1])))).reshape(shape[:-1] + (1,))
+        v = v / _np.linalg.norm(v, axis=-2, keepdims=True)
+        v_T = _np.swapaxes(v, -1, -2)
+        U = I - 2 * _np.matmul(v, v_T)
+        return _np.matmul(_np.matmul(U, D), _np.swapaxes(U, -1, -2))
+
+    def get_grad_A(A, ind):
+        inv_A = _np.linalg.tensorinv(A, ind)
+        d_inv_A = _np.ones_like(inv_A)
+        axes1 = len(A.shape) - ind
+        axes2 = ind
+        inv_A_trans_axes = tuple(_np.arange(len(A.shape)))[axes1:] + tuple(_np.arange(len(A.shape)))[:axes1]
+        inv_A_trans = _np.transpose(inv_A, inv_A_trans_axes)
+        temp_tensor = -_np.tensordot(inv_A_trans, d_inv_A, axes = axes1)
+        return _np.tensordot(temp_tensor, inv_A_trans, axes = axes2)
+
+    shapes = [
+        (0, 0, 0),
+        (1, 0, 0),
+        (1, 1, 1),
+        (1, 2, 2),
+        (1, 6, 2, 3),
+        (1, 20, 4, 5),
+        (1, 24, 3, 8),
+        (1, 30, 5, 6),
+        (2, 1, 1),
+        (2, 1, 1, 1),
+        (2, 2, 10, 4, 5),
+        (2, 4, 6, 3, 8),
+        (2, 5, 8, 4, 10),
+        (2, 12, 5, 3, 4, 5),
+        (3, 1, 1, 1),
+        (3, 2, 3, 4, 24),
+        (3, 3, 4, 5, 2, 3, 10),
+        (3, 2, 4, 10, 40, 2)
+    ]
+    dtypes = ['float32', 'float64']
+    for hybridize, shape, dtype, in itertools.product([False, True], shapes, dtypes):
+        rtol = 1e-3 
+        atol = 1e-5
+        ind = shape[0]
+        test_tensorinv = TestTensorinv(ind=ind)
+        if hybridize:
+            test_tensorinv.hybridize()
+
+        if 0 not in shape:
 
 Review comment:
   In numpy, tensorinv op can not calculate shape like (1, 0, 0)

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services